Seba Susan

A Simple Single Seeded Region Growing Algorithm for Color Image Segmentation using Adaptive Thresholding

In this paper, we present a region growing technique for color image segmentation. Conventional image segmentation techniques using region growing requires initial seeds selection, which increases computational cost & execution time. To overcome this problem, a single seeded region growing technique for image segmentation is proposed, which starts from the center pixel of the image as the initial seed. It grows region according to the grow formula and selects the next seed from connected pixel of the region. We use intensity based similarity index for the grow formula and Otsus Adaptive thresholding is used to calculate the stopping criteria for the grow formula. We apply the proposed method to the Berkley segmentation database images and discuss results based on Lius F-factor that shows efficient segmentation.

A non-extensive entropy feature and its application to texture classification

Abstract This paper proposes a new probabilistic non-extensive entropy feature for texture characterization, based on a Gaussian information measure. The highlights of the new entropy are that it is bounded by finite limits and that it is non-additive in nature. The non-additive property of the proposed entropy makes it useful for the representation of information content in the non-extensive systems containing some degree of regularity or correlation. The effectiveness of the proposed entropy in representing the correlated random variables is demonstrated by applying it for the texture classification problem since textures found in nature are random and at the same time contain some degree of correlation or regularity at some scale. The gray level co-occurrence probabilities (GLCP) are used for computing the entropy function. The experimental results indicate high degree of the classification accuracy. The performance of the new entropy function is found superior to other forms of entropy such as Shannon, Renyi, Tsallis and Pal and Pal entropies on comparison. Using the feature based polar interaction maps (FBIM) the proposed entropy is shown to be the best measure among the entropies compared for representing the correlated textures.

Color segmentation by fuzzy co-clustering of chrominance color features

Abstract This paper presents a novel color segmentation technique using fuzzy co-clustering approach in which both the objects and the features are assigned membership functions. An objective function which includes a multi-dimensional distance function as the dissimilarity measure and entropy as the regularization term is formulated in the proposed fuzzy co-clustering for images (FCCI) algorithm. The chrominance color cues a ⁎ and b ⁎ of CIELAB color space are used as the feature variables for co-clustering. The experiments are conducted on 100 natural images obtained from the Berkeley segmentation database. It is observed from the experimental results that the proposed FCCI yields well formed, valid and high quality clusters, as verified from Liu’s F -measure and Normalized Probabilistic RAND index. The proposed color segmentation method is also compared with other segmentation methods namely Mean-Shift, NCUT, GMM, FCM and is found to outperform all the methods. The bacterial foraging global optimization algorithm gives image specific values to the parameters involved in the algorithm.

Unsupervised detection of nonlinearity in motion using weighted average of non-extensive entropies

Motion estimation and motion analysis have an important role to play for detecting abnormal motion in surveillance videos. In this paper, we propose to use the non-extensive entropy to detect any unnaturalness in the motion over correlated video frames since it has already been proved to represent the correlated textures successfully. To achieve this end, we utilize the temporal correlation property of motion vectors over three consecutive frames to detect any motion disturbance using a weighted average of the non-extensive entropies. It is proved by the experimental results on the state-of-the-art database that the non-extensive entropy is most apt for detecting any disturbance in the continuance of motion vectors in between frames. The advantage of our approach is that no training period or normalcy reference is required since a relative disturbance in the magnitudes of motion vectors over a three-frame window gives an alarm.

An adaptive single seed based region growing algorithm for color image segmentation

In this paper an adaptive single seed based region growing algorithm (ASSRG) is proposed for color image segmentation. The proposed method starts with the center pixel of the image as the initial seed. The region growing formula uses three homogeneity criteria local, global and relative, in two steps to label the pixel to a region. It first checks for the color similarity of the pixel with respect to the connected labelled pixel and secondly with the mean value of a growing region. If the similarity criterion is fulfilled then this pixel is included in the growing region. Otherwise the similarity of the pixel with respect to its 8-neighbors is compared with respect to the mean value of a growing region. If the pixel is closer to the growing region as compared to its neighbors then it is included in the growing region, otherwise it is labelled as boundary pixel. After one region is completely grown, the next seed pixel is selected from the boundary pixel stack. Region merging is performed to reduce over segmentation in the results. We have applied our algorithm to Berkley images with successful results and the evaluation of the segmented images has been done using Lius F-factor, total number of regions segmented and time taken by the algorithm. A fuzzy rule based modification of the algorithm is also proposed to further improve results. The proposed algorithm is also compared with SSRG algorithm using Otsus threshold, SRGRM algorithm and MRG region growing techniques and is shown to outperform all methods.

