Search

Diabetic retinopathy, age-related macular degeneration and glaucoma, are the leading causes of visual impairment or blindness of the population across different ages. Retinal fundus imaging is a clinically regular tool for the diagnosis of retinal diseases. In the interest of having a comprehensive understanding of the fundus condition, it is valuable to leverage multiple fundus images from different modalities. However, a direct fusion of the multi-source fundus images eases to mis-align the physiological structure or spatial position due to possible eyeball rotations or head movements. The problem turns out to be more severe if the images were corrupted by ill conditions on eyes, such as micro-bleeding and plaques. To tackle this problem, we propose a multi-source registration model for retinal fundus images. Our proposed method considers multiple correspondences and dual structural constraints during the registration process. The method firstly selects adequate feature points by an adjustable threshold selection strategy. Then a feature-guided correspondence estimation model is established to build complementary features. Finally, their spatial transformation is built by using mean shift evolution. The evolution is guided by Tikhonov regularization on dual geometric structures. It overcomes the mess of mean shift vector field and mitigating the ill-posed displacement in field recovery. We have conducted our method on the collected 220 multi-source retinal fundus image pairs, which involve minor and larger displacement or severe retinopathy lesions, as well as additive different intensities of Gaussian noises. Extensive experiments demonstrate that the proposed method consistently outperforms seven feature-based methods.

A smart healthcare sensor cloud is an amalgamation of the body sensor networks and the cloud that facilitates the early diagnosis of diseases and the real-time monitoring of patients. Sensitive data of the patients which are stored in the cloud must be free from outliers that may be caused by malfunctioned hardware or the intruders. This paper presents a machine learning-based scheme for outlier detection in smart healthcare sensor clouds. The proposed scheme is a hybrid of clustering and classification techniques in which a two-level framework is devised to identify the outliers precisely. At the first level, a density-based scheme is used for clustering while at the second level, a Gaussian distribution-based approach is used for classification. This scheme is implemented in Python and compared with a clustering-based approach (Mean Shift) and a classification-based approach (Support Vector Machine) on two different standard datasets. The proposed scheme is evaluated on various performance metrics. Results demonstrate the superiority of the proposed scheme over the existing ones.

Aerospace has become an indispensable part of exploring the universe. To guarantee the safety of the materials on aerospace, the recognition and the detection of hypervelocity impact (HVI) damage is needed. Processing bottlenecks and the lack of precision acquisition tools have restricted detection to the HVI damage of space debris for on-orbit spacecraft. This paper proposes an optimal multi-target image segmentation based feature extraction method to detect the HVI damage. For the collected infrared data, an improved mean shift clustering algorithm is given to adaptively classify the temperature change characteristics in the sampled data of the infrared video stream and reconstruct the image to obtain the infrared reconstruction image reflecting the damage characteristics. Moreover, a multi-objective optimization segmentation model is proposed to segment the defects of infrared reconstructed images to further extract the geometric shape, depth and other features of various defects. To obtain accurate segmentation results of the infrared reconstructed thermal images, the segmentation objective functions are set with respect to noise elimination and detail reservation. A weight vectors adjustment algorithm is developed to further improve the accuracy of damage segmentation in the multi-objective optimization segmentation model, to derive better balance between detail retention and noise removal. The experimental results of the algorithm verify the effectiveness and benefits of the proposed method.

With the rapid development of the Internet platform, the Internet economy has become a new situation for evaluating the poverty/investment slow effect. What has brought development to the Internet economy is a variety of e-commerce platforms. However, in the face of huge Internet users and different users’ shopping habits, reasonably personalized e-commerce platform recommendations for users can improve the user’s website shopping experience. However, due to the huge user data and the diversity of e-commerce platforms, it is a huge challenge to reasonably recommend e-commerce platforms to users. The quality of the e-commerce’s personalized recommendations also affects the purchase conversion rate. Clustering algorithm is an algorithm involved in grouping data in machine learning. The same set of data has the same attributes and characteristics, and the attributes or features between different sets of data will be relatively large. In this paper, by using the mean shift clustering algorithm to characterize the behavior data. According to the characteristics of the grouping, we can recommend the e-commerce platform commonly used between the groups. However, using mean shift clustering for personalized recommendation faces the problem of too high user data dimensions. Therefore, we first conduct behavioral analysis of user data by leveraging the well-known Hopfield Neural Network. We define user behavior sequences for user behavior data and classify user behavior. We transform the grouped user behavior into an embedded vector, and linearly transform the embedded vectors of different lengths into the same semantic space. We process the vectors in the semantic space through the self-attention layer and perform mean shift clustering. Experiments show that our method can reduce the complexity of user poverty distribution effect toward complex data and improve the quality of personalized recommendation.

