Vellore Institute of Technology (VIT) is a private university located in&nbsp;Tamil Nadu, India. Founded in 1984, as Vellore Engineering College, the institution offers 20 undergraduate, 34 postgraduate, four integrated and four research programs. It has campuses in Vellore, Amravati, Bhopal and Chennai.

VIT is one of the top ranked private universities in India according to NIRF, THE and QS Rankings.&nbsp;Govt. of India has recognized&nbsp;VIT, Vellore as an&nbsp;Institution of Eminence. This has allowed VIT to take independent quality initiatives and move up in world ranking.

&nbsp;

&nbsp;

VIT University

In this article, an attempt is made to combine both transform and spatial domain features to increase the accuracy, sensitivity, and specificity for the classification of retinal diseases. Papilloedema, macular edema, glaucoma, diabetic retinopathy (DR), and central retinal vein occlusion (CRVO) are the leading cause of blindness in humans. To identify these diseases in a mass screening process, it consumes intense labor and time. In this proposed work, an automated computer aided algorithm is developed to classify the normal from the abnormal (Papilledema, macular edema, glaucoma, DR, and CRVO) images. For any classification task, an increase in data size, number of classes, and dimension of the feature space affect the performance of any classifier. A single classifier is generally unable to handle the wide variability and scalability of the data in any problem. In the proposed algorithm, the feature fusion technique is employed to reduce the input data to the classifier. The features are extracted from wavelet packet transform (WPT) and intensity hue saturation (IHS). A support vector machine (SVM)-based learning algorithm is used to train the classifier using the extracted features. The proposed method yields better results than the existing classifier fusion method with an overall accuracy, sensitivity, specificity, positive prediction value (PPV), negative prediction value (NPV), and area under ROC curve (AUC) of 96.3%, 95.8%, 97.8%, 96.4%,98.1%, and 0.971 respectively. © 2019 Wiley Periodicals, Inc.

World University root node

Department of Instrumentation

School of Electrical Engineering

Vellore Campus

International Journal of Imaging Systems and Technology

Computer-aided diagnosis of retinal diseases using multidomain feature fusion

This paper proposes a uniform ultimate bounded (UUB) controller framework using model reference adaptive control for visual servoing of the ball on plate system. To address the major challenges in designing a control scheme for visual servoing applications including inter-axis coupling, exogenous disturbances, and plant perturbations due to modelling errors, a robust model reference adaptive PID control scheme using e1 modification method is put forward. The key advantage of this methodology is its ability to yield asymptotic stability of the closed loop system without prior information on the plant perturbations. Moreover, exploiting the Erzberger׳s perfect model following condition, the algorithm obtains the pseudo inverse of the system to make the system track different test trajectories. The stability and convergence of the proposed scheme are proved using Schwarz׳s inequality condition and Frobenius norms. To evaluate the tracking performance, two test cases namely reference following during exogenous disturbance and tracking under plant perturbations are validated. Simulation results accentuate that the proposed scheme yields satisfactory tracking response even during plant perturbation and exogenous disturbances. © 2016 The Franklin Institute

VIT Unitversity

Journal of the Franklin Institute

Uniform ultimate bounded robust model reference adaptive PID control scheme for visual servoing

This paper presents a robust method for recognizing human faces with varying illumination as well as partial occlusion. In the proposed approach, a dual-tree complex wavelet transform (DTCWT) is employed to normalize the illumination variation in the logarithm domain. In order to minimize the variations under different lighting conditions, appropriate low frequency DTCWT subbands are truncated and the rest of the directional subbands are used to reconstruct the stable invariant face. Using the fundamental concept that patterns from a single object class lie in a linear subspace, we develop class specific dictionaries using principal component analysis (PCA) based subspace learning on illumination invariant faces. By representing the pre-processed probe image against each dictionary using l1 regularization into PCA reconstruction, target face and sparse noises are effectively factorized. Then, identification decision is made in favor of a class with minimum reconstruction error. Evaluations on challenging probe images demonstrate that the proposed method performs favorably against several state of the art methods. © 2015, Springer Science+Business Media New York.

Multimedia Tools and Applications

Robust face identification using DTCWT and PCA subspace based sparse representation

Robust eye state classification in real-time is very crucial for automatic driver drowsiness detection to avoid road accidents. In this paper, we propose partial least squares (PLS) analysis based eye state classification method and its real-time implementation on resource constraint digital video processor platform, to monitor the eye state during all time driving conditions. The drowsiness is detected using percentage of eye closure (PERCLOS) metric. In this approach, face in the infrared (IR) image is detected using Haar features based cascaded classifier and within the face, eye is detected. For binary eye state classification, PLS analysis is applied to obtain the low dimensional discriminative subspace, within which simple PLS regression score based classifier is used to classify test vector into open and closed. We compared our algorithm to recent methods on challenging test sequences and the result shows superior performance. The results obtained during on-vehicle testing show that the proposed system achieves significant improvement in classification accuracy at nearly 3 frames per second. © 2015, Springer Science+Business Media New York.

