Ravi Kant Kumar

Selection of User-Dependent Cohorts Using Bezier Curve for Person Identification

The traditional biometric systems can be strengthened further with exploiting the concept of cohort selection to achieve the high demands of the organizations for a robust automated person identification system. To accomplish this task the researchers are being motivated towards developing robust biometric systems using cohort selection. This paper proposes a novel user-dependent cohort selection method using Bezier curve. It makes use of invariant SIFT descriptor to generate matching pair points between a pair of face images. Further for each subject, considering all the imposter scores as control points, a Bezier curve of degree n is plotted by applying De Casteljau algorithm. As long as the imposter scores represent the control points in the curve, a cohort subset is formed by considering the points determined to be far from the Bezier curve. In order to obtain the normalized cohort scores, T-norm cohort normalization technique is applied. The normalized scores are then used in recognition. The experiment is conducted on FEI face database. This novel cohort selection method achieves superior performance that validates its efficiency.

BCP-BCS: Best-Fit Cascaded Matching Paradigm with Cohort Selection Using Bezier Curve for Individual Recognition

The concept of cohort selection has been emerged as a very interesting and potential topic for ongoing research in biometrics. It has the capability to provide the traditional biometric systems to having a higher performance rate with lesser complexity and cost. This paper describes a novel matching technique incorporated with Bezier curve cohort selection. The Best-Fit matching with dynamic threshold has been proposed here to reduce the number of false match. This algorithm is applied for matching of Speeded Up Robust Feature (SURF) points detected on face images to find out the matching score between two faces. After that, Bezier curve is applied as a cohort selection technique. All the cohort scores are plotted in a 2D plane as if these are the control points of a Bezier curve and then a Bezier curve of degree n is plotted on the same plane using De Casteljau algorithm where number of control point is \(n+1\). A template contains more discriminative features more it is having distance from the curve. All the templates having score point far from the curve are included into the account of cohort subset. For each enrolled user a specific cohort subset is determined. As long as the subset is formed, T-norm cohort score normalization technique is applied to obtain the normalized scores which are further used for person identification and verification. Experiments are conducted on FEI face database and results are showing dominance over the non-cohort system.

Novel methodology for guiding attention of faces through relative visual saliency (RVS)

Identification of a human face in a crowded flux plays an important role in the context of surveillance. Considerable amount of research has been carried out on face identification in different applications. Accordingly, different researchers propose new algorithms. This paper attempts to showcase a novel methodology through which any face may be identified in a large crowd of human face. This proposed technique is based on relative visual saliency which is evaluated on the intensity values and respective spatial distance of the faces. In addition to visual saliency, top-down and bottom-up approaches to visual attention are also presented and explained in the context of face identification. Both of these two approaches are considered to be made a significant contribution while visual saliency is measured for attention-based face identification. Experiment has been carried out on test image dataset. The results are satisfactory and accuracy has also been measured. The evaluation made with the proposed approach exhibits quite encouraging results and accuracy leads to a future model of human face tracking and recognition system.

Constrained maximization of saliency of intended object for guiding attention

Saliency of an object in an image determines the attentiveness of the object with respect to human visual system. Saliency, in the absence of any external stimuli, is determined by contrast of features (like intensity, color, etc.) of the object with its surroundings. This paper proposes a novel technique to enhance the saliency of an object, which is not intrinsically salient. In order to enhance the attentiveness, the feature values of a used-defined target object (whose saliency has to be enhanced) should have more differences with the feature values of its surrounding objects. But too much modification of these values of the target object will destroy the naturalness of the image. So this problem of enhancing saliency of a target object is treated as a maximization problem under some constraints.

A Bezier Curve Cohort Selection Strategy for Face Pair Matching

The matching of two face images without any prior information is very much challenging task unlike a verification or identification system where already some knowledge about the images of each subjects are stored in the systems database. This paper proposes a methodology to enrich the performance of a face pair matching system by utilizing the complementary information collected from a set of cohort face images with the help of Bezier Curve cohort selection algorithm. A pair of face images is given as input to the system. Each image is compared with a predefined cohort pool to form two separate set of cohort scores. Further these set of cohort scores are passed through Bezier curve cohort selection method which provide two suitable cohort subsets. Afterwards a cross normalization is accomplished in conjunction with T-norm score normalization method then the absolute normalized difference between the paired face images is determined. On the basis of this normalized difference, it is finally decided whether the input face pair is from same person or not. The system is investigated with FEI face database and the results are quite impressive.