In this paper, the Probabilistic Mapped Mean-Shift Algorithm is proposed to detect anomalous data in public datasets and local hospital children’s wellness clinic databases. The proposed framework consists of two main parts. First, the Probabilistic Mapping step consists of k-NN instance acquisition, data distribution calculation, and data point reposition. Truncated Gaussian Distribution (TGD) was used for controlling the boundary of the mapped points. Second, the Outlier Detection step consists of outlier score calculation and outlier selection. Experimental results show that the proposed algorithm outperformed the existing algorithms with real-world benchmark datasets and a Children’s Wellness Clinic dataset (CWD). Outlier detection accuracy obtained from the proposed algorithm based on Wellness, Stamps, Arrhythmia, Pima, and Parkinson datasets was 93%, 94%, 80%, 75%, and 72%, respectively.

In this paper, we introduce a methodology for the detection and segmentation of automobiles in urban scenarios. We use the LiDAR Velodyne HDL-64E to scan the surroundings. The method is comprised of three steps: (1) remove facades, ground plan, and unstructured objects, (2) smoothing data using robust principal component analysis (RPCA), and finally, (3) unstructured objects model and indexing. The dataset is partitioned into training with 4500 objects and test with 3000 objects. Mean Shift thresholds, the filter, the Delaunay parameters, and the histogram modelling are optimized via ROC analysis. It is observed that the car scan quality affects our method to a lesser degree when compared with state-of-the-art methods.

As the internet and data mining technologies are developing rapidly, how to provide various students with high-quality education services has become the hotspot in the internet environment. In order to promote the characteristics of online education and enhance the quality of personalized e-learning, in this paper, we propose a novel algorithm named MK-means by exploiting the cluster-wise weighing co-association matrix mechanism and improving the K-means algorithm based on the mean shift theory. The experimental results on the UCI’s Iris and Wine test sets demonstrate its effectiveness and superiority, finding that the total F-measure of MK-means achieves better performance than the Hierarchical Clustering, FCM, K-means, SOM, and X-means algorithms. Finally, the new algorithm combined with the student model explains the clustering results in detail from perspectives of cognitive model and knowledge map respectively and can extend to support the personalized e-learning in a wide range.

Image segmentation has been an important procedure in object-based image analysis (OBIA), which takes image object as a processing unit. The spatial scale in image segmentation has great importance in OBIA. Due to the high heterogeneity and large dynamic range of polarimetric synthetic aperture radar (PolSAR) images, it is often difficult to choose optimal spatial scales. This letter proposes a polarimetric semivariogram-based spatial scale selection method for PolSAR image segmentation. The optimal spatial bandwidth parameter in the mean-shift algorithm is pre-estimated based on the combined polarimetric and statistical analysis of PolSAR images. By implementing a quantitative evaluation of segmentation result, the effectiveness of the proposed method in optimal spatial bandwidth selection for PolSAR image segmentation is verified. Experiments on both the EMISAR and UAVSAR L-band PolSAR data sets testify the validity of the proposed adaptive optimal bandwidth selection strategy for PolSAR images.

Least squares support vector machine (LSSVM) is a machine learning algorithm based on statistical theory. Itsadvantages include robustness and calculation simplicity, and it has good performance in the data processingof small samples. The LSSVM model lacks sparsity and is unable to handle large-scale data problem, this articleproposes an LSSVM method based on mixture kernel learning and sparse samples. This algorithm reduces theinitial training set to a sub-dataset using a sparse selection strategy. It converts the single kernel function in theLSSVM model into a mixed kernel function and optimizes its parameters. The reduced sub-dataset is used fortraining LSSVM. Finally, a group of datasets in the UCI Machine Learning Repository were used to verify theeffectiveness of the proposed algorithm, which is applied to real-world power load data to achieve better fittingand improve the prediction accuracy.

This paper introduces a method based on centroid clustering and contour analysis to extract area and height measurements on buildings from the 3D model generated by oblique images. The method compr...

Moving object tracking is a computation-intensive operation that requires accelerating hardware solution. In this work, a high-performance design for mean shift based moving object tracking algorithm and its FPGA implementation is done. Here, associated circuits are utilized as intellectual-property cores to implement an embedded system-on-a-chip (SoC) framework for real-time moving object tracking application. Real-time video with $$640 \times 480$$ resolution at 60 frames per second is captured and buffered in SDRAM, and processing is performed on the temporal frames. In the implemented FPGA-SoC framework, PowerPC processor embedded inside the FPGA device is used for the platform configuration, IPs control, and running the application program. The design require 30.08% slices, 35.14% BRAMs, and 43.75% DSP48E slices of Xilinx Virtex-5 xc5vfx70t FPGA device on the ML-507 platform. The computed power is 48.3 mW.

Unman aerial vehicle (UAV) LiDAR has been widely used in the field of forestry. Individual tree extraction is a key step for forest inventory. Although many individual tree extraction methods have been proposed, the individual tree extraction accuracy is still low due to the complex forest environments. Moreover, many parameters in these methods generally need to be set. Thus, the degree of automation of the methods is generally low. To solve these problems, this paper proposed an automatic mean shift segmentation method, in which the kernel bandwidths can be calculated self-adaptively. Meanwhile, a hierarchy mean shift segmentation technique was proposed to extract individual tree gradually. A plot-level UAV LiDAR tree dataset was adopted for testing the performance of the proposed method. Experimental results showed that the proposed method can achieve better individual tree extraction result without any parameter setting. Compared with the traditional mean shift segmentation method, both the completeness and mean accuracy of the proposed method are higher.

