Kishor M. Bhurchandi

3-D face recognition: features, databases, algorithms and challenges

Face recognition is being widely accepted as a biometric technique because of its non-intrusive nature. Despite extensive research on 2-D face recognition, it suffers from poor recognition rate due to pose, illumination, expression, ageing, makeup variations and occlusions. In recent years, the research focus has shifted toward face recognition using 3-D facial surface and shape which represent more discriminating features by the virtue of increased dimensionality. This paper presents an extensive survey of recent 3-D face recognition techniques in terms of feature detection, classifiers as well as published algorithms that address expression and occlusion variation challenges followed by our critical comments on the published work. It also summarizes remarkable 3-D face databases and their features used for performance evaluation. Finally we suggest vital steps of a robust 3-D face recognition system based on the surveyed work and identify a few possible directions for research in this area.

ReBiS - Reconfigurable Bipedal Snake robot

Robots capable of switching between snake-like and bipedal motion have advantages of greater manoeuvrability. This paper introduces ReBiS (Reconfigurable Bipedal Snake) robot, a novel modular design mechanism which can quickly transform between various configurations without rearrangement of modules. This paper documents the design as well as the gaits implemented on ReBiS. Possible gaits are divided into three categories; snake gaits, transforming gaits and walking gaits. An example gait, belonging each of the three categories, is implemented and presented here. Experimental verification demonstrated that the reconfiguration of this robot is swift and without reshuffling of modules.

Expression invariant face recognition using contourlet transform

The performance of many state-of-the-art expression invariant face recognition systems hampers when fewer faces are available in the gallery. This paper addresses the issue of expression invariant face recognition with small gallery set. The contourlet transform is an established tool for capturing contour-like edges. The contourlet transform generates prominent features as its local and directional properties have strong resemblance with human visual cortex. We have proposed a novel approach that fuses the features from spatial domain and contourlet transform domain. Feature extraction is performed by employing the LBP and WLD descriptors. The experiments are performed on the JAFFE face database and the Yale face database. The results indicate that the proposed feature level fusion approach yields a robust feature vector and exemplary recognition rates.

Low Cost E-Book Reading Device for Blind People

Document describes e-book reading device for visually impaired people using two electronic refreshable Braille display units. A blind person has access to literature using Braille script, which consists of a alphabet being represented by combination of six pierced dots. Referred device uses a novel method of two tactile refreshable Braille cells consisting of six solenoid pins actuators, for conversion of alphanumeric character into Braille character. Secure Disk (SD) Card is used as a storage media. Text is printed with each word alternatively printed on either display unit. Characters are printed on a specific unit with a speed of 2 characters per second. The user reads words by placing both hands, one on each of the two tactile units. This device demonstrates functionality of AVR ATMEGA16 to design a robust embedded system. The advantages of the design are low complexity, portability and potential low cost of manufacture.

Performance evaluation of wavelet, ridgelet, curvelet and contourlet transforms based techniques for digital image denoising

Digital images always inherit some extent of noise in them. This noise affects the information content of the image. Removal of this noise is very important to extract useful information from an image. However noise cannot be eliminated, it can only be minimized due to overlap between the signal and noise characteristics. This paper reviews image denoising algorithms which are based on wavelet, ridgelet, curvelet and contourlet transforms and benchmarks them based on the published results. This article presents the techniques, parameters used for benchmarking, denoising performance on standard images and a comparative analysis of the same. This paper highlights various trends in denoising techniques, based on which it has been concluded that a single parameter Peak Signal to Noise Ratio (PSNR) cannot exactly represent the denoising performance until other parameters are consistent. A new robust parameter Performance measure ‘P’ is presented as a measure of denoising performance on the basis of a new concept named Noise Improvement Rectangle followed by its analysis. The results of the published algorithms are presented in tabular format in terms of PSNR and P which facilitates readers to have a bird’s eye view of the research work in the field of image denoising and restoration.

