An Enhanced Image Dehazing Scheme for Scene Classification

Abstract

Hazy images retrieved in today s world are processed in a number of applications to automate multiple benefits. Starting off with scene images, captured images are rendering service to a number of real time problems and solutions. Scene and geographical analysis domain play a significant role in addressing the needs of growing populations. In computer vision applications, texture, color, contrast, illuminations, fog, haze, and multiple visual features are considered for understanding the images. Due to environmental factors, the visibility of images is degraded. Many researchers have proposed multiple algorithms for enhancement in gradient domain as part of dehazing but fails to restore it with good quality preserving the color and contrast features. When an automated system attempts to increment the gradient values, the original low dynamic range input evolves into high dynamic range images. The linear dynamic range compression employed on images may not return with expected quality, making the image more shadowy blurring some of the details. The color restoration method of the two channels excluding the lightness channel is done proportional to the inverse of the transmission which itself does not consider the color, contrast and color distortion of the entire image. The intention of the automated system is affected as quality of colour and contrast features are decreased. These limitations have opened the research area for applying a new aspect that best fits for compression, enhancement of the visible attributes and classification of the scene images in the existing system. newlineA lossy compression technique is used instead of linear dynamic range compression for the reduction of more than required enhancements. Images are converted into simpler numerical representations and channelized