An abnormal rise in glucose levels may lead to diabetes. Around 30 million people are diagnosed with this disease in our country. In this perspective Indian Council of Medical Research funded by Registry of People with diabetes in India have taken an initiative and come up with numerous solutions but unfortunately neither of them has taken shape. Initially, the behavior of chemical reaction between glucose with chemical agent is estimated and tracked in the region of interest via mean shift algorithm using spatial and range information. This color change is related to plasma glucose concentration (plas), diastolic blood pressure, (pres.) Triceps skin fold thickness(skin), 2_hour serum insulin(insu), Body mass index and age. These features obtained from these 768 instances are classified using Naïve Bayes Algorithm. The results are compared with our previous work, an integrated system of K means and Naïve Bayes approach in terms of sensitivity, specificity, precision, and F-measure. It is worth noticing that our integration of mean-shift clustering and classification gives promising results with an utmost accuracy rate of 99.42% even after removing nearby duplicates in predefined clusters.

English text is difficult to recognize under the interference of blurred background, so it is necessary to improve the fixed-point tracking technology of English text. Based on machine learning algorithms, this paper studies the fixed-point tracking model of English reading text based on mean shift and multi-feature fusion. The target tracking algorithm based on mean shift obtains the description of the target model and the candidate model by calculating the pixel feature probability in the target area and the candidate area. Then, it uses the similarity function to measure the similarity between the initial frame target model and the current candidate model, selects the candidate model that maximizes the similarity function, and obtains the target model mean offset vector. Finally, it continuously iteratively calculates the offset vector based on this vector, and finally converges to the true position of the target, thereby achieving the effect of tracking. In general, it is verified that the model constructed in this paper works well through control experiments.

Change detection (CD) for multitemporal remote sensing images can find change trends and reveal the development patterns. However, the typical methods based on algebraic operation, image transformation, or segmentation may not yield satisfactory results due to the spectral variability and noise complexity. In a sense, it can be considered that multitemporal images are composed of unchanged and changed regions, as well as noise. In the light of this, a stepwise subtraction method for CD is proposed, based on auto-updating multitemporal matrix factorization (AMMF) and spectral constraint, to remove the unchanged regions and noise from the original images step by step. The unchanged regions are first identified by AMMF, during which the distribution and subspace information of each temporal image are regularized to encode the spatio-temporal correlation. Then, mean shift smoothness is adopted as a spectral constraint so as to remove the noise. In this way, the changed regions have been highlighted so that the change map can be obtained by a postsegmentation method. Experiments have been conducted on three multitemporal data sets, including images from Quick Bird, aerial, and GF-1, indicating that the proposed method is effective and robust, which is superior to some state-of-the-art methods.

Accurate segmentation of infrared thermal images is a challenging issue in the breast cancer detection. This paper presents and evaluates novel segmentation method for breast cancer detection using...

Genetically modified crops have been planted commercially on a large scale since 1996. However, the food safety issue of genetically modified crops remains controversial. Conventional genetically modified crops’ detection methods require a plenty of detective time and complex operations that cannot rapidly identify. Previous reports show that combining terahertz time-domain spectroscopy and supervised learning has advanced to identify genetically modified crops, but supervised learning requires large data to train the model. To solve the above problem, we proposed an unsupervised learning method, PCA-mean shift, to identify genetically modified crops. Principal component analysis was employed to reduce the absorbance data dimensionality. After principal component analysis, the first three principal components were used as the input of mean shift. At last, our proposed method had 100% identification accuracy, and K-means had 98.75% identification accuracy. The comparison results demonstrated that PCA-mean shift outperforms K-means. Therefore, PCA-mean shift combined with terahertz time-domain spectroscopy is a potential identification tool for genetically modified crops’ identification.

Network traffic processing is an automated method for arranging and optimizing network traffic, based on the parameters. The traffic data is gathered to begin the study of the component of network traffic. Subsequently, the clustering and grouping process is carried out to evaluate network traffic. Continuous evaluation of the patterns of network traffic remained a daunting challenge during traffic classification. However, existing approaches have not been able to reduce time consumption and improve clustering accuracy for network traffic analysis. In order to resolve these problems, a Density-based Mean Shift Clustering and Deep Packet Inspection Classification (DMSC-DPIC) methodology is implemented to perform an efficient network traffic analysis. In addition, the classification model DPI has been developed to identify network Traffic by payloading data points with minimum time as real as well as non-real-time traffic. In the DPI classification model, data points are grouped into various groups by analyzing associated points throughout the session. The experimental assessment of the proposed methodology DMSC-DPIC is carried out with the CAIDA anonymized Internet Traces Dataset and achieves improved efficiency compared with state-of-the-art work in terms of clustering precision, classification time and communications overhead.

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.

Computer-based fabric segmentation has recently increased significantly in diverse fields of which textiles design and engineering is no exception. Extracting appropriate information from printed f...

