Search

Multi-feature segmentation has demonstrated its superiority against one-dimensional feature approaches based on only grayscale information. Mean shift (MS) is an algorithm that has been used commonly for multi-feature segmentation. In spite of its interesting results, MS maintains a computational cost that is prohibitive for segmentation scenarios where the feature map consists of multi-dimensional features. In this paper, a new competitive segmentation algorithm for grayscale images is introduced. The proposed approach considers a two-dimensional feature map that includes the grayscale value and the local variance for each pixel in the image. To reduce the computational cost, the Mean shift (MS) algorithm is modified to operate with a very limited number of points from all available data. Under such conditions, two sets of elements are differentiated: involved data (the reduced dataset considered in the MS operation) and not involved data (the rest of the available data). Different from the classical MS, which employs Gaussian functions, in our approach, the process of estimating the feature map is carried out using a more accurate approach such as the Epanechnikov kernel function. Once the MS results are obtained, they are generalized to include the not involved data. Therefore, each unused element is assigned to the same cluster of the closest used data. Finally, clusters with the fewest elements are fused with other neighboring clusters. The proposed segmentation method has been compared with other state-of-art algorithms considering the full number of images from the Berkeley dataset. Experimental results confirm that the proposed scheme produces segmented images with a 50% better quality of visual perception approximately two times (≈ 1.8 − 2) faster than its competitors.

Optimization is used in different fields of engineering to solve complex problems. In image processing, multilevel thresholding requires to find the optimal configuration of thresholds to obtain accurate segmented images. In this case, the use of two-dimensional histograms is helpful because they permit us to combine information from the image preserving different features. This paper introduces a new method for multilevel image thresholding segmentation based on the improved version of the owl search algorithm (iOSA) and 2D histograms. The performance of the iOSA is enhanced with the inclusion of a new strategy in the optimization process. Moreover, in the initialization step, it is applied the opposition-based learning. Meanwhile, the 2D histograms permit to maintain more information of the image. Considering such modifications, the iOSA performs a better exploration of the search space during the early iterations, preserving the exploitation of the prominent regions using a self-adaptive variable. The iOSA is employed to allocate the optimal threshold values that segment the image by using the 2D Renyi entropy as an objective function. To test the efficiency of the iOSA, a set of experiments were performed which validate the quality of the segmentation and evaluate the optimization results efficacy. Moreover, to prove that the iOSA is a promising alternative for optimization and image processing problems, statistical tests and analyses were also conducted.

The natural aggregation algorithm (NAA) is a new efficient population-based optimizer. The NAA has a competent performance when compared to other well-established optimizers. However, a problem of concern is NAA lack of exploitation in its local search. In this article, we propose an improved version of NAA. The modifications made are: hypercubes with displacement and shrink mechanism applied in each shelter, we designed a new movement operator to search inside the hypercubes, an improved readjustment of the algorithm’s parameters and “leave shelter” formula of NAA, to better mimic the aggregation behavior. To prove the effectiveness of the modified hypercube natural aggregation algorithm (HYNAA), we compared with classics optimizers, such as PSO, DE and ABC, state of the art, such as CMA-ES, MSA and NAA himself with a benchmark of 28 functions. The said functions consist of five unimodal, 19 multimodal and four hybrids, and we compared them on 30, 50 and 100 dimensions. We also made extra comparisons against NAA in 500 and 1000 dimensions to contrast the ability of the hypercubes to reduce the dimensional complexity. Finally, we tested two trajectory optimization problems. Experimental results and statistical tests demonstrate that the performance of HYNAA is significantly better than that of other optimizers.

This paper proposes a general multi-objective thresholding segmentation methodology for color images and a quality metric designed to prevent and quantify the overlapping effect of segmented images. Multi-level thresholding (MTH) has been used to segment color images in recent years; this process considers each channel as a single grayscale image and applies the MTH independently. Although this method provides competitive results, the inherent relationship among color channels is disregarded. Such approaches generate spurious classes on overlapping regions, where new colors are generated, especially on the borders of the objects. The proposed multi-objective color thresholding (MOCTH) approach performs image segmentation while preserving the relationship between image channels. MOCTH is aimed to reduce the overlapping effect on segmented color images without performing additional post-processing. To measure the overlapping classes on a thresholded color image, the overlapping index is proposed to quantify the pixels affected. The presented approach is analyzed on two color spaces (RGB and CIE L*a*b*) using three multi-objective algorithms; they are NSGA-III, SPEA-2, and MOPSO. Results provide evidence pointing out to a better segmentation from MOCTH over the traditional single-objective approaches while reducing overlapped areas on the image.

This chapter presents the main concepts of metaheuristic schemes for Wireless Sensor Networks (WSNs). WSNs are multi-functional, low-cost, and low-power networks and rely on communications among devices, from sensor nodes to one or more sink nodes. Sink nodes, sometimes called coordinator nodes or root nodes, may be more robust and have larger processing capacity than the other nodes.

The world, as we know, it would not exist without the development of machine learning (ML). Its application in processing information gives the human more advantages than in any other age. The use of ML brings the opportunity to perform categorizations into a big amount of elements and into areas that the human simply can not do it without this technology. But the development of ML algorithms is complex since it requires specific theoretical bases and the development of skills to achieve the tuning of these algorithms. Given the wide variety of algorithms and techniques to perform element classification, this chapter focuses on the study of Convolutional Neural Networks (CNN), and the development of tests that facilitate the understanding of the effects of their basic parameters. The classification objects used are images, and only the multiple class classification model is considered.