Journal of Signal Processing Systems

Real-Time Vision Based Driver Drowsiness Detection Using Partial Least Squares Analysis

Over recent years, the demand for face video analytics in surveillance cameras is increasing dramatically due to the need for real-time event detection in commercial and security applications. The main objective of this paper is to implement robust face detection and identification system on a programmable resource constrained digital video processor. In order to reduce the unnecessary computations involved in Viola and Jones (V-J) face detector, we have proposed skin detection based search window reduction technique. After that, to identify the detected faces, we have proposed sparse prototypes based face identification algorithm which can address partial occlusion and expression variation issues effectively. Furthermore, within the face identification framework, a simple rejection ratio has been proposed to reject the invalid test faces. Experimental results of the developed embedded vision system demonstrate that the proposed methods can achieve satisfactory performance in uncontrolled settings. © 2015 Inderscience Enterprises Ltd.

International Journal of Signal and Imaging Systems Engineering

Robust embedded vision system for face detection and identification in smart surveillance

Video based face identification is a challenging problem as it needs to learn a robust model to account for face appearance change caused by pose, scale, expression and illumination variations. This paper proposes a novel video based face identification framework by combining fast sparse representation and incremental learning. For robust face identification, we proposed class specific subspace model based sparse representation which gives dense target coefficients and sparse error coefficients. Each subspace model is learned by using Principal Component Analysis (PCA). The test face is identified by using residual errors obtained for each model. For video based face identification, to harness the temporal information we integrated incremental face learning with our proposed face identification algorithm. By using reconstruction error and sparse error coefficients we have formulated new decision rules using rejection ratio and occlusion ratio respectively which helps in effective subspace model update in online. Numerous experiments using static and video datasets show that our method performs efficiently. © Springer International Publishing Switzerland 2014.

Advances in Intelligent Systems and Computing

Real-time video based face identification using fast sparse representation and incremental learning

In aerially captured video, low resolution, very small object size and poor image quality may pose tracking as very difficult problem. Based on fundamental principle that single object class lie on a linear subspace, we propose a least squares estimation-based linear model which represents probe candidate as class specific galleries. In this method, tracking is considered as a binary classification problem, which consists of target and background class. Inclusion of background templates improves the discriminative strength of the object model. By projecting target patches on estimated target coefficient vector, all candidates are reconstructed. The squared sum of reconstruction error is used as an observation likelihood function to determine the maximum a posteriori (MAP) estimation for current frame. To contain the target appearance change during tracking online multiple dictionary learning strategy is proposed. Experiment results of proposed method shows better tracking accuracy and computation when compared to other representative tracking methods. Copyright © 2014 Inderscience Enterprises Ltd.

International Journal of Computational Vision and Robotics

Least squares estimation-based adaptive observation model for aerial visual tracking applications

Facial landmark detection is very crucial for different kinds of video analytics in robotic applications. However, resource limited characteristics of an embedded system pose many challenges to implement a standalone robotic vision systems like human machine interaction, driver drowsiness monitoring and surveillance applications. The main objective of this paper is to propose effective implementation strategies for the extended Active Shape Model (ASM) based facial landmark detection algorithm using programmable resource constraint digital video processor and evaluate it in real-time. In extended ASM approach, robust detection of face, eyes and mouth is very important to initialize the shape model which defined by facial landmarks. Using this initialized shape model, the landmarks of the given test face are obtained through minimizing the distance with learned shape model iteratively. For different face parts detection, we have used robust and fast Viola-Jones detector (V-J detector) and optimized it at various levels to meet the computation and memory constraints. The implemented embedded vision system has been evaluated using visible light camera in real-world scenarios and its robustness and time consumption has been reported in detail. Through different experiments, we show that the proposed implementation framework achieves more than 98% face parts detection rate and mean error of less than 2 pixels in facial landmarks accuracy. © 2014 IEEE.

International Conference on Embedded Systems, ICES 2014

Embedded implementation of facial landmarks detection using extended active shape model approach

Effective tracking is still a big challenge due to lack of robust descriptors which captures discriminative features in non-controlled environment. We propose a novel descriptor based on Gabor wavelet and Partial Least Squares (PLS) discriminant analysis. Multi scale and multi orientation Gabor wavelets can extract selective local frequencies effectively in spatial and frequency domain. Due to the large dimension of feature vectors, dimensionality reduction is done using class aware PLS analysis. Unlike unsupervised Principal Component Analysis (PCA), PLS based subspace model learns target effectively by explicitly knowing the class labels of target and background region feature vectors. Tracking is done using particle filter and similarity between target and candidates is measured using low dimensional subspace model. To combat the target changes during tracking, novel static and dynamic target as well as background update strategy is used. Experimental results of various dataset demonstrate that the proposed tracker improves robustness and accuracy against representative trackers. © 2013 The Authors. Published by Elsevier Ltd.