Expression invariant face recognition using semidecimated DWT, Patch-LDSMT, feature and score level fusion

This paper addresses the issue of human face recognition in presence of expression variations, which pose a great challenge to face recognition systems. Typically, the discriminant features lie in both spatial as well as transform domain. In this paper, we propose combination of Discrete Wavelet Transform (DWT) and proposed Semi-decimated Discrete Wavelet Transform (SDWT) to develop an expression invariant face recognition algorithm followed by a novel wavelet coefficients enhancement function. The wavelet coefficients are boosted using the proposed coefficients enhancement function and extracted using the Weber Local Descriptors (WLD). This enhances weak skin edges based features, resulting in increased probability of recognition. The proposed algorithm also exploits spatial domain features using our customized version of Complete Local binary patterns (CLBP) named Patch Local Difference Sign Magnitude Transform (Patch-LDSMT) applied on complete images and physiologically meaningful overlapping regions of human facial images for the first time. Feature level fusion of the wavelet based features and Patch-LDSMT yields a robust feature vector whose dimensionality is reduced using Linear Discriminant Analysis (LDA). Comprehensive experimentation is carried out on the JAFFE, CMU-AMP, ORL, Yale, Cohn-Kanade (CK) and database collected by us. Benchmarking analysis illustrates that the proposed face recognition algorithm offers much better rank one recognition performance when compared with the current state-of-the-art expression invariant face recognition approaches.

Design analysis and development of low cost underactuated Robotic hand

This paper represents the control and driving mechanism of IVLABS Robotic hand, which aims to grab different objects performing various types of grasps while retaining the resemblance of a human hand; thus it is able to function as a Prosthetic hand. Meanwhile, it is vital that such hands be highly functional, light weight, provide ease of attachment and control for people and have minimum wear and tear. But their cost generally makes them unaffordable to a larger section of people. Hence, this paper focuses on the design of hand, that is cost-effective yet imparts maximum functionality by using simple actuation and using proximity sensors on fingers; replacing generally used Electromyography (EMG) sensors for user controlled grabbing. Being these the objectives we designed the tendon-driven under-actuated fingers and 3D-printed the hand model and carried out various grabbing experiments.

A novel Quantized Gradient Direction based face image representation and recognition technique

This paper presents a novel quantized gradient based local feature descriptor, named Local Quantized Gradient Direction (LQGD) descriptor and the subsequent Partitioned Gradient Histogram, for facial image representation. The 8 bit LQGD descriptor accommodates eight levels quantized gradient magnitude and direction information from the horizontal and vertical gradients at local facial image pixels using 3×3 neighborhoods. The subsequent novel partitioned histogram based feature detection using the proposed descriptor offers separation in feature space resulting in recognition performance improvement. The technique is also robust to rotation, scale variations and noise due to typical preprocessing, background minimization and the descriptor itself. Spatial and transform domain feature level fusion is used for further performance improvement. The benchmarking of the proposed technique has been done using publicly available YEL and JAFFE databases with other contemporary techniques. The proposed technique outperforms the other published contemporary techniques.

Substitute Eyes for Blind with Navigator Using Android

Our aim is to develop an affordable technology which is cheap and can be a substitute eyes for blind people. As a first step to achieve this goal we decided to make a Navigation System for the Blind. Our device consists of the following 2 parts: 1) Embedded Device: can be used to detect local obstacles such as walls/cars/etc. using 2 ultrasonic sensors to detect the obstacles and vibrator motors to give tactile feedback to the blind. 2) Android App: will give the navigation directions. Can be installed on any android device: cellphone/tablet/etc.

A biologically inspired technique for sampling of color images

Color space dimensionality possesses main problem in fast processing of color images so appropriate sampling of color images is very important. Unlike the existing statistical sampling algorithm, in this paper, a biologically inspired non-linear color image sampling technique has been proposed using non-uniform quantization of RGB space. Response of human retinal receptors to various light intensities is non-linear in nature. Buschbaum has qualitatively presented the non-linear tan-sigmoid model of the human vision as against the logarithmic and power law models. An experiment has been carried out on certified normal color vision observers in broad day light conditions to model their color vision. Readings of this experiment were used to compute the parameters of Red, Green and Blue color vision non-linearity presented by Buchsbaum. These parametric non-linearity equations were used to sample the color images and other applications of the work have been proposed. The non-linearity equations with respective parameters represent the models of Red, Green and Blue color vision receptors. Physiological limitations and facts of human vision have been utilized to compute the parameter.