Clustering involves the process of dividing a collection of abstract objects into a number of groups, which are integrated with similar elements. Several clustering methods have been introduced and developed in the literature with different performance levels. Among such approaches, density algorithms present the best advantages, since they are able to find clusters from a dataset under different scales, shapes, and densities without requiring the number of groups as input. On the other hand, there are processes that humans perform much better than deterministic approaches or computers. In fact, humans can visually cluster data exceptionally well without the necessity of any training. Under this unique capability, clustering, instantaneous, and effortless for humans, represents a fundamental challenge for artificial intelligence. In this chapter, a simple clustering model inspired by the way in which the human visual system associates patterns spatially is presented. The model, at some abstraction level, can be characterized as a density grouping strategy. The approach is based on Cellular Neural Networks (CNNs), which have demonstrated to be the best models for emulating the human visual system. In the method, similar to the biological model, CNN is used to build especially groups while an automatic mechanism tries different resolution scales to find the best possible data categorization. Different datasets have been adopted to evaluate the performance of the algorithm. Their results are also compared with popular density clustering techniques from the literature. Computational results demonstrate that the CNN approach presents competitive results in comparison with other algorithms regarding accuracy and robustness.

In the last years, medical image processing has become an important tool for health care. In particular, the analysis of retinal vessel images has become crucial for achieving a more accurate diagnosis and treatment for several cardiovascular and ophthalmological deceases. However, this task is extremely hard and time-consuming, often requiring human supervised segmentation of fundus images as well as some degree of professional skills. An automatic yet accurate procedure for retinal vessel segmentation is essential to assist ophthalmologists with illness detection. Several retinal vessel segmentation methods have been developed with satisfactory results. Nevertheless, the image preprocessing techniques implemented in this kind of procedure is known to have poor performance, mainly due to the complex nature of retinal vessel imaging. To improve the image preprocessing stage, a fast and accurate approach for retinal vessel segmentation is presented. This approach aims to enhance the contrast between retinal vessels and the image’s background by implementing a natural inspired technique called Lateral Inhibition (LI). Additionally, a cross-entropy minimization procedure based on Differential Evolution (DE) is applied to find the appropriate threshold and then used to define whether an image pixel is a vessel or non-vessel. The described method has been tested by considering two well-known image databases: DRIVE and STARE. The obtained results are compared against those of other related methods in order to prove the accuracy, effectiveness, and robustness of the presented approach.

The hyperparameters of a convolutional neural network always have been important, because the performance of the convolutional neural network depends largely on them. To find manually the optimal value requires a lot of work, experience, and time. Metaheuristic algorithms are approximate general-purpose search and optimization algorithms, they are used to find a good solution into search space. In the book chapter, the hyperparameters of a convolutional neural network are optimized. A metaheuristic algorithm is used to finds the best solution, that is, the best hyperparameter value for higher accuracy of the convolutional neural network (CNN). The results of four metaheuristic algorithms that have the same objective are compared, which are PSO (Particle Swarm Optimization), ABC (Artificial Bee Colony), ALO (Ant Lion Optimization), and BA (Bat Algorithm). Each algorithm uses the same convolutional neural network architecture as well as the same dataset (MNIST database) also each algorithm is run 30 times to perform the corresponding statistics. Finally, the results of the algorithms are shown.

Chest radiography is one of the most widely used imaging techniques for the detection and diagnosis of lung diseases. However, the correct extraction of information in these images is a tedious and challenging task that depends on the experience of radiologists. Due to this, in recent years, computer-aided diagnostic tools have been implemented to improve the quality and productivity in the diagnosis of radiological tests. The segmentation process represents an important step for any automatic method of information extraction because it simplifies the representation of images. Recently, the use of the threshold segmentation technique has become popular due to its simplicity in implementation and precision. In this work, is proposed an efficient segmentation approach based on the Whale Optimization Algorithm (WOA) to treat the maximization of two statistical criteria, the Otsu’s variance and the Kapur’s entropy. The method is tested for the segmentation of X-ray images to visually diagnose pneumonia, determining details such as the scope, location of the infection, and some complications that may arise. To validate the efficacy of the approach with both statistical criteria, it is tested with chest radiography images from a database and compared with four other metaheuristic algorithms. Comparisons are made to verify the quality of the segmented image using the PSNR, SSIM and FSIM metrics. According to the experimental results, the WOA algorithm used to maximize the variance of Otsu presents a better segmentation compared to the other algorithms.

This chapter approaches the topic of Extreme Learning Machines (ELM); this technique is implemented in a single layer feedforward neural network and provides to the neural network with a fast performance since it is not necessary to adjust weights and hidden layer biases. ELM is implemented and combined with three metaheuristic algorithms. The selected algorithms belong to the family of Swarm intelligence such as Particle Swarm Optimization (PSO), Artificial Bee Colony (ABC) and Grey Wolf Optimization (GWO); these algorithms are applied to optimize the hidden biases and input weights and are implemented for improving the performance along with techniques of Extreme learning machines in data classification. Also, the performance and behavior of each algorithm can be observed. The data employed in this work consist of information about Australian credit, diabetes detection and hearth disease. The investigation result show that GWO obtained better results in the first dataset and PSO are better in the second and third datasets.

Various methods are so complex to be handled quantitatively; However, human beings have achieved by using simple rules that are extracted from their experiences. Fuzzy logic resembles human reasoning in its use of information to generate inaccurate decisions. Diffuse logic incorporates an alternative way of processing that allows complex systems to be modeled using a high level of abstraction originating from human knowledge and experiences. Recently, several of the new evolutionary computing algorithms have been proposed with exciting results. Several of them use operators based on metaphors of natural or social elements that evolve candidate solutions. Although humans have demonstrated their potential to solve complicated optimization problems of everyday life, they are not mechanisms to include such aptitudes into an evolutionary optimization algorithm. In this chapter, a methodology to implement human intelligence based on strategy optimization is presented. Under this approach, a procedure carried out is codified in rules based on Takagi-Sugeno diffuse inference system. So, to implement fuzzy practices, they express the conditions under which candidate solutions are evolved into new positions. To show the capability and robustness of the proposed approach, it is compared to other well-known evolutionary methods. The comparison examines several benchmark functions (benchmark) that are generally considered within the literature of evolutionary algorithms. The results confirm a high performance of the method in the search for a global optimum of different benchmark functions.