An abundance of various segmentation techniques are available in the literature, that cater to wide range of image understanding applications. The paper proposes a unified way of systematic categorization of the research work on image segmentation called Multi-Faceted Hierarchical Image Segmentation Taxonomy (MFHIST), which consist of six facets presented in a hierarchical manner - scope, requirement, control, feature, image representation and approach specifications. Every scope is exemplified with research works from the literature for better understanding. The paper gives an illustration of populating MFHIST, to provide the reader a quick grasp of few important state-of-art image segmentation research works and their adaptations. As a case study, the illustrations display a limited version to uncover the journey of basic to modern adaptations in the areas region based segmentation approach, such as Markov Random Fields, Spectral Clustering, Active Contour Model, Mean Shift Clustering. The other segmentation approaches have not been considered here, owing to the enormous volume of works in the past four to five decades and limitation in articulating all of them using MFHIST. The performance analysis of the algorithms using quantitative metrics is not in the present scope and will be considered in future version of MFHIST.

Segmentation of the breast ultrasound (BUS) image is an important step for subsequent assessment and diagnosis of breast lesions. Recently, Deep-learning-based methods have achieved satisfactory performance in many computer vision tasks, especially in medical image segmentation. Nevertheless, those methods always require a large number of pixel-wise labeled data that is expensive in medical practices. In this study, we propose a new segmentation method by dense prediction and local fusion of superpixels for breast anatomy with scarce labeled data. First, the proposed method generates superpixels from the BUS image enhanced by histogram equalization, a bilateral filter, and a pyramid mean shift filter. Second, using a convolutional neural network (CNN) and distance metric learning-based classifier, the superpixels are projected onto the embedding space and then classified by calculating the distance between superpixels’ embeddings and the centers of categories. By using superpixels, we can generate a large number of training samples from each BUS image. Therefore, the problem of the scarcity of labeled data can be better solved. To avoid the misclassification of the superpixels, $K$ -nearest neighbor (KNN) is used to reclassify the superpixels within every local region based on the spatial relationships among them. Fivefold cross-validation was taken and the experimental results show that our method outperforms several often used deep-learning methods under the condition of the absence of a large number of labeled data (48 BUS images for training and 12 BUS images for testing).

Image enhancement corresponds to processing an image to obtain an image with more perceptual details. In this paper, a social spider optimization (SSO) based scheme is proposed to generate an enhanced image which contains higher contrast and minimum change of entropy with respect to the original image. The proposed method employs histogram equalization with a modified cumulative distribution function to obtain a mapping function. A three-step process is followed to modify the original histogram. The first step is to segment the image histogram into two parts using the Otsu's thresholding method. Then both of the upper and lower histograms are weighted as well as thresholded to control the level of enhancement. The constraint parameters for modification are obtained by SSO. After applying the constrain parameters on the histograms, mean shift correction is performed to ensure there is a minimum level of mean shift from input image to output image. The results indicate that proposed method achieves better color preservation, and balanced contrast enhancement in comparison to existing techniques. The proposed scheme also leads to significant feature enhancement, low contrast boosting, and brightness preservation in the enhanced image, while preserving the natural feel of the original image.

The article is an overview of the direction of graphic image processing based on clustering algorithms. The analysis of prospects of application of algorithms of cluster analysis in digital image processing, in particular, at segmentation and compression of graphic images, and also at recognition of images in transport sphere of activity is carried out. Comparative modeling of such algorithms of cluster analysis as K-means, Mean-Shift (clustering of average shift) and DBSCAN (based on density of spatial clustering for applications with noise) on various types of data is carried out. The simulation was performed on synthetic datasets in a Jupyter Notebook environment using the Scikit-learn library. In particular, four data sets were generated in this environment, to which these clustering algorithms were applied. The simulation results showed that the K-means algorithm can effectively describe relatively simple shapes. In contrast, the mean shift does not require assumptions about the number of clusters and the shape of the distribution, but its performance depends on the choice of scale parameters. The DBSCAN algorithm can successfully detect more complex shapes, which emphasizes one of the strengths of this algorithm - the clustering of arbitrary data. The disadvantages of the selected algorithms are also given and it is indicated on which types of images they effectively work with the estimation of computational speed.

In this paper, an Interactive Teaching Learning Based Optimization (In-TLBO) algorithm is proposed for tracking multiple objects with several challenges. The performance of the four other competitive approaches, such as the Mean Shift (MS), Particle Swarm Optimization (PSO), Sequential PSO (SPSO), and Adaptive Gaussian Particle Swarm Optimization (AGPSO) are investigated for comparison. The quantitative and qualitative analyses of these approaches have been performed to demonstrate their efficacy. The comparison of various performance measures includes the convergence rate, tracking accuracy, Mean Square Error (MSE) and coverage test. To assess the dominance of the proposed In-TLBO approach, Sign and Wilcoxon test are also performed. These two non-parametric tests reveal considerable advancement of the proposed Interactive TLBO (In-TLBO) over other four competitive approaches. In-TLBO shows significant improvement over the MS and PSO algorithms with a level of significance α = 0.05, and over SPSO, with a level of significance α = 0.1 by considering detection rate as winning parameter. The analyses of comparative results demonstrate that the proposed approach effectively tracks similar objects in the presence of many real time challenges.