Automatic texture defect detection using Gaussian mixture entropy modeling

In this paper we propose a new unsupervised, automated texture defect detection that does not require any user-inputs and yields high accuracies at the same time. To achieve this end we use the non-extensive entropy with Gaussian gain as the regularity index, computed locally from texture patches through a sliding window approach. The optimum window size is determined by modeling the entropy values by a two-mode Gaussian mixture model and checking for the minimum entropy of the mode-probabilities. The outlier entropy values corresponding to defective areas are defined as those that exceed thrice the standard deviation, as is the norm in statistics. The result is automatic defect detection with no manual intervention. Empirical results on defective texture images from the Brodatz database provide accurate localization of the defect as compared to Chetverikov and Hanburys maximal regularity method, which requires manual setting of threshold parameters for each type of texture despite of being a benchmark for texture defect detection.

Fuzzy Co-Clustering of medical images using bacterial foraging

A novel modification of the Fuzzy Clustering for Categorical Multivariate date (FCCM) algorithm termed as dasiaFuzzy Co-Clustering Algorithm for Images (FCCI)psila is proposed for clustering of medical images. The main aim of this work is to segment regions of interest in histo-pathological images which consist of groups of similar cells indicating some form of abnormality in the animal tissue. The proposed method relies on improved colour clustering results when FCCI is applied on images as compared to the conventional clustering techniques. The method also categorizes different types of lesions based on the co-clustering results. The objective function is optimized using the bacterial foraging algorithm which gives image specific values to the parameters involved in the algorithm. The colour segmentation results are found to be more accurate, producing well formed and valid clusters having ldquocrisppsila values of membership function with lesser number of iterations. The algorithm results in distinct co-clusters ranked in the order of their memberships.

Dynamic Growth of Hidden-Layer Neurons Using the Non-extensive Entropy

In this paper we present a dynamic neural network that dynamically grows the number of the hidden-layer neurons based on an increase in the entropy of the weights during training. The weights are normalized to probability values prior to the computation of the entropy. The entropy being referred is the non-extensive entropy proposed recently by Susan and Hanmandlu for the representation of structured data. Incrementally growing the hidden layer as per requirement leads to better tuning of network weights and high classification performance as proved by the empirical results.

A novel Fuzzy Entropy based on the Non-Extensive entropy and its application for feature selection

A novel Fuzzy Entropy is defined in this work that is based on the recently proposed probabilistic Non-Extensive entropy for correlated texture patterns. The main reason behind the success of the Non-Extensive entropy was its nonlinear one-sided Gaussian Information Gain function which is non-additive and hence is suitable for representing correlated data structures. The properties of the new fuzzy entropy are found to satisfy the basic properties set in literature for fuzzy entropies. In addition the feature selection using the proposed fuzzy entropy is also discussed along with its merits. It is specially noted that the fuzzy version of the non-extensive entropy retains its non-additivity property for crisp values of the membership function.

Fuzzy match index for scale-invariant feature transform (SIFT) features with application to face recognition with weak supervision

A fuzzy match index for scale-invariant feature transform (SIFT) features is proposed in this study that cumulatively involves all the test SIFT keypoints in the decision-making process. The new fuzzy SIFT classifier is adapted successfully for robust face recognition from complex backgrounds without using any face cropping tools and using only a single training template. The further incorporation of entropy weights ensures that the facial features have a greater role in the soft decision-making as compared with the background features. The highlights of the authors’ work are: (i) The development of a novel highly efficient fuzzy SIFT descriptor matching tool; (ii) incorporation of feature entropy weights to highlight the contribution of facial features; (iii) application to robust face recognition from uncropped images having diverse backgrounds with a single template for each subject. The authors thus allow for weak supervision of the face recognition experiment and obtain high accuracy for 20 subjects of the CALTECH-256 face database, 133 subjects of the labelled faces for the wild dataset and 994 subjects of the FERET database, with state-of-the-art comparisons indicating the supremacy of the authors’ approach.