Attending Prominent Face in the Set of Multiple Faces Through Relative Visual Saliency

Visual saliency determines the extent of attentiveness of a region in a scene. In the context of attending faces in the crowd, face components and its dominance features decide the focus on attention. Attention boosts up the recognition and identification process in a crowd and hence plays an excelling role in the area of visual surveillance and robotic vision. Using different computer vision-based techniques, enormous researches have been carried out on attention, recognition, and identification of the human face in context of different applications. This paper proposes a novel technique to analyze and explore the prominent face in the set of multiple faces (crowd). The proposed method stretched out the solution, using the concept of relative visual saliency, which has been evaluated on the various parameters of face as a whole and its componentwise too. These parameters are face area, spatial location, intensity, hue, RGB values, etc. The proposed work furnishes satisfactory results. The assessment made with this approach shows quite encouraging results which may lead to a future model for robotic vision and intelligent decision-making system.

Combined Effect of Cohort Selection and Decision Level Fusion in a Face Biometric System

There are different parameters which degrade the performance of a face biometric system due to their variations. The baseline biometric systems can get relief to some extent from this kind of negative effect by utilizing the information of the cohort images and fusion methods. But to achieve the set of suitable cohorts for each and every enrolled person is a task of great challenge. Determining the cohort subset using k-means clustering cohort selection based on the matching proximity is presented in this paper. SIFT and SURF are used as facial features to represent each face image and to calculate the similarity score between two face images. The clusters having highest and lowest centroid value are fused using union rule to form the target, user dependent cohort subset. The query-claimed matching scores are normalized with the help of T-norm cohort normalization technique. The scores after normalization are used in recognition separately for SIFT as well as SURF. Finally, the responses from the classifier for these two different features are fused at decision level to cover up the shortcomings of the cohort selection method if any. The experimental execution is done on FEI face database. This integrated face biometric system gains a significant hike in performance that evidences its effectiveness over baseline.

A Master Map: An Alternative Approach to Explore Human’s Eye Fixation for Generating Ground Truth Based on Various State-of-the-Art Techniques

Saliency map is an efficient way to represent the salient objects in an image. In the area of object-based saliency, several researches have been accomplished. In these works, normal input images are taken in which salient region or arousal object are easily perceived by us. These experiments are validated based on ground truth obtained either from volunteers either through human eye fixation machine or based on viewer’s voting. But for complex images, salient locations are very confusing; therefore preparing ground truth is very difficult. In such images, results are varying with different state-of-the-art saliency model. To address this problem, this paper implements combine strategies to achieve composite saliency map which can incorporate the properties of every individual component maps. Fusion of saliency maps in this way can be utilized to generate good ground truth information which can be an alternate way of preparing ground truth unlike based on volunteers voting or through eye fixation machine. As this approach incorporates the concepts of reusability, therefore it may reduce the time and cost in the preparation of ground truth.

Emotion Recognition Through Facial Gestures - A Deep Learning Approach

As defined by some theorists, human emotions are discrete and consistent responses to internal or external events which have significance for an organism. They constitute a major part of our non-verbal communication. Among the human emotions, happy, sad, fear, anger, surprise, disgust and neutral are the seven basic emotions. Facial expressions are the best way to exhibit emotions. In this era of booming human-computer interaction, enabling the machines to recognize these emotions is a paramount task. There is an amalgamation of emotions in every facial expression. In this paper, we identified the different emotions and their intensity level in a human face by implementing deep learning approach through our proposed Convolution Neural Network (CNN). The architecture and the algorithm here yield appreciable results that can be used as a motivation for further research in computer based emotion recognition system.

Heterogeneous Face Detection

Face detection is the process of determining the location of human faces in an image. Like human visual system, a face detection system should also be capable of achieving the detection task irrespective of illumination, absence of texture, orientation and camera distance. Detecting faces in heterogeneous, infrared and thermal images is a challenging job due to variation in texture, orientation, lighting condition, intensity etc. Many researchers have worked and proposed various methods for visible faces in the domain of face detection. However, face detection with the existing algorithm in heterogeneous face images is found quite difficult one. This paper attempts to improve the accuracy of an existing face detection algorithm which was basically designed for detecting visible faces, can now be used to detect heterogeneous faces by applying various image enhancement techniques before face detection. The proposed improved algorithm is suitable for heterogeneous face images which include thermal images, infrared images in the crowd. Attempts for the same have been tested on test image dataset. The experimental results are found to be encouraging.