Extremal Regions of Extremum Levels (EREL) are regions detected from a set of all extremal regions of an image. Maximally Stable Extremal Regions (MSER) which is a novel affine covariant region detector, detects regions from a same set of extremal regions as well. Although MSER results in regions with almost high repeatability, it is heavily dependent on the union-find approach which is a fairly complicated algorithm, and should be completed sequentially. Furthermore, it detects regions with low repeatability under the blur transformations. The reason for the latter shortcoming is the absence of boundaries information in stability criterion. To tackle these problems we propose to employ prior information about boundaries of regions, which results in a novel region detector algorithm that not only outperforms MSER, but avoids the MSER's rather complicated steps of union-finding. To achieve that, we introduce Maxima of Gradient Magnitudes (MGMs) and use them to find handful of Extremum Levels (ELs). The chosen ELs are then scanned to detect their Extremal Regions (ER). The proposed algorithm which is called Extremal Regions of Extremum Levels (EREL) has been tested on the public benchmark dataset of Mikolajczyk [1]. Our experimental evaluations illustrate that, in many cases EREL achieves higher repeatability scores than MSER even for very low overlap errors.

A fundamental problem in retinal image registration (RIR) is the determination of correspondences between image pairs. State-of-the-art RIR methods use local features like scale invariant feature transform (SIFT) to find corresponding points. However, SIFT suffers from the quantity and quality of the detected points. On the other hand, the attention of human visual systems directs to regions instead of points for feature matching. Being aware of these issues, this paper presents a new structure-based region detector and describes a robust RIR framework. It is based on a robust watershed segmentation which obtains closed-boundary regions within a clean vascular structure map. Since vascular structure maps are relatively stable in partially overlapping and temporal image pairs, the regions are unaffected by viewpoint, content and illumination variations. The regions are approximated by convex polygons, so that robust boundary descriptors are achieved to match them. Experimental results on different datasets show that our approach is comparable or superior to SIFT-based methods in terms of efficiency, accuracy and speed.

A fundamental problem of image registration is the determination of corresponding points. The scale‐invariant feature transform (SIFT) is a well‐known algorithm in this regard. However, SIFT suffers from quantity, quality and distribution of the detected points when facing with high‐resolution and low‐contrast colour retinal fundus images. This study introduces an improved SIFT algorithm which identifies adequate, stable and distinctive keypoints with uniform distribution in the overlapped areas. The keypoint of the proposed method is a selection strategy of the difference of Gaussian (DoG) extremum points according to a stability score to guarantee the feature qualities. The stability score is based on the DoG values of extremum points and their vesselness measures in the relevant Gaussian images. Since the selected points lie on the vessels, which are relatively stable between image pairs, the points are unaffected by illumination and content variations of retinal backgrounds. The detected points are introduced to an integrated outlier rejection method. Then, the correspondences determine the geometric transformation parameters. The authors examined quantitatively and qualitatively the performance of this algorithm on four datasets including temporal and partially overlapping image pairs. The experimental results show the outperformance of the approach over similar methods in terms of efficiency, positional accuracy and speed.

A problem of computer vision applications is to detect regions of interest under different imaging conditions. The state-of-the-art maximally stable extremal regions (MSERs) detects affine covariant regions by applying all possible thresholds on the input image, and through three main steps including: (1) making a component tree of extremal regions' evolution; (2) obtaining region stability criterion; and (3) cleaning up. The MSER performs very well, but, it does not consider any information about the boundaries of the regions, which are important for detecting repeatable extremal regions. We have shown in this paper that employing prior information about boundaries of regions results in a novel region detector algorithm that not only outperforms MSER, but avoids the MSER's rather complicated steps of enumeration and the cleaning up. To employ the information about the region boundaries, we introduce maxima of gradient magnitudes (MGMs) which are shown to be points that are mostly around the boundaries of the regions. Having found the MGMs, the method obtains a global criterion for each level of the input image which is used to find extremum levels (ELs). The found ELs are then used to detect extremal regions. The proposed algorithm which is called extremal regions of extremum levels (EREL) has been tested on the public benchmark data set of Mikolajczyk. The obtained experimental results show that the inclusion of region boundaries through MGMs, results in a detector that detects regions with high repeatability scores and is more robust against noise compared with MSER.