In recent years, image processing techniques have been an essential tool for health care. A good and timely analysis of retinal vessel images has become relevant in the identification and treatment of diverse cardiovascular and ophthalmological illness. Therefore, an automatic and precise method for retinal vessel and optic disc segmentation is crucial for illness detection. This task is arduous, time-consuming, and generally developed by an expert with a considerable grade of professional skills in the field. Various retinal vessel segmentation approaches have been developed with promissory results. Although, most of such methods present a deficient performance principally due to the complex structure of vessels in retinal images. In this work, an accurate and hybrid methodology for retinal vessel and optic disc segmentation is presented. The method proposed a fusion of two different schemas: the lateral inhibition (LI) and Differential Evolution (DE). LI is used to improve the contrast between the retinal vessel and background. Followed by the minimization of the cross-entropy function to find the threshold value, these performed by the second schema the DE algorithm. To test the performance and accuracy of the proposed methodology, a set of images obtained from three public datasets STARE, DRIVE, and DRISHTI-GS have been used in different experiments. Simulation results demonstrate the high performance of the proposed approach in comparison with related methods reported in the literature.

Shape recognition in images represents one of the complex and hard-solving problems in computer vision due to its nonlinear, stochastic and incomplete nature. Classical image processing techniques have been normally used to solve this problem. Alternatively, shape recognition has also been conducted through metaheuristic algorithms. They have demonstrated to have a competitive performance in terms of robustness and accuracy. However, all of these schemes use old metaheuristic algorithms as the basis to identify geometrical structures in images. Original metaheuristic approaches experiment several limitations such as premature convergence and low diversity. Through the introduction of new models and evolutionary operators, recent metaheuristic methods have addressed these difficulties providing in general better results. This paper presents a comparative analysis on the application of five recent metaheuristic schemes to the shape recognition problem such as the Grey Wolf Optimizer (GWO), Whale Optimizer Algorithm (WOA), Crow Search Algorithm (CSA), Gravitational Search Algorithm (GSA) and Cuckoo Search (CS). Since such approaches have been successful in several new applications, the objective is to determine their efficiency when they face a complex problem such as shape detection. Numerical simulations, performed on a set of experiments composed of images with different difficulty levels, demonstrates the capacities of each approach.

In machine learning and data mining, feature selection (FS) is one of the most important tasks required to select the most relevant instances from a dataset. In other words, FS is used to reduce the amount of information, creating a subset that represents the entire pool of data. The accuracy of the FS is reflected in a good classification of the information. This article presents an improved version of the social spider optimization (SSO) algorithm. The SSO tends to fail in local optima during the iterative process and is not possible to avoid this situation in the standard form. The proposed version avoids selecting the irrelevant features that demerit the performance of the FS. To achieve this goal, the opposition-based learning is used, in which there is a rule used to increase the exploration of the search space and the prominent zones in a determined neighborhood. The proposed algorithm is called opposition-based social spider optimization (OBSSO), and it has been tested over different mathematical problems. Moreover, the OBSSO, also, has been tested and compared with similar approaches using different datasets with specific information selected from UCI repository. The experimental results provide the evidence of the capabilities of the OBSSO for solving complex optimization problems.

Optimization is a very important issue to efficiently solve engineering problems. The clear comprehension of the elements and processes of optimizer algorithms is fundamental to ensure that engineer students are capable to apply, tune, and design new algorithms. This paper presents an educational platform based on LEGO© EV3 and MATLAB to assist the learning of the principles of classical and metaheuristic optimization algorithms at undergraduate level, by providing a simple and easy-to-follow teaching setup. The proposed study aims to accompany students through the learning of optimization fundamentals by building hands-on robotic experiments. The proposed educational platform has been successfully applied to several undergraduate courses within the Electronics Department at the University of Guadalajara. The description of each experiment and the evaluation of their impact in the student performance are both provided in the paper.

Extracting knowledge is the multidisciplinary process of identifying novel, significant, potentially useful, and consistent information in data. One of the most interesting techniques in the fields of extracting knowledge and machine learning are the self-organization maps (SOMs). They have the capacity of mapping complex high-dimensional relations onto a reduced lattice preserving the topological organization of the initial data. On the other hand, Evolutionary approaches provide an effective alternative to solve complex optimization problems in different application domains. One important characteristic in the application of evolutionary methods to real-world problems is its high demand for function evaluations before obtaining a satisfying solution. In their operation, evolutionary techniques produce new solutions without extracting useful knowledge from a large number of solutions already generated. The use of acquired knowledge during the evolution process could significantly improve their performance in conducting the search strategy toward promising regions or increasing its convergence properties. This paper introduces an evolutionary optimization algorithm in which knowledge extracted during its operation is employed to guide its search strategy. In the approach, a SOM is used as extracting knowledge technique to identify the promising areas through the reduction of the search space. Therefore, in each generation, the proposed method uses a subset of the complete group of generated solutions seen so-far to train the SOM. Once trained, the neural unit from the SOM lattice that corresponds to the best solution is identified. Then, by using local information of this neural unit an entire population of candidate solutions is produced. With the use of the extracted knowledge, the new approach improves the convergence to difficult high multi-modal optima by using a reduced number of function evaluations. The performance of our approach is compared to several state-of-the-art optimization techniques considering a set of well-known functions and three real-world engineering problems. The results validate that the introduced method reaches the best balance regarding accuracy and computational cost over its counterparts.

