Search

Face recognition has assigned a special place to itself because of its low intrusiveness, low cost and effort and acceptable accuracy. There are several methods for recognition and appearance based methods is one of the most popular one. Unfortunately most of the papers that have been published these years have just shown the results on the databases that are all without any noise and all of focus. But it is clear that for a real system all these problems can happen, so finding methods that are robust to such problems is important. In this paper we show that linear appearance based methods are robust to an acceptable degree to problems such as, when the camera is moving or it is defocus and when the image is influenced with Gaussian noise. For linear appearance based methods we chose Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA) and Multiple Exemplar Discriminant Analysis (MEDA) that has shown better performance than other appearance based methods. [ABSTRACT FROM AUTHOR]

In this paper, a recently developed fiber tracking algorithm to be used with diffusion tensor (DT) fields acquired via magnetic resonance imaging (MRI) is improved and applied to real brain DT-MR images. The method performs satisfactorily in regions where branching and crossing fibers exist and offers the capability of reporting a probability value for the computed tracts. This certainty figure takes into account both the anisotropy and the information provided by all the eigenvectors and eigenvalues of the diffusion matrix at each voxel. In previous papers of the authors, a simpler algorithm was applied only to elementary synthetic DT-MR images. As now presented, this algorithm is now adequately used with more intricate synthetic images and is applied to real white matter DT-MR images with successful results. A novel neural network is used to estimate the main parameters of the algorithm. Numerical experiments show a performance gain over previous approaches, specially with respect to convergence and computational load. The tracking of white matter fibers in the human brain will improve the diagnosis and treatment of many neuronal diseases. [ABSTRACT FROM AUTHOR]

Moving object detection is very important for video surveillance. In this paper, we present a new real time motion detection algorithm that is based on the integration of accumulative optical flow and double background filtering method (long-term background and short-term background) to achieve better performance. The accumulative optical flow method is used to obtain and keep a stable background image to cope with variations on environmental changing conditions and the double background filtering method is used to eliminate the background information and separate the moving object from it. The biggest advantage of this algorithm is that it does not need to learn the background model from hundreds of images and can handle quick image variations without prior knowledge about the object size and shape. The algorithm has high capability of anti-interference and preserves high accurate rate detection at the same time. The effectiveness of the proposed algorithm for motion detection is demonstrated in a simulation environment and the evaluation results are reported in this paper. [ABSTRACT FROM AUTHOR]

Our work explores the use of several text categorization techniques for classification of manufacturing quality defect and service shop data sets into fixed categories. Although our work was in the area of manufacturing quality the technique is applicable to free form, short text summaries of data that may be stored in a database, file, or document. We refer to these types of text as "technical text passages". These summaries may not follow standard grammar conventions; they commonly contain abbreviations, technical phrases, misspelled words and industry specific acronyms. Typical types of text to be classified include aircraft engine repair shop findings, industrial manufacturing quality problems and corrective actions, and standardization of attributes in a bill-of-materials. In this paper, we will present our results in using machine learning and rule based algorithms to categorize text. Our results show that the rules based approach is as good as several machine learning approaches. For example, using Support Vector Machine algorithms we were able to achieve 82% accuracy on validation set, using 1,645 training samples and 823 validation samples. Each category had 50 or more samples. Using rule-based approach we were able to achieve 80% accuracy. [ABSTRACT FROM AUTHOR]

In this paper, a novel approach to Arabic letter recognition is proposed. The system is based on the classified vector quantization (CVQ) technique employing the minimum distance classifier. To prove the robustness of the CVQ system, its performance is compared to that of a standard artificial neural network (ANN)-based solution. In the CVQ system, each input letter is mapped to its class using the minimum Euclidean distance. Simulation results are provided and show that the CVQ system always produces a lower Mean Squared Error (MSE) and higher success rates than the current ANN solutions. [ABSTRACT FROM AUTHOR]