Viseme (Visual Phoneme) clustering and analysis in every language is among the most important preliminaries for conducting various multimedia researches as talking head, lip reading, lip synchronization and computer assisted pronunciation training applications. With respect to the fact that clustering and analyzing visemes are language dependent processes, we concentrated our research on Persian language, which indeed has suffered from lack of such study. In this paper, we used a hierarchical approach for clustering visemes in Persian language based on principal component analysis of a polynomial kernel matrix considering coarticulation effect. Having obtained feature vector of each phoneme, we applied unweighted pair group method with arithmetic mean to each projected viseme on constructed manifold. Then furthest neighbor of the weight value as a result of reconstruction is set as the criterion for comparing viseme dissimilarity. In order to indicate the robustness of the proposed algorithm, a set of experiments was conducted on Persian databases in which two syllables were examined. Comparing the results of the clustering algorithm with that of the perceptual test given by an expert proves a reasonable evaluation of the proposed algorithm.

In this paper, we propose a method for kernel-based object tracking in order to deal with partial occlusion. We use particle filter to estimate target position accurately. The incremental Bhattacharyya Dissimilarity (IBD) based stage is designed to consistently distinguish the particles located in the object region from the others placed in the background. While the target is occluded by background or other objects, the kernel parameters change which adaptively improves the target model. In addition, the number of particles increases in the next frame. To attain the appropriate accuracy in tracking, we use multi-feature to describe the target. The color histogram feature is robust to scale, orientation, partial occlusion and non-rigidity of the object. However, this feature is sensitive to illumination variations. Therefore, we utilize the combination of color histogram and generalized LBP for object edge points to describe an appropriate target model. The performance of this method is evaluated for real world scenarios such as PETS benchmark.

Facial image analysis is one of the areas that have been received considerable attention in recent decades. In addition to areas such as face recognition, gender classification, emotion recognition, and age estimation, there are new applications that have not been studied yet. Family similarity recognition is a new trend that has been studied in this paper for the first time. Local Gabor Binary Pattern Histogram Sequence (LGBPHS) has led to many important advances in face recognition, including over looking generalizability and training issues. Given the current status of this study, two approaches were considered: (1) holistic approach and (2) component-based approach, which embodies the practical principles of theory. In order to model facial family manifold, LGBP is used both for the holistic view and components of the face. For recognition, histogram intersection is used to measure the similarity of different LGBPHSes and the nearest neighborhood is exploited for final clustering. In order to prove the efficiency of the proposed method three set of experiments are conducted in which both subjective and algorithmic issues are considered. It is observed in the course of experiments that the proposed method outperforms the subjective test up to 15%, and outperforms the observed state of the art face recognition methods up to 25.06%.

Characters connectivity is a problem in automated printed Farsi/Arabic script recognition. This paper introduces a novel scheme based on wavelet transform to solve segmentation of printed Farsi/Arabic words into characters. Our novel algorithm employs a new wavelet transform by which the extracted wavelet coefficients are exploited, in detecting, underlying horizontal edges and base line. Projection of horizontal edges and their location on base line provide the segmentation points. A classification method distinguishes true segmenting points. New algorithm is robust against noise, gray level, font and size of characters. Simulation results provide a comparison between new algorithm and three schemes, closed contour, structural and holistic, in terms of precision, speed and robustness against Gaussian noise. Experimental Results indicate superiority of our scheme in terms of precision and show that new algorithm improves recognition speed by a factor of at least 2.5 times.

This paper presents a novel holistic handwritten Farsi /Arabic word recognition scheme in situation where we face with word rotation and scale change. Image words features are extracted by exploiting rotation and scale invariance characteristics of M-Band packet wavelet transform performed on polar transform version of images of handwritten Farsi/Arabic words. The extracted features construct a feature vector for each word image. This vector is employed in recognition phase by finding the similar words based on the least Mahalanobis distance of feature vectors. This scheme is robust against rotation and scaling. Experimental results, obtained from testing different handwritten texts with various orientations and scales, show that proposed scheme outperforms Fourier-wavelet and Zernike moments algorithms. The robustness of new scheme has been tested with images corrupted by Gaussian noise and compared with similar schemes. Experimental results show that the accuracy of our algorithm reaches 95.8 percents.

This paper proposes a new feature vector for non-segmentation shape-based image indexing and retrieval. This feature considers correlation between neighboring edges. Hence it includes information of continuous edges and lines of images and describes major shape properties of images. This scheme is effective and robustly tolerates translation, scaling, color, illumination, and small viewing position variations. Experimental results show superiority of proposed scheme over several other indexing methods. Averages of precision and recall rates of this new indexing scheme for retrieval as compared with edge direction histogram are 1.26 and 1.04 times, respectively.

This correspondence compares the computational complexity of the pair-wise nearest neighbor (PNN) and Linde-Buzo-Gray (LBG) algorithms by deriving analytical expressions for their computational times. It is shown that for a practical codebook size and training vector sequence, the LBG algorithm is indeed more computationally efficient than the PNN algorithm.