Multi-Objective Evolutionary Algorithms (MOEAs) are known to solve problems where two or more conflicting goals are involved. To accomplish it, MOEAs incorporate strategies to determinate optimal trade-offs between each of the objective functions. In this paper, an Unassisted image Thresholding (UTH) methodology is proposed based on MOEAs. UTH takes advantage of the trade-off mechanisms present on MOEAs to perform the image thresholding while simultaneously determinating the number thresholds required to segment each image and the best placement of each threshold along the histogram of the image. The image thresholding problem is commonly addressed as the search for the best possible thresholds able to partition a given image into a finite number of homogeneous classes. Such approach requires the assistance of a designer to determinate the number of threshold values that will properly segment the image. However, as images can vary significantly, the definition of an optimal number of thresholds should be performed for each image. Thus, a methodology able to determinate both the number of thresholds and the best placement of each value contributes to a general segmentation scheme. In the proposed approach, UTH redefines the thresholding problem as a multi-objective task with two conflicting goals. The first goal is the quality of the segmented image, and it is computed as a non-parametric criteria to evaluate candidate threshold points. The second goal is the normalized number of threshold points. Since the number of thresholds is not fixed, a particle encoding the thresholds with variable length is used. The strategy of UTH is coupled with three MOEAs namely NSGA-III, PESA-II and MOPSO using as the non-parametric criteria the Cross Entropy. According to the results, the UTH NSGA-III formulation outperforms UTH-PESA-II and UTH-MOPSO regarding convergence and quality of the resulting image.

Breast cancer is one of the most common diseases and the second cause of death in women around the world. The presence of a cancerous tumor increase temperature in the region near to it, such heating is then transferred to the skin surface. In this sense, screening seeks to help in cancer diagnostic process before symptoms became evident in a person, different imaging techniques are employed for this purpose (Mammography, Ultrasonography, X-ray, Magnetic Resonance, etc). In the past decade, thermography has shown its major potential to early diagnosis of breast diseases. Thermographic images provide information related to vascular or physiological changes and have some advantages regarding other diagnostic methods; they are non-ionizing, non-invasive, passive, painless and real-time screening. On the other hand, thresholding has been widely used to solve several problems. It is regularly the first step in the process of image analysis that uses histograms to classify the pixels in the image. Segmentation of medical digital images has been stated as an important task for several medical applications. This paper proposes a segmentation technique for thermographic images that consider the spatial information of the pixel contained in the image. This approach employs a novel optimization technique called the Dragonfly Algorithm to compute the best thresholds that segment the image. The experimental results exhibit a well-performance of the proposal in comparison to the other methods over a set of randomly selected thermograms retrieved from the Database for Research Mastology with Infrared Image. The presented proposed approach could provide a highly reliable clinical decision support, which aims to help clinicians in performing a diagnosis using thermography images.

Class imbalance is an important problem in many domains such as disease classification, network intrusion detection, fraud detection, and spam filtering. While dealing with imbalanced datasets, traditional supervised machine learning algorithms do not often provide acceptable results. Several approaches are used to handle the class imbalance problem. Of these, undersampling approaches are mostly followed by the researchers in which the number of instances in the majority class gets reduced. Selection of instances from the majority class can be considered as an optimization problem. To this end, in this paper, we present an undersampling approach based on widely-used Particle Swarm Optimization (PSO). The majority class samples are first clustered to form the initial undersampled set. The samples to be selected are then optimized using PSO to give the best model. The parameters of PSO are fine tuned using Learning Automata. Appropriate metrics suitable for class imbalance problems have been used to construct the fitness function for optimizing the undersampled training set. The proposed method has achieved 2% to 10% performance improvement over most of the contemporary methods on various datasets with imbalance ratios ranging from 5 to 130, thus showing that the method is robust and useful in practical scenarios. The code of the proposed method can be accessed via https://github.com/kkg1999/Undersampling .

Coronavirus disease 2019 (COVID-19) is a contagious disease caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). It may cause severe ailments in infected individuals. The more severe cases may lead to death. Automated methods which can detect COVID-19 in radiological images can help in the screening of patients. In this work, a two-stage pipeline composed of feature extraction followed by feature selection (FS) for the detection of COVID-19 from CT scan images is proposed. For feature extraction, a state-of-the-art Convolutional Neural Network (CNN) model based on the DenseNet architecture is utilised. To eliminate the non-informative and redundant features, the meta-heuristic called Harris Hawks optimisation (HHO) algorithm combined with Simulated Annealing (SA) and Chaotic initialisation is employed. The proposed approach is evaluated on the SARS-COV-2 CT-Scan dataset which consists of 2482 CT-scans. Without the Chaotic initialisation and the SA, the method gives an accuracy of around 98.42% which further increases to 98.85% on the inclusion of the two and thus delivers better performance than many state-of-the-art methods and various meta-heuristic based FS algorithms. Also, comparison has been drawn with many hybrid variants of meta-heuristic algorithms. Although HHO falls behind a few of the hybrid variants, when Chaotic initialisation and SA are incorporated into it, the proposed algorithm performs better than any other algorithm with which comparison has been drawn. The proposed algorithm decreases the number of features selected by around 75% , which is better than most of the other algorithms.

Many processes are too complex to be manipulated quantitatively; however, humans succeed by using simple rules of thumb that are extracted from their experiences. Fuzzy logic emulates the human reasoning in the use of imprecise information to generate decisions. Unlike traditional approaches, which require a mathematical understanding of the system, fuzzy logic comprises an alternative way of processing, which permits modeling complex systems through the use of human knowledge. On the other hand, several new metaheuristic algorithms have recently been proposed with interesting results. Most of them use operators based on metaphors of natural or social elements to evolve candidate solutions. In this paper, a methodology to implement human-knowledge-based optimization strategies is presented. In the scheme, a Takagi-Sugeno Fuzzy inference system is used to reproduce a specific search strategy generated by a human expert. Therefore, the number of rules and its configuration only depend on the expert experience without considering any learning rule process. Under these conditions, each fuzzy rule represents an expert observation that models the conditions under which candidate solutions are modified in order to reach the optimal location. To exhibit the performance and robustness of the proposed method, a comparison to other well-known optimization methods is conducted. The comparison considers several standard benchmark functions which are typically found in scientific literature. The results suggest a high performance of the proposed methodology.