Hyperspectral anomaly detection methods based on representation model have attracted much attention in recent years. In the method, a background dictionary is used to represent each pixel linearly, and the residual is taken as the abnormal level of the pixel. Therefore, building a functional model with strong representation ability and a matching background dictionary is an important factor for success of the method. In the existing methods, the lack of feature utilization, the contamination of background dictionary and the dependence on prior knowledge lead to the instability of anomaly detection results, which are difficult to be applied in practice. To address the issues, this paper proposed a novel hyperspectral anomaly detection method which consists of two interconnected components: a new anomaly detection function model through the combination of LRR and CR, and the supporting background dictionary which does not need prior knowledge. The new anomaly detection model decomposed a two-dimensional normalized hyperspectral image into background and anomaly components, and reconstructed the background through a background dictionary and the corresponding coefficient matrix. The representation coefficient matrix was constrained by global low-rank and local collaborative attributes which helps background modeling. A distance weight matrix was also included to enhance the dense representation of background dictionary atoms which are similar to the testing pixel. The anomaly part was constrained by column sparsity simultaneously due to the global sparsity of anomalous targets in hyperspectral images. To further clarify the physical meaning of the model, another version that imposes non-negative and sum-to-one constraints on coefficient matrix was also proposed. The supporting background dictionary was designed for practical and reliable purposes and was implemented by dual mean shift clustering which automatically estimates parameter without prior-knowledge. The experimental results show that the proposed method can effectively improve the result of anomaly detection, and has the best ROC, AUC, and separation degree between background and anomalies for four real datasets. The average AUC of four datasets of the proposed algorithm is 2.6% higher than the method based on CR and 3.2% higher than the method based on LRR. Moreover, the proposed algorithm has stability and reliability under complex background, as well as the application on a larger hyperspectral scene shows that it offers great potential in practical use.

Single molecule localization microscopy (SMLM) techniques transcend the diffraction limit of visible light by localizing isolated emitters sampled stochastically. This time-lapse imaging necessitates long acquisition times, over which sample drift can become large relative to the localization precision. Here we present a novel, efficient, and robust method for estimating drift using a simple peak-finding algorithm based on mean shifts that is effective for SMLM in 2 or 3 dimensions.

Computer vision technology can automatically detect and recognize objects on the ground or on a court, such as balls, players, and lines, using camera sensors. These are non-contact sensors, which do not interfere with an athlete’s movement. The game elements detected by such measuring equipment can be used for game analysis, judgment, context recognition, and visualization. This paper proposes a method to automatically track the position of stones in curling sport images using computer vision technology. The authors extract the optimal feature vector of the mean-shift tracking algorithm by obtaining the optimal histogram from the color and edge information of the curling stone, thereby adaptively controlling the number of bins in the histogram. After evaluating the performance of the curling stone tracking method among 1424 image frames from curling sport videos, the authors found that the proposed method improved detection rate (overlap threshold = 0.9) by 14.85% compared to the general mean-shift method.

We study properties of the mean shift (MS)-type algorithms for estimating modes of probability density functions (PDFs), via regarding these algorithms as gradient ascent on estimated PDFs with adaptive step sizes. We rigorously prove convergence of mode estimate sequences generated by the MS-type algorithms, under the assumption that an analytic kernel function is used. Moreover, our analysis on the MS function finds several new properties of mode estimate sequences and corresponding density estimate sequences, including the result that in the MS-type algorithm using a Gaussian kernel the density estimate monotonically increases between two consecutive mode estimates. This implies that, in the one-dimensional case, the mode estimate sequence monotonically converges to the stationary point nearest to an initial point without jumping over any stationary point.

With the emergence of consumer RGB-D sensors, discriminative modeling has been shown to perform well in estimating human body pose. However, articulated hand pose estimation remains a challenging problem, mostly due to its high flexibility, occlusions, noisy data, and small area of the fingertips. In this paper, we present an efficient discriminative-based scheme to improve the performance of hand pose estimation from a single depth image. The proposed scheme is inspired by decision forest-based framework, but with several well-motivated modifications. Specifically, we propose a method to estimate 2D in-plane orientation of the hand, which is then utilized to enforce the depth comparison features and make them invariant to in-plane rotation. Subsequently, we investigate the use of random decision forests (RDF) and mean shift algorithm to predict a primary version of hand parts and joint locations. Based on this primary prediction, an adaptive spatial clustering method is applied to correct the misclassified regions, and to deliver the final estimation of hand pose. Along with the proposed scheme, we further develop a new set of highly-distinctive features for static finger sign recognition by utilizing the estimated hand pose configurations and RGB information. The proposed features are straightforward and can effectively capture different aspects of hand pose, such as links from each joint to the closest joints and orientation of each hand part. Extensive experiments on several challenging datasets demonstrate that our approach, compared to decision forest-based methods, is able to provide more precise estimation of hand poses (with up to 21% improvement in joint localization accuracy), and can efficiently recognize more complex static finger signs (93.85% mean recognition accuracy on a challenging 34-finger sign dataset). Our approach is also robust to illumination, inter-hand occlusion, scale, and rotation variance.

