Praveen Sankaran

Fusion based Multi Scale RETINEX with Color Restoration for image enhancement

Image fusion is a technique of combining two or more images so that the combined image is better enhanced than all these images. We propose that a fusion based approach on Multi Scale Retinex with Color Restoration(MSRCR) would give better image enhancement. Lower dynamic range of a camera as compared to human visual system causes images taken to be extremely dependent on illuminant conditions. MSRCR algorithm enhances images taken under a wide range of nonlinear illumination conditions to the level that a user would have perceived it in real time.

An automated multi Scale Retinex with Color Restoration for image enhancement

The dynamic range of a camera is much lesser than that of human visual system. This causes images taken by the camera to look different from how the scene would have looked to a naked eye. Multi Scale Retinex with Color Restoration (MSRCR) algorithm enhances images taken under a wide range of nonlinear illumination conditions to the level that a user would have perceived it in real time. But there are parameters used in this enhancement method that are image dependent and have to be varied based on the images under consideration. In this paper we propose a completely automated approach for MSRCR by obtaining parameter values from the image being enhanced.

Tracking and Recognizing Multiple Faces Using Kalman Filter and ModularPCA

Abstract Real-time tracking and recognizing multiple faces in complex environments has the ability to provide e_cient security automation to large areas. Previous research has shown that Kalman filter techniques paired with the traditional face detection methods can be used to track one or more faces in a viewing region, but prove unreliable under variant conditions due to the inability to reliably distinguish between multiple trackers. A real-time face tracking and recognition system is presented that is capable of processing multiple faces simultaneously. The proposed system utilizes the Kalman filter for tracking and uses a low-level recognition system to properly distinguish between the many trackers. This low-level system is created using a face database of twenty unrelated people trained using Modular Principal Component Analysis (MPCA) and classification is performed using a feature correlation metric. After tracking the faces, they are then analyzed by a high-level face recognition subspace which is created using a large database of people and processed using Adaptive MPCA. The overall system is shown to provide reliable tracking of more than one person and to allow a more accurate recognition rate due to the ability to create a time-average of the recognized faces.

Visual Information Fidelity in evaluating Retinex enhancement algorithms

Developing image enhancement algorithms require that there exists a proper evaluation scheme to compare different algorithms. Most objective algorithms fail because they are not designed for the subjective nature of the enhancement problem. This necessitates an evaluation scheme that would correlate with scores provided by a group of human observers. In this paper, we consider the quality assessment of images which are visually improved using various versions of Retinex algorithm. The objective of this paper is twofold. To check the suitability of available current distortion based quality assessment methods to evaluate nonlinear image enhancement algorithms and to check if consistently better objective scores are obtained for better Retinex algorithms. To this end, we find correlation between mean opinion score obtained from subjective evaluation and objective quality score for each of the methods under consideration.

A novel edge detection approach on active contour for tumor segmentation

Active contour methods for image segmentation allow a contour to deform iteratively to partition an image into various regions. A novel region-based active contour model proposed by Kaihua Zhang which take advantages of both geodesic active contours and Chan-Vese model is discussed in this paper. This method uses region-based signed pressure force (SPF) to stop the contours at weak or blurred edges. In practise, this method is useful only for images with clear boundary. Second, this method is strongly depends on parameter α which controls the propagation of moving curve. α value depends on size of region to be extracted. There is no fixed α value which works well for all images. In this paper, we proposed new method of edge detection for active contours based on local adaptive threshold technique via variational energy minimization to stop the contour at desired object boundary. By using proposed method, we can control the evaluation of curve with moderate value of α for all images. As an application, our method has been used for tumor segmentation from magnetic resonance images and experimental results show desirable performances of our method.

Application of Quad Tree for low bitrate compression

Quad Tree algorithm is an efficient method for image compression at lower bit rates. It divides the image into four equal quadrants based on a threshold. The threshold is chosen from image characteristics using Otsu thresholding method. The image regions obtained from the Quad Tree decomposition are approximated by using a first order polynomial. Polynomial approximation coefficients now represent the pixel values. Its performance is better than JPEG at lower bitrates.

A clustered locally linear approach on face manifolds for pose estimation

Data points with small variations between them are assumed to lie close to each other on a smooth varying manifold in the feature space. Such data are hard to classify into separate classes . A sequence of face pose images with closely varying pose angles can be considered as such data. The pose angles when large enough create images that are largely differing from each other, and thus, the sequence of face images can be assumed to be on or near a nonlinear manifold. In this paper, we propose an unsupervised pose estimation method for face images based on clustered locally linear manifolds using discriminant analysis. We divide the data into multiple disjointed, locally linear and separable clusters. The problem of identifying which cluster to use is solved by dividing the entire process into two steps. The first step or projection using the entire smooth manifold identifies a rough region of interest. We use clustering techniques on entire data to form the pose-dependent classes which are then used to find the first set of discriminant functions. The second step or second projection uses trained cluster(s) from this neighbourhood to obtain a second set of discriminant functions. The idea behind such an approach is that the local neighbourhood would be linear and provide better between-class separation, and hence, the classification problem would now be simpler.

Face pose estimation for driver distraction monitoring by automatic clustered linear discriminant analysis

Smooth varying data is hard to classify/divide to separate classes since there is small separation. Large number of close and adjacent poses create smooth varying manifolds. Manual class formation by selecting different data points from entire database into different training classes will affect the error rate in smooth varying data classification. This paper proposes classification of smooth varying data based on clustering and discriminant analysis. The clustering process results in different clusters which can be used for classification based on discriminant analysis. The automated class formation based on the data points in the manifold reduces effort of manual clustering and it gives very comparable results. This pose estimation can be used as a measure of driver distraction monitoring.

An Effective Surround Filter for Image Dehazing

Atmospheric moisture, dust, smoke and vapor result in haze which tends to produce a distinctive gray or bluish hue and diminishes visibility. Acquired images can be used in applications such as surveillance, object identification, classification etc. only if the effect of weather is removed from them. One of the popular existing haze removal algorithms uses a dark channel prior based approach. Though this approach gives very good results, it is computationally complex. Retinex theory, which is widely used in image enhancement, can be applied effectively for haze removal also. Retinex theory is based on illumination -- reflectance model. It also makes use of the theory of homomorphic filtering [1] which simultaneously normalizes the brightness across an image and increases contrast. We combine the ideas of dark channel prior and Retinex methods to obtain a haze removal technique that gives good results, and is computationally simple compared to the existing methods. Image quality assessment methods help us compare the quality of the dehazed images.

Color Image Dehazing using Surround Filter and Dark Channel Prior

Abstract Outdoor images are often degraded by haze, resulting in a distinctive gray or bluish hue which diminishes visibility. Of the existing haze removal methods, the ones that are effective are computationally complex and memory intensive. In this paper, we propose a simple haze removal technique, whose computational complexity is that of a simple convolution. To this purpose, a center surround filter is employed to improve speed and memory requirements of the transmission estimation in image dehazing. This can be useful for real time applications such as driver assistance, runway hazard detection and surveillance. The proposed technique relies on deriving an alternative transmission estimate by filtering the input image in three different color spaces, namely RGB, Lab and HSV. The effectiveness of the proposed method is compared with that of other state of the art methods using a subjective quality assessment method and a number of objective quality assessment methods.