This paper presents two multi-criteria optimization techniques: the Multi-Objective Crow Search Algorithm (MOCSA) and an improved chaotic version called Multi-Objective Chaotic Crow Search Algorithm (MOCCSA). Both methods MOCSA and MOCCSA are based on an enhanced version of the recently published Crow Search Algorithm. Crows are intelligent animals with interesting strategies for protecting their food hatches. This compelling behavior is extended into a Multi-Objective approach. MOCCSA uses chaotic-based criteria on the optimization process to improve the diversity of solutions. To determinate if the performance of the algorithm is significantly enhanced, the incorporation of a chaotic operator is further validated by a statistical comparison between the proposed MOCCSA and its chaotic-free counterpart (MOCSA) indicating that the results of the two algorithms are significantly different from each other. The performance of MOCCSA is evaluated by a set of standard benchmark functions, and the results are contrasted with two well-known algorithms: Multi-Objective Dragonfly Algorithm and Multi-Objective Particle Swarm Optimization. Both quantitative and qualitative results show competitive results for the proposed approach.

We use an evolutionary mechanism to improve the image segmentation process.We optimize the minimum cross entropy with an evolutionary method for image segmentation.We test the approach in multidimensional spaces.An alternative method for MR brain image segmentation is proposed.Comparisons and non-parametric test support the experimental results. Segmentation is considered the central part of an image processing system due to its high influence on the posterior image analysis. In recent years, the segmentation of magnetic resonance (MR) images has attracted the attention of the scientific community with the objective of assisting the diagnosis in different brain diseases. From several techniques, thresholding represents one of the most popular methods for image segmentation. Currently, an extensive amount of contributions has been proposed in the literature, where thresholding values are obtained by optimizing relevant criteria such as the cross entropy. However, most of such approaches are computationally expensive, since they conduct an exhaustive search strategy for obtaining the optimal thresholding values. This paper presents a general method for image segmentation. To estimate the thresholding values, the proposed approach uses the recently published evolutionary method called the Crow Search Algorithm (CSA) which is based on the behavior in flocks of crows. Different to other optimization techniques used for segmentation proposes, CSA presents a better performance, avoiding critical flaws such as the premature convergence to sub-optimal solutions and the limited exploration-exploitation balance in the search strategy. Although the proposed method can be used as a generic segmentation algorithm, its characteristics allow obtaining excellent results in the automatic segmentation of complex MR images. Under such circumstances, our approach has been evaluated using two sets of benchmark images; the first set is composed of general images commonly used in the image processing literature, while the second set corresponds to MR brain images. Experimental results, statistically validated, demonstrate that the proposed technique obtains better results in terms of quality and consistency.

Clustering is one of the most popular methods of machine learning. The process of clustering involves the division of a set of abstract objects into a certain number of groups which integrated with objects of similar characteristics. Therefore, a cluster integrates objects which are similar to them, but dissimilar to the elements that belong to the rest of the clusters. Several clustering methods have proposed in the literature with different performance levels. All these techniques use as similarity criterion the Euclidean distance among cluster elements. However, there exist diverse scenarios where the Euclidean distance cannot be utilized appropriately to separate the elements in groups. Under such conditions, traditional cluster methods cannot directly apply. On the other hand, the operations of dilate and erode are a set of non-linear operators that modify the shape of a data group in the feature space, to obtain a monolithic object. Although morphological operations have demonstrated its importance in several engineering fields as image processing, its use as a clustering technique has been practically overlooked. In this work, an alternative clustering algorithm is proposed to group elements without considering the distance as a similarity criterion. In our approach, the data were separated into different groups by considering morphological operations. Under this scheme, the procedure allows the integration of data points, which present a spatial connection. Since the proposed algorithm does not use the distance in its functioning, it solves complex clustering problems which traditional clustering algorithms cannot.

1Departamento de Electronica, Universidad de Guadalajara, CUCEI, Avenida Revolucion 1500, 44430 Guadalajara, JAL, Mexico 2Departamento de Ingenieŕia de Software e Inteligencia Artificial, Facultad Informatica, Universidad Complutense, 28040 Madrid, Spain 3Centro Universitario de Tonala, Universidad de Guadalajara, Avenida Nuevo Periferico No. 555, Ejido San Jose Tatepozco, 48525 Tonala, JAL, Mexico 4Institut fur Informatik, Freie Universitat Berlin, Arnimallee 7, 14195 Berlin, Germany

Multilevel thresholding image segmentation has received considerable attention in several image processing applications. However, the process of determining the optimal threshold values (as the preprocessing step) is time-consuming when traditional methods are used. Although these limitations can be addressed by applying metaheuristic methods, such approaches may be idle with a local solution. This study proposed an alternative multilevel thresholding image segmentation method called VPLWOA, which is an improved version of the volleyball premier league (VPL) algorithm using the whale optimization algorithm (WOA). In VPLWOA, the WOA is used as a local search system to improve the learning phase of the VPL algorithm. A set of experimental series is performed using two different image datasets to assess the performance of the VPLWOA in determining the values that may be optimal threshold, and the performance of this algorithm is compared with other approaches. Experimental results show that the proposed VPLWOA outperforms the other approaches in terms of several performance measures, such as signal-to-noise ratio and structural similarity index.

The increasing use of renewable energy due to the excessive use of fossil fuels causing high quantities pollution has caused the growth of research fields in this area. One of the most important is the use of solar cells because of their unlimited source of power. The performance of a solar cell directly depends on its design parameters, so that, the solar cells parameter estimation is a complex task due to its non-linearity and high multimodality. Optimization techniques are widely used to solve complex problems efficiently.