Accurate recognition of green fruit targets is one of the key technologies for fruit growth monitoring and yield estimation. To solve the problem of fruit misidentification due to the similarity between fruit skin and leaf colors, a progressive detection method of green apples in natural environments was proposed. Image enhancement based on fuzzy set theory was carried out to make the fruit targets more salient in the whole image. Then, the fruit areas were roughly determined by the attention-based information maximization (AIM) algorithm, and the recognized apple regions were cropped according to the adaptive pixel-extending method to remove the background information. After that, accurate segmentation of fruit targets was accomplished by fusing the illumination-invariant image and R-component of the cropped image. To evaluate the performance of this method, it was compared with the illumination invariance theory-based algorithm, mean shift algorithm, K-means clustering algorithm, manifold ranking algorithm and GrabCut algorithm. The test was conducted using 200 green apple images under different growth statuses. Experimental results showed that the segmentation rate of the proposed method was 86.91%, which was 3.26%, 6.35%, 16.43%, 3.08% and 4.7% higher than those of the other five methods, respectively. The false positive rate and false negative rate were 0.88% and 10.53%, which gained an advantage over those of the other five segmentation algorithms. The localization error was 3.65%. In conclusion, the proposed method can accurately segment green fruit targets, which can lay the foundation for intelligent management of fruits over the entire growing season.

Accurate left ventricle landmark localization in cardiac MRI plays a vital role in computer-aided diagnosis of heart disease. Typical classification models hardly deal with artifacts and low discrimination of landmark regions by using only local image information, while regression models suffer from ambiguity between random samples and label, also the imbalance of samples. To overcome this limitation, this paper proposes a left ventricle landmark localization and identification method in cardiac MRI based on deep distance metric learning and CNN (convolutional neural network) regression. The method includes sample generation and regression stages, where super-pixel over-segmentation and unsupervised deep metric learning are integrated to extract the embedding information of local images, then a dual-channel salient sample mining module is designed by integrating specific super-pixel patches and grid patches extracted by the embedded triplet network. The weights of this triplet network are fed to build the CNN regression model, and the salient samples are employed to predict the landmark coordinate point cloud clusters. Furthermore, the centroid of each point cloud cluster is refined as landmark by Mean Shift iteration. The proposed method and close related methods were thoroughly evaluated on the CAP (cardiac atlas project) data set, and experimental results show that the proposed method achieves the competitive accuracy and outperforms the state-of-the-art classification and regression-based models.

Background: Current Augmented Reality systems in liver and bowel surgeries, are not accurate enough to classify the hidden parts such as gallbladder and uterus which are behind the liver and bowel. Therefore, we aimed to improve the visualization accuracy of bowel and liver augmented videos to avoid the unexpected cuttings on the hidden parts. Methodology: The proposed system consists of an Enhanced hybrid recursive matching and λ-parameterization techniques to improve the visualization. In addition, Mean Shift Filter is also added to improve the matching process while image registration. Results: Results proved that, the accuracy is improved in terms of liver and bowel surgeries Visualization errors about 0.53 mm and 0.22 mm respectively. Similarly, it can produce 2 more frames/sec compared to the current system. Conclusion: The proposed system worked towards the visualization of gallbladder and uterus while liver and bowel surgeries. So, this study solved the visualization issues, which are caused by neighbouring and hidden parts.

Online short-term rental platforms, such as Airbnb, have been becoming popular, and a better pricing strategy is imperative for hosts of new listings. In this paper, we analyzed the relationship between the description of each listing and its price, and proposed a text-based price recommendation system called TAPE to recommend a reasonable price for newly added listings. We used deep learning techniques (e.g., feedforward network, long short-term memory, and mean shift) to design and implement TAPE. Using two chronologically extracted datasets of the same four cities, we revealed important factors (e.g., indoor equipment and high-density area) that positively or negatively affect each property's price, and evaluated our preliminary and enhanced models. Our models achieved a Root-Mean-Square Error (RMSE) of 33.73 in Boston, 20.50 in London, 34.68 in Los Angeles, and 26.31 in New York City, which are comparable to an existing model that uses more features.

Feature extraction and fault recognition of vibration signals are two important parts of bearing fault diagnosis. In this article, a fault diagnosis method based on teager energy entropy of each wavelet subband and improved fuzzy C-means is proposed. First, bearing vibration signal is decomposed into wavelet packet and normalized teager energy entropy feature matrix is constructed as clustering index. Principal component analysis is applied to the high-dimensional teager energy entropy feature matrix, and the principal components are determined by cumulative contribution rate to construct feature vectors. Then, the mean-shift method is used to search for the high probability density region of principal components so as to determine the cluster number and cluster center. Finally, fuzzy C-means is used to update the clustering center and membership value, and confirm the optimal clustering center and the type of clustering. Through simulated and experimental analysis, the proposed method has two advantages. The feature vector constructed by this method has better specificity than wavelet energy entropy. The initial clustering center of fuzzy C-means is confirmed by the mean-shift method, which can improve the clustering performance of fuzzy C-means and solve the misclassification without preknowing the number of categories.

