Rajkiran Gottumukkal

An improved face recognition technique based on modular PCA approach

A face recognition algorithm based on modular PCA approach is presented in this paper. The proposed algorithm when compared with conventional PCA algorithm has an improved recognition rate for face images with large variations in lighting direction and facial expression. In the proposed technique, the face images are divided into smaller sub-images and the PCA approach is applied to each of these sub-images. Since some of the local facial features of an individual do not vary even when the pose, lighting direction and facial expression vary, we expect the proposed method to be able to cope with these variations. The accuracy of the conventional PCA method and modular PCA method are evaluated under the conditions of varying expression, illumination and pose using standard face databases.

A flexible and efficient hardware architecture for real-time face recognition based on eigenface

We describe a flexible and efficient multilane architecture for real-time face recognition system based on modular principal component analysis (PCA) method in a field programmable gate array (FPGA) environment. We have shown in Gottumukkal R., and Asan K.V., (2004) that modular PCA improves the accuracy of face recognition when the face images have varying expression and illumination. The flexible and parallel architecture design consists of multiple processing elements to operate on predefined regions of a face image. Each processing element is also parallelized with multiple pipelined paths/lanes to simultaneously compute weight vectors of the non-overlapping region, hence called multilane architecture. The architecture is able to recognize a face image from a database of 1000 face images in 11ms.

Real time face detection from color video stream based on PCA method

We present a face detection system capable of detection of faces in real time from a streaming color video. Currently this system is able to detect faces as long as both the eyes are visible in the image plane. Extracting skin color regions from a color image is the first step in this system. Skin color detection is used to segment regions of the image that correspond to face regions based on pixel color. Under normal illumination conditions, skin color takes small regions of the color space. By using this information, we can classify each pixel of the image as skin region or non-skin region. By scanning the skin regions, regions that do not have shape of a face are removed. Principle Component Analysis (PCA) is used to classify if a particular skin region is a face or a non-face. The PCA algorithm is trained for frontal view faces only. The system is tested with images captured by a surveillance camera in real time.

Multi-lane architecture for eigenface based real-time face recognition

Abstract The concept of simultaneously processing different non-overlapping spatial regions of an image and combining the results to obtain a final image is used in this paper. We apply this concept to the domain of face recognition using Principal Component Analysis (PCA). We have shown in [1] that modular PCA improves the accuracy of face recognition when the face images have varying expression and illumination. In this work we design and implement the modular PCA algorithm for face recognition in a Field Programmable Gate Array (FPGA) environment. Since modular PCA processes non-overlapping regions of a face image to produce weight vectors, we design a parallel architecture where each parallel path has a processing element to process a predefined region of a face image. Each processing element computes a weight vector from a face image region and pre-computed eigenvectors; hence the processing element is also parallelized where each path works on one eigenvector and the face image region to compute one element in the weight vector. Each of these paths is pipelined to process the pixels from the face image region and corresponding elements from the eigenvectors in a faster manner. We name this design having pipelined parallel paths as multi-lane architecture. The architecture is able to recognize a face image from a database of 1000 face images in 11 ms.

Learning skin distribution using a sparse map

We present a new skin modeling technique based on SNoW (sparse network of Winnows) for accurate and robust skin region detection. A skin distribution map (SDM) representing the sparse network is trained with skin pixels to learn their distribution in a color space. We then train the SDM with non-skin pixels to unlearn the distribution of the non-skin pixels, which overlap with the skin pixels in the color space. This skin model can be used for skin detection on any color space. We have found the accuracy of skin detection using SDM to be slightly better than that using the skin probability map (SPM) method. The main advantage of using the SDM method over the SPM method is that the complexity, memory requirements and time for skin detection are reduced significantly.

Skin color constancy for illumination invariant skin segmentation

Accuracy of skin segmentation algorithms is highly sensitive to changes in lighting conditions. When the lighting condition in a scene is different from that in the training examples, miss-classification rate of the skin segmentation algorithms is high. Using color constancy approach we aim to compensate for skin color variations to achieve accurate skin color segmentation. Skin color constancy is realized in an unsupervised manner by using the color changes observed on face regions under different illuminations to drive the model. By training on a few faces of different ethnicities, our model is able to generalize the color mapping for any unseen ethnicity. The color changes observed are used to learn the color mapping from one lighting condition to the other. We show the proof of concept of unsupervised skin color constancy on faces from the PIE database. Skin segmentation with and without color compensation was performed on the PIE database. Results are presented which show improved skin segmentation accuracy after performing color compensation.

An automated feature-localisation algorithm for a feature-specific modular approach for face recognition

Novel techniques for accurate location of the eyes and nose of a person in a complex-lighting environment are presented in this paper. An adaptive progressive thresholding technique is applied to spot the darkest regions representing the eyes in a face. The nose region is located by performing cumulative histogram-based thresholding of the gradient image formed below the eye region. A feature-specific modular Principal Component Analysis (PCA) approach on face images is performed with the identified features for face recognition. Principal components are extracted from non-overlapping modules of the image and are concatenated to make a single signature vector to represent the face in a particular viewing angle. Additional principal components are extracted from the key facial features and are added as an extension to the signature vector. The feature-specific modular PCA approach is capable of recognising faces in varying illumination conditions and facial expressions, as the modular components represent the local information of the facial regions.

Illumination invariant faces

We create a model of joint color changes in face images due to lighting variations. This is done by observing how colors of an individuals face with fixed pose and expression are mapped to new colors under different lighting conditions. One of the challenges we are dealing with in this work is that the scenes are not constant for different lighting. Hence we cannot observe the joint color changes of the scenes. However all the scenes have a human subject with approximately frontal pose, so we use the color changes observed on a human subjects face to learn the color mapping. The joint color mappings are represented in a low dimensional subspace obtained using singular value decomposition (SVD). Using these maps the detected face from a new image can be transformed to appear as if taken under canonical lighting condition.