Through the history, humans have been succeeded by solving multiple problems during their day to day life. They use simple rules of thumb from their past experiences to solve several difficulties. Under such circumstances, many researchers have tried to emulate the human reasoning based on mathematical approaches. Based on simple if-then rules, fuzzy logic is one of the disciplines in artificial intelligence which emulates the human reasoning in terms of linguistic variables. In fuzzy logic, linguistic variables represent natural language variables which humans commonly used to specify semantic rules from several processes. On the other hand, metaheuristics have been proposed as alternative search mechanisms to find optimal solutions for complex optimization problems where classical mathematical methodologies present some limitations by working under multimodal surfaces. This chapter presents a novel metaheuristic algorithms called Fuzzy Logic Optimization Algorithm (FLOA). The proposed algorithm models the search strategy which an expert human in optimization could follow to solve optimization problems based on simple if-then rules. The FLOA, uses a Takagi-Sugeno inference model, where the output is a weighted sum of four fuzzy rules; Attraction, repulsion, perturbation and randomness. The performance of the proposed method is compared against the performance results of several state-of-art metaheuristics, evaluating several test functions. The numerical results are statistical validated using a non-parametric framework to eliminate the random effect.

In recent years swarm intelligence emulate the behavior of insects or animal. In this chapter, an optimization algorithm called Locust Search (LS) is presented. The LS is inspired of the behavior of the locust swarms. In the algorithm consider two different behaviors: solitary and social. This tow types of behavior interact with each other in ways to allow find solution to a complex optimization problem. In order to illustrate the efficiency and robustness the LS was compared with other well-known optimization algorithms. The algorithm was proved with several benchmark functions.

For a wide-range of image processing tasks (such as feature tracking, object recognition, stereo matching and remote sensing), Template Matching (TM) serves as a crucial strategy to allow the localization and recognition of objects and/or patterns within digital images. The purpose of a TM approach is to find the location (region) within a source image which manifest the best possible resemblance to a given sub-image (commonly known as image template). Any TM algorithm (independently of their particularities) incorporates two important elements: 1. a search strategy; and 2. a similarity measurement. In its simplest form, a TM method involves exhaustive search processes were a similarity measurement (such as the Normalized Cross-Correlation) is applied for each available location within the source image. Intuitively, such an extensive amount of function evaluations has a strong impact on the algorithms computational cost, an issue that happens to be extremely detrimental for many real-world applications. In the last few years, Evolutionary Algorithms (EAs) have been proposed as an alternative to aid on the search process of TM approaches by reducing the amount of locations that are evaluated within a given source image; however, it is known that many of these methods carry with them several critical flaws which could negatively impact the TM process, including an insufficient exploration of the source image (which often leads to premature convergence). In this chapter, the swarm optimization algorithm known as Locust Search (LS) is applied to aid on solving the of template matching. The LS method includes a series of unique evolutionary operators that allows to explicitly avoid the concentration of search agents toward the best-known solutions, which in turn allows these agents to better explore of the available image’s search region. According to a series of experimental results, when compared to other TM approaches which integrates EAs as a part of their search strategy, LS achieves the best between estimation accuracy and computational load.

In a computer vision one problem is image segmentation as an alternative the problem has been handled whit optimization algorithm. Most of the methods have two difficulties (1) sub-optimal results (2) the number of classes is previously known. In this chapter presented an algorithm that automatic selection of this classes for image segmentation whit a new objective function in gaussian mixture model. The optimization algorithm called Locust Search (LS), is based on behavior of locust swarms, presented a balance between exploration and exploitation. The method was tested over several images to validate the efficacy over other techniques.

Swarm intelligence is a computer science field which emulates the cooperative behavior of natural systems such as insects or animals. Many methods resulting from these models have been suggested to solve several complex optimization problems. In this chapter, a metaheuristic approach known as the Social Spider Optimization (SSO) is analyzed for solving optimization problems. The SSO method considers the simulation of the collective operation of social-spiders. In SSO, candidate solutions represent a set of spiders which interacts among them based on the natural laws of the colony. The algorithm examines two different kinds of search agents (spiders): males and females. According to the gender, each element is conducted by a set of different operations which imitate different behaviors that are commonly observed in the colony.

After millions of years of evolution, nature has developed a wide variety of interesting structures, each with their own singularities and properties. Such structures provide several unique and innovative models which may be extended to solve complex engineering problems. One example of such structures is the so-called orb webs, built by many species of a spider as a part of their survival tactics. Orb webs are highly optimized structure, specifically devised to capture prey by efficiently covering a whole area with sticky threads. In this paper, a new feature descriptor called spider local image features (SLIF) is proposed. In the proposed approach, feature vectors are built by selectively extracting pictorial information from a set of previously detected interest point. This is achieved by considering a set of efficiently distributed sampling points, which emulate the intersection nodes formed by the threads of an orb web structure. The SLIF method produces simple low-dimensional feature descriptors, which are robust to several image transformation and distortions, such as scaling, rotation, bright shifts and viewpoint changes. In order to illustrate the proficiency and robustness of the proposed approach, it is compared to other well-known feature description methods, such as those presented on the scale-invariant feature transform, speeded-up robust features, binary robust scalable keypoints and fast retina keypoints. The comparison examines several different images, commonly considered as a benchmark within the image matching literature. Our experimental results evidence SLIF’s high performance and robustness against common image transformations and distortions and further show its viability for many of computer vision applications.

This paper proposes a new algorithm called Multimodal Flower Pollination Algorithm (MFPA). Under MFPA, the original Flower Pollination Algorithm (FPA) is enhanced with multimodal capabilities in order to find all possible optima in an optimization problem. The performance of the proposed MFPA is compared to several multimodal approaches considering the evaluation in a set of well-known benchmark functions. Experimental data indicate that the proposed MFPA provides better results over other multimodal competitors in terms of accuracy and robustness.

In this paper, a new segmentation method based on the mean shift (MS) algorithm is presented. The proposed approach divides the image into two sets of pixels: operative elements and inactive elements. In its first phase, the MS scheme considers only the operative elements. In its second stage, the results obtained by the MS method with the operative data are used to include the inactive data. During this stage, each inactive pixel is assigned to the cluster corresponding to the nearest operative pixel. As a final operation, clusters that maintain the minimal number of elements are blended with other nearby clusters. Our method has been tested against other current segmentation methods using test images extracted from the Berkley dataset. Numerical experiments demonstrate that our approach exhibits better performance in terms of consistency, quality, velocity and accuracy.

Agent-based modeling is a relatively new approach to model complex systems composed of agents whose behavior is described using simple rules. As a consequence of the agent interactions emerges a complex global behavioral pattern not explicitly programmed. In the last decade, an increasing number of metaheuristic techniques have been reported in the literature where authors claim their novelty and their abilities to perform as powerful optimization methods. Although these schemes emulate very different processes or systems, the rules used to model individual behavior are very similar. The idea behind the design of many metaheuristic methods is to configure a recycled set of rules that has demonstrated to be successful in previous approaches for producing new optimization schemes. Such common rules have been designed without considering the final global result obtained by the individual interactions. On the other hand, agent-based systems provide a solid theory and a set of consistent models that allow characterizing global behavioral patterns produced by the collective interaction of the individuals from a set of simple rules. Under this perspective, several agent-based concepts and models that generate very complex global search behaviors can be used to produce or improve efficient optimization algorithms. In this paper, a new metaheuristic algorithm based on agent systems principles is presented. The proposed method is based on the agent-based model known as “Heroes and Cowards”. This model involves a small set of rules to produce two emergent global patterns that can be considered in terms of the metaheuristic literature as exploration and exploitation stages. To evaluate its performance, the proposed algorithm has been tested in a set of representative benchmark functions, including multimodal, unimodal, and hybrid benchmark formulations. The competitive results demonstrate the promising association between both paradigms.

The techniques of halftoning are widely used in marketing because they reduce the cost of impression and maintain the quality of graphics. Halftoning converts a digital image into a binary image conformed by dots. The output of the halftoning contains less visual information, a possible benefit of this task is the reduction of ink when graphics are printed. The human eye is not able to detect the absence of information, but the printed image stills have good quality. The most used method for halftoning is called Floyd-Steinberger, and it defines a specific matrix for the halftoning conversion. However, most of the proposed techniques in halftoning use predefined kernels that do not permit adaptation to different images. This article introduces the use of the harmony search algorithm (HSA) for halftoning. The HSA is a popular evolutionary algorithm inspired by the musical improvisation. The different operators of the HSA permit an efficient exploration of the search space. The HSA is applied to find the best configuration of the kernel in halftoning, meanwhile, as an objective function, the use of the structural similarity index (SSIM) is proposed. A set of rules are also introduced to reduce the regular patterns that could be created by non-appropriate kernels. The SSIM is used due to the fact that it is a perception model used as a metric that permits comparing images to interpret the differences between them numerically. The aim of combining the HSA with the SSIM for halftoning is to generate an adaptive method that permits estimating the best kernel for each image based on its intrinsic attributes. The graphical quality of the proposed algorithm has been compared with classical halftoning methodologies. Experimental results and comparisons provide evidence regarding the quality of the images obtained by the proposed optimization-based approach. In this context, classical algorithms have a lower graphical quality in comparison with our proposal. The results have been validated by a statistical analysis based on independent experiments over the set of benchmark images by using the mean and standard deviation.

The increasing deployment of smart meters to collect load profile data in real-time is generating new opportunities for analysing electric power distribution system. Storing, managing and analysing large volumes of collected data, however, is challenging. Measured data is high-dimensional in nature and may contain hidden information and complex patterns that need to be interpreted. In this chapter, a method that combines dimensionality reduction (DR) technique with Particle Swarm Optimization (PSO) algorithm for clustering load profile electricity data is presented. The DR techniques allows to obtain a low-dimensional data model that can be used to project representative electricity load (REL) data onto an easily interpretable 3D space. The PSO algorithm and a validation index algorithm is then applied to obtain an optimal number of clusters. The presented framework methodology is applied to clustering historical REL data. The REL data allows to evaluate the ability of linear and nonlinear DR techniques to extract the relevant information that may be useful to visualize and improve the clustering data process.

As a consequence of increased electricity market, the distribution systems have been restated to consider alternative generation sources. These sources with a maximum capacity of less than 100 MW are defined as distributed generation (GD). Most DGs are usually connected to the electrical network without planned and dispatched control, since traditional distribution systems were not designed to consider any other generation sources.

Metaheuristic is a set of soft computing techniques which considers the design of intelligent search algorithms based on the analysis of several natural and social phenomena.

Several systems are extremely complicated to be handled quantitatively. In spite of this humans undergo them by using simplistic rules that are obtained from their own experiences.

Swarm intelligence (SI) is a research field which has recently attracted the attention of several scientific communities. An SI approach tries to characterize the collective behavior of animal or insect groups to build a search strategy. These methods consider biological systems, which can be modeled as optimization processes to a certain extent. The Social Spider Optimization (SSO) is a novel swarm algorithm that is based on the cooperative characteristics of the social spider. In SSO, search agents represent a set of spiders which collectively move according to the biological behavior of the colony. In most of SI algorithms, all individuals are modeled considering the same properties and behavior. In contrast, SSO defines two different search agents: male and female. Therefore, according to the gender, each individual is conducted by using a different evolutionary operation which emulates its biological role in the colony. This individual categorization allows reducing critical flaws present in several SI approaches such as incorrect exploration-exploitation balance and premature convergence. After its introduction, SSO has been modified and applied in several engineering domains. In this paper, the state of the art, improvements, and applications of the SSO are reviewed.