The build up of water ice on aircraft flight surfaces poses a significant safety risk. As a result, much effort has gone into studying this problem in order to understand how individual droplets contribute to the accretion process. One approach has been to capture the moment of impact of a supercooled droplet onto a surface placed in an icing tunnel. However, this produces a large number of images that must be analysed manually. This paper describes the development of an automated analysis system, employing image processing techniques, that is capable of classifying the impact images without operator input. Using a carefully chosen feature vector and K-means clustering algorithm, the classification results from the automated system are comparable with that achieved using the manual approach. [ABSTRACT FROM AUTHOR]

Applications of the Particle Swarm Optimization (PSO) to solve image processing problem with a reference to a new automatic enhancement technique based on real-coded particle swarms is proposed in this paper. The enhancement process is a non-linear optimization problem with several constraints. The objective of the proposed PSO is to maximize an objective fitness criterion in order to enhance the contrast and detail in an image by adapting the parameters of a novel extension to a local enhancement technique. The feasibility of the proposed method is demonstrated and compared with Genetic Algorithms (GAs) based image enhancement technique. The obtained results indicate that the proposed PSO yields better results in terms of both the maximization of the number of pixels in the edges and the adopted objective evaluation. Computational time is also relatively small in the PSO case compared to the GA case. [ABSTRACT FROM AUTHOR]

Pose problem is a big challenge for applying face recognition technology under real world conditions. In this paper, appearance based approach was proposed to recognize face across front and non-frontal view images by reconstructing frontal view features. Statistical learning method based on sample images is applied to find transformation matrix which encapsulated general knowledge of pose transition in feature subspace, therefore, different view feature vectors constituted linear equations and transformation matrix can be solved from the equations by least square (LS) approach. Experimental results on popular FERET and CMU databases showed that the proposed method could cope with the head rotation roughly within half profile view. Compared with model based approaches, this method is not dependent on heavy computation and has merit of easy implementing in live conditions. [ABSTRACT FROM AUTHOR]

Face recognition is a biometric identification method which among the other methods such as, finger print identification, speech recognition, signature and hand written recognition has assigned a special place to itself. In principle, the biometric identification methods include a wide range of sciences such as machine vision, image processing, pattern recognition neural networks and has various applications in film processing, control access networks and etc. There are several methods for recognition and appearance based methods is one of them. One of the most important algorithms in appearance based methods is linear discriminant analysis (LDA) method. One of the drawbacks for LDA in face recognition is the small sample size (SSS) problem so it is suggested to first reduce the dimension of the space using methods among which, principal component analysis (PCA) is the most popular one. In this paper we show that there exist stronger methods such as discrete cosine transform (DCT). [ABSTRACT FROM AUTHOR]

This paper presents a multi step-size monic normalization equation-error linear filter. We also extend the idea to nonlinear adaptive filter and derive a multi step-size monic normalization equation-error bilinear filter (MSS MNEEBF). The algorithms enjoy fast convergence behavior and can remove biased estimates associated with conventional equation-error adaptive filters. Simulation results validate the usefulness of our algorithms. [ABSTRACT FROM AUTHOR]

This paper presents a new method for the signature analysis of mechanical watch movements. Contrary to the existing method, it analyzes the time-frequency features of mechanical watch movement through a combination of reassigned scalogram and Finite Element Analysis (FEA). By mapping the signal into a 2D domain of time and frequency, the reassigned scalogram reveals the frequency components of the shocks at different time. On the other hand, FEA gives the modal frequencies of the movements. By comparing the frequency components at different shocks to the modal frequencies of the movement, possible malfunctions of the movement can then be detected. A number of demonstration examples are included. [ABSTRACT FROM AUTHOR]

In this paper, we attempt to develop image watermarking algorithms which are portable to a variety of applications such as copyright protection, fingerprinting and identification. Therefore, we require that the watermark be binary and be not only detectable but also extractable. The embedding technique is based on the parent-child structure of the multiwavelet transform called "triple tree" and this technique does not require the original image in the watermark extraction. The experimental results show that the watermark. [ABSTRACT FROM AUTHOR]