Visual tracking is an art of tracking a moving object over video frames using non-stationary cameras, for which feature descriptors of the target are computed and applied to a motion model. In this paper, SURF is used to compute the feature descriptors and Mean-Shift algorithm as a motion model. Mean-shift algorithm is fundamentally a logical approach to track the object on an image frame where the appearance is described by histograms. Mean-Shift tracking can directly be applied to SURF features but there is a big constraint of availability of an adequate number of feature keypoints for a given object. To get over this limitation, the re-projection technique is implemented in this paper for tracking an object in any video recorded from a mobile or stationary camera. Re-projection, online updates the histogram of object template for every frame by superimposing the feature keypoints from upcoming frames to the first frame. The object can be tracked using SURF descriptors without using any secondary information (color, texture, optical flow, gradient, etc.) regarding the object, which makes it computationally inexpensive to be used in real-time systems. Also, the SURF feature is calculated only for the target object that is to be tracked, this further reduces computational complexity. The simulation results prove the effectiveness of the work.

This review considers previous research, regarding the background and applications of 3D face-tracking systems with a focus on stereo camera-based systems. Stereo cameras are less expensive than laser ranging systems, and they are widely available on devices such as smart phones. This review aims to spur further development and applications of face tracking in this domain. Many studies on face tracking have used concepts such as the Kanade-Lucas-Tomasi method, particle filters, tracking-learning-detecting, probability hypothesis density, mean shift/cam shift, and others. As imaging constraints are relaxed, facial tracking becomes more challenging. This review presents an exposition of the most common challenges in face tracking, such as occlusion and clutter, pose variations, changes in facial resolution, illumination variations, and facial deformation. Five forms of pose estimation are discussed: appearance template methods, detector arrays, flexible models, geometric methods, and tracking methods. Applications of the listed 3D face tracking systems are also discussed, including face modelling, film editing, access control, security, and surveillance.

Long-term object tracking encounters complex scene changes, such as deformation, short-term departure from sight, occlusion, and lighting changes, resulting in complex and unstable tracking. To improve the accuracy and success rate of long-term object tracking in complex scenes, an improved continuously adaptive mean shift (CAMShift) algorithm was proposed. The joint probability density distribution of the target model was obtained by using the Bhattacharrya coefficient to calculate the contribution of the color features and texture features. Combining with the fused target model and Kalman filter, the target position was obtained by implementing CAMShift algorithm. Finally, a template pool was designed to store high-confidence tracking results. The target template was updated online by retrieving the initial frame from the template pool to recover re-detection after tracking drift or failure. The accuracy of the proposed algorithm was verified by simulation analysis. Results show that the distance precision and success rate of the proposed algorithm are 0.9 and 0.83, respectively. The proposed algorithm effectively solves long-term target tracking problems affected by complex scenes, such as occlusion, similar colors, and deformation. This study provides references for the automatic detection of traffic incidents in the intelligent traffic monitoring system.

Integrating multiple different yet complementary modalities has been proved to be an effective way for boosting tracking performance. Many previous works just consider the fusion of different features from identical image or identical features from images with different spectrum alone, which makes them be quite distinct from each other and be hard to be integrated naturally. In this study, we propose a unified tracking framework to integrate multiple different modalities via innovative use of spatiogram, where the spatiogram is formed by weighting each bin of histogram with the mean and covariance of the locations of the pixels that contribute to that bin. Specifically, each modal target and its candidate are first represented by second-order spatiogram, and their similarity is measured by the weighted Bhattacharayya coefficient. Next, an objective function is built by integrating all modal similarities, then a joint center-shift formula of the target is gained by performing Taylor expansion and gradient minimization on the objective function. Finally, the optimal target location is gained recursively by applying the mean shift procedure. Besides, a fast fuzzy logic system is designed to adaptively adjust the weight of each modality, and a model update scheme based on particle filter is developed to capture the appearance variations. Our framework allows the modality to be original gray of pixel or other extracted feature from single image or different spectral images, and provides the flexibility to arbitrarily add or remove modality. Experimental results on three challenging public datasets demonstrate clearly the robustness and effectiveness of the proposed method.

Mean-Shift Super Resolution (MSSR) is a principle based on the Mean Shift theory that improves the spatial resolution in fluorescence images beyond the diffraction limit. MSSR works on low- and high-density fluorophore images, is not limited by the architecture of the detector (EM-CCD, sCMOS, or photomultiplier-based laser scanning systems) and is applicable to single images as well as temporal series. The theoretical limit of spatial resolution, based on optimized real-world imaging conditions and analysis of temporal image series, has been measured to be 40 nm. Furthermore, MSSR has denoising capabilities that outperform other analytical super resolution image approaches. Altogether, MSSR is a powerful, flexible, and generic tool for multidimensional and live cell imaging applications.

In the current decade, many automobile crashes are occurred due to the non-professional drivers and the negligence of legal drivers during driving. To reduce the automobile crashes, illegal driver recognition and legal driver’s attention estimation requires significant research attention in the field of computer vision. In this paper, a driver recognition and assistance system is proposed to monitor the driver’s attention and drowsiness based on face and eyes tracking. For that, initially, the driver is recognized before starting the automobile through SVM, where features are extracted by using uniform LBP and Gabor filters. After recognize the driver, the driver is allowed to drive. Furthermore, from the recognized face, the eye’s pupils are detected and tracked to estimate the attention of the driver. Here, color feature is utilized to track the face and eyes through the mean shift algorithm. The system generates the alarm for the illegal driver and awareness alarm of the driver in the case of driver face angle and eye’s fatigue. The effectiveness of the proposed system is demonstrated through the real-time experiment. The proposed system is evaluated in the different lighting conditions and presented outcomes demonstrate the adequacy.

In the process of moving target tracking, when the trajectory changes or the target is occluded during the moving process, the traditional tracking algorithm is prone to loss of target tracking or poor tracking effect. This paper proposes an improved algorithm based on the interactive multi-model Kalman filter algorithm. The improved algorithm integrates the interactive multi-model Kalman filter algorithm and the Mean Shift filter algorithm to estimate the position of moving targets, solves the problem of target occlusion, and improves the accuracy of target tracking. Experimental results show that the improved algorithm can locate and track targets quickly and effectively.

Accurately detecting human falls of elderly people at an early stage is vital for providing early alert and avoid serious injury. Towards this purpose, multiple triaxial accelerometers data has been used to uncover falls based on an unsupervised monitoring procedure. Specifically, this paper introduces a one-class support vector machine (OCSVM) scheme into human fall detection. The main motivation behind the use of OCSVM is that it is a distribution-free learning model and can separate nonlinear features in an unsupervised way need for labeled data. The proposed OCSVM scheme was evaluated on fall detection databases from the University of Rzeszow's. Three other promising classification algorithms, Mean shift, Expectation-Maximization, k-means, were also assessed based on the same datasets. Their detection performances were compared with those obtained by the OCSVM algorithm. The results showed that the OCSVM scheme outperformed the other methods.

As big data technology penetrates all aspects of our lives; more and more industries are beginning to apply big data technology. Its robust data analysis, resource acquisition, and predictive capabilities have brought tremendous business activities as a brand-new data management technology. This article analyzes the big data technology application in business activities and introduces the principles of two essential algorithms in big data technology, clustering algorithm, and principal component analysis. Moreover, it introduces applying the mean shift clustering and principal component analysis in clustering algorithm to analyze the commercial data. Finally, a comprehensive analysis of the advantages and disadvantages of big data technology is made.

A robust visual algorithm is proposed in order to solve the problem of tracking drift or failure easily and lower accuracy in the load detection of overhead cranes based on visual detection with illumination change. Firstly, the load position information of the overhead cranes can be obtained by detecting the spherical marker attached on the load, which is insensitive to rotation and tilt. Then, a hierarchical strategy which tracking and detecting to the target is used for load location information acquisition. In the tracking module, the insensitivity of texture features to light changes is used, and the improved mean shift algorithm of adaptive linear fusion of color and texture features combined with linear prediction method are applied to track spherical markers continuously under illumination changes. In the detection module, an improved random incremental circle detection is used to detect spherical markers in real time. Meanwhile, the homomorphic filtering and image processing algorithms are applied to eliminate the impact of illumination changes, which can improve the robustness of the algorithm with the changing of illumination. Finally, the real-time swing angle are calculated based on the spatial geometry method. Experimental results demonstrate that the method can detect the swing angle of the load under illumination change effectively.

Nowadys, image processing plays vital role in real time application by enhancing image clarity, removes noise, compresses and decompresses it etc. Image segmentation is one of the techniques that process an image by dividing it into its component regions or objects. There are many algorithms proposed for color image segmentation. Watershed method results in over segmentation of image. Enhanced method, Marker-controlled watershed solves this problem. Mean Shift algorithm results in blurring of image. This paper proposes a hybrid approach to solve the above problems. We carried out our experiment on different colored images. Firstly, mean shift is implemented which finds density regions which are highest that represents clusters centered on modes, followed by Marker controlled watershed procedure. Results were compared against watershed and mean-shift method and conclusion was drawn that best results were obtained for hybrid approach.

Identification and tracking of dynamic objects are essential concepts in the field of robotic vision. This paper presents the design of a four-wheeled robot that can detect a target object, track the object, and maintain a constant distance from it. In this design, S5P6818 system-on-a-chip (SoC) is used as the core processing unit with embedded Linux OS, and a USB video class (UVC) camera is used for image acquisition. The robot captures real-time images and tracks the object based on a joint tracking algorithm of enhanced Continuously Adaptive Mean Shift (CAMShift) and Kalman filter (KF). A combination of Bang-bang controller and proportional-integral-derivative (PID) controller is used to control the robot's real-time motion. The experimental results show that the robot can track a target object and keep a certain distance in real-time, proving its effectiveness and robustness.