Vijayan K. Asari

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 pipelined architecture for real-time correction of barrel distortion in wide-angle camera images

An efficient pipelined architecture for the real-time correction of barrel distortion in wide-angle camera images is presented in this paper. The distortion correction model is based on least-squares estimation to correct the nonlinear distortion in images. The model parameters include the expanded/corrected image size, the back-mapping coefficients, distortion center, and corrected center. The coordinate rotation digital computer (CORDIC) based hardware design is suitable for an input image size of 1028/spl times/1028 pixels and is pipelined to operate at a clock frequency of 40 MHz. The VLSI system will facilitate the use of a dedicated hardware that could be mounted along with the camera unit.

Adaptive and integrated neighborhood-dependent approach for nonlinear enhancement of color images

A novel image enhancement algorithm called AINDANE (adaptive and integrated neighborhood dependent approach for nonlinear enhancement) for improving the visual quality of digital images captured under extremely low or nonuniform lighting conditions is proposed. AINDANE is comprised of two separate processes, namely, adaptive luminance enhancement and adaptive contrast enhancement, to provide more flexibility and better control over image enhancement. Adaptive luminance enhancement is a global intensity transformation based on a specifically designed nonlinear transfer function, which is self-tuned by the histogram statistics of the input image. This process largely increases the luminance of darker pixels and compresses the dynamic range of the image at the same time. Adaptive contrast enhancement tunes the intensity of each pixel based on its relative magnitude with respect to the neighboring pixels. This process is also adaptively controlled by the global statistics of the image. A color restoration process, based on the relationship between the spectral bands and the luminance of the original image, is applied to convert the enhanced intensity image back to a color image.

Letters: Ratio rule and homomorphic filter for enhancement of digital colour image

Digital colour image enhancement using a homomorphic filter provides good dynamic range compression, but it fails in colour rendition. In this letter, we propose to perform natural colour rendition in a digital colour image that is enhanced by a homomorphic filter. A novel neural network learning algorithm, named Ratio rule, is used to carry out the natural colour rendition process. The Ratio rule learns to restore colour in an image by representing the colour relationships of each pixel in the original image as a line of attraction in the state space. The dynamics of the neural network is then used as an associative memory for recalling the natural colour characteristics of image pixels after homomorphic filtering. Several experiments on benchmark problems show that the new technique of image enhancement can generate natural colour in images from their original colour information.

Neural network based skin color model for face detection

This paper presents a novel neural network based technique for face detection that eliminates limitations pertaining to the skin color variations among people. We propose to model the skin color in the three dimensional RGB space which is a color cube consisting of all the possible color combinations. Skin samples in images with varying lighting conditions, from the Old Dominion University skin database, are used for obtaining a skin color distribution. The primary color components of each plane of the color cube are fed to a three-layered network, trained using the backpropagation algorithm with the skin samples, to extract the skin regions from the planes and interpolate them so as to provide an optimum decision boundary and hence the positive skin samples for the skin classifier. The use of the color cube eliminates the difficulties of finding the non-skin part of training samples since the interpolated data is consider skin and rest of the color cube is consider non-skin. Subsequent face detection is aided by the color, geometry and motion information analyses of each frame in a video sequence. The performance of the new face detection technique has been tested with real-time data of size 320/spl times/240 frames from video sequences captured by a surveillance camera. It is observed that the network can differentiate skin and non-skin effectively while minimizing false detections to a large extent when compared with the existing techniques. In addition, it is seen that the network is capable of performing face detection in complex lighting and background environments.

Modified luminance based MSR for fast and efficient image enhancement

A luminance based multi scale retinex (LB/spl I.bar/MSR) algorithm for the enhancement of darker images is proposed in this paper. The new technique consists only the addition of the convolution results of 3 different scales. In this way, the color noise in the shadow/dark areas can be suppressed and the convolutions with different scales can be calculated simultaneously to save CPU time. Color saturation adjustment for producing more natural colors is implemented. Each spectral band can be adjusted based on the enhancement of the intensity of the band and by using a color saturation parameter. The color saturation degree can be automatically adjusted according to different types of images by compensating the original color saturation in each band. Luminance control is applied to prevent the unwanted luminance drop at the uniform luminance areas by automatically detecting the luminance drop and keeping the luminance up to certain level that is evaluated from the original image. Down-sized convolution is used for fast processing and then the result is re-sized back to the original size. Performance of the new enhancement algorithm is tested in various images captured at different lighting conditions. It is observed that the new technique outperforms the conventional MSR technique in terms of the quality of the enhanced images and computational speed.

Facial Recognition Using Multisensor Images Based on Localized Kernel Eigen Spaces

A feature selection technique along with an information fusion procedure for improving the recognition accuracy of a visual and thermal image-based facial recognition system is presented in this paper. A novel modular kernel eigenspaces approach is developed and implemented on the phase congruency feature maps extracted from the visual and thermal images individually. Smaller sub-regions from a predefined neighborhood within the phase congruency images of the training samples are merged to obtain a large set of features. These features are then projected into higher dimensional spaces using kernel methods. The proposed localized nonlinear feature selection procedure helps to overcome the bottlenecks of illumination variations, partial occlusions, expression variations and variations due to temperature changes that affect the visual and thermal face recognition techniques. AR and Equinox databases are used for experimentation and evaluation of the proposed technique. The proposed feature selection procedure has greatly improved the recognition accuracy for both the visual and thermal images when compared to conventional techniques. Also, a decision level fusion methodology is presented which along with the feature selection procedure has outperformed various other face recognition techniques in terms of recognition accuracy.

A multisensor image fusion and enhancement system for assisting drivers in poor lighting conditions

A system of multisensor image fusion and enhancement for visibility improvement is proposed in this paper for helping drivers driving at night or under bad weather conditions. Video stream captured by a CCD camera is enhanced, then aligned and fused with another stream captured by a thermal camera to improve the visibility of roads in extremely low lighting conditions. A nonlinear image enhancement technique capable of dynamic range compression and contrast enhancement is developed to enhance the visible images prior to fusion. The thermal image and the enhanced visible image are then aligned based on prior information obtained on image registration process. Pixel-level multiresolution based image fusion method is applied to merge source images. After image fusion, a color restoration is performed on fused images with the chromatic information of visible images. The entire image processing and analysis system is being installed in an FPGA environment. Preliminary results obtained in various experiments conducted with the proposed system are encouraging

An integrated neighborhood dependent approach for nonlinear enhancement of color images

We propose a new image enhancement algorithm INDANE (integrated neighborhood dependent approach for nonlinear enhancement of color images), which is applied to improve the visibility of the dark regions in digital images. INDANE is a combination of two independent processes: luminance enhancement and contrast enhancement. The luminance enhancement, also regarded as a process of dynamic range compression, is essentially an intensity transformation based on a specifically designed nonlinear transfer function. The contrast enhancement transforms the pixels intensity based on the relationship between the center pixel and its surrounding pixels. For color images, they are firstly converted to intensity (grayscale) images, and then are treated by the proposed image enhancement process.

Recurrent neural network as a linear attractor for pattern association

We propose a linear attractor network based on the observation that similar patterns form a pipeline in the state space, which can be used for pattern association. To model the pipeline in the state space, we present a learning algorithm using a recurrent neural network. A least-squares estimation approach utilizing the interdependency between neurons defines the dynamics of the network. The region of convergence around the line of attraction is defined based on the statistical characteristics of the input patterns. Performance of the learning algorithm is evaluated by conducting several experiments in benchmark problems, and it is observed that the new technique is suitable for multiple-valued pattern association.