This paper presents a predictive control strategy for an image-based visual servoing scheme that employs evolutionary optimization. The visual control task is approached as a nonlinear optimization problem that naturally handles relevant visual servoing constraints such as workspace limitations and visibility restrictions. As the predictive scheme requires a reliable model, this paper uses a local model that is based on the visual interaction matrix and a global model that employs 3D trajectory data extracted from a quaternion-based interpolator. The work assumes a free-flying camera with 6-DOF simulation whose results support the discussion on the constraint handling and the image prediction scheme.

As an alternative to classical techniques, the problem of image segmentation has also been handled through evolutionary methods. Recently, several algorithms based on evolutionary principles have been successfully applied to image segmentation with interesting performances. However, most of them maintain two important limitations: (1) they frequently obtain suboptimal results (misclassifications) as a consequence of an inappropriate balance between exploration and exploitation in their search strategies; (2) the number of classes is fixed and known in advance. This paper presents an algorithm for the automatic selection of pixel classes for image segmentation. The proposed method combines a novel evolutionary method with the definition of a new objective function that appropriately evaluates the segmentation quality with respect to the number of classes. The new evolutionary algorithm, called Locust Search (LS), is based on the behavior of swarms of locusts. Different to the most of existent evolutionary algorithms, it explicitly avoids the concentration of individuals in the best positions, avoiding critical flaws such as the premature convergence to suboptimal solutions and the limited exploration-exploitation balance. Experimental tests over several benchmark functions and images validate the efficiency of the proposed technique with regard to accuracy and robustness.

Machine learning has been widely used in image analysis and processing for the purpose of letting the computer recognize specifics aspects like color, shape, textures, size, and position. Such procedures allow to have algorithms capable of identifying objects, find relationships, and perform tracking. The present chapter executes an analysis of the one image attribute that is listed among the basic aspects of an image and the color. The color is chosen due to the importance given by humans to this attribute. Also the color is used main filter criteria in object recognition and tracking. In this section the effect of the color is studied from the point of view of the most popular color spaces available in image processing. It will be tested the effect of the selection of a given color space one of the most common clustering machine learning algorithms. The chosen algorithm is K-means ++, a variation of the popular K-means, which allows a more fair evaluation since it mitigates some of the randomness in the final cluster configuration. Every advantage or weakness will be exposed, so it can be known what color spaces are the right choice depending on the desired objective in the image processing.

In order to improve the performance of solar energy systems, accurate modeling of current vs. voltage ( I – V ) characteristics of solar cells has attracted the attention of various researches. The main drawback in accurate modeling is the lack of information about the precise parameter values which indeed characterize the solar cell. Since such parameters cannot be extracted from the datasheet specifications, an optimization technique is necessary to adjust experimental data to the solar cell model. Considering the I – V characteristics of solar cells, the optimization task involves the solution of complex non-linear and multi-modal objective functions. Several optimization approaches have been proposed to identify the parameters of solar cells. However, most of them obtain sub-optimal solutions due to their premature convergence and their difficulty to overcome local minima in multi-modal problems. This paper proposes the use of the ABC (artificial bee colony) algorithm to accurately identify the solar cells' parameters. The ABC algorithm is an evolutionary method inspired by the intelligent foraging behavior of honey bees. In comparison with other evolutionary algorithms, ABC exhibits a better search capacity to face multi-modal objective functions. In order to illustrate the proficiency of the proposed approach, it is compared to other well-known optimization methods. Experimental results demonstrate the high performance of the proposed method in terms of robustness and accuracy.

In this paper, we will present a mathematical analysis of the transition proportion for the normal threshold (NorT) based on the transition method. The transition proportion is a parameter of NorT which plays an important role in the theoretical development of NorT. We will study the mathematical forms of the quadratic equation from which NorT is computed. Through this analysis, we will describe how the transition proportion affects NorT. Then, we will prove that NorT is robust to inaccurate estimations of the transition proportion. Furthermore, our analysis extends to thresholding methods that rely on Bayes rule, and it also gives the mathematical bases for potential applications of the transition proportion as a feature to estimate stroke width and detect regions of interest. In the majority of our experiments, we used a database composed of small images that were extracted from DIBCO 2009 and H-DIBCO 2010 benchmarks. However, we also report evaluations using the original (H-)DIBCO׳s benchmarks.

Clustering involves the process of dividing a collection of abstract objects into a number of groups, which are integrated with similar elements. Several clustering methods have been introduced and developed in the literature with different performance levels. Among such approaches, density algorithms present the best advantages, since they are able to find clusters from a dataset under different scales, shapes, and densities without requiring the number of groups as input. On the other hand, there are processes that humans perform much better than deterministic approaches or computers. In fact, humans can visually cluster data exceptionally well without the necessity of any training. Under this unique capability, clustering, instantaneous and effortless for humans, represents a fundamental challenge for artificial intelligence. In this paper, we develop a simple clustering model inspired by the way in which the human visual system associates patterns spatially. The model, at some abstraction level, can be characterized as a density grouping strategy. The approach is based on Cellular Neural Networks (CNNs) which have demonstrated to be the best models for emulating the human visual system. In the proposed method, similar to the biological model, the CNN is used to build especially groups while an automatic mechanism tries different resolution scales to find the best possible data categorization. Different datasets have been adopted to evaluate the performance of the proposed algorithm. Their results are also compared with popular density clustering techniques from the literature. Computational results demonstrate that the proposed CNN approach presents competitive results in comparison with other algorithms regarding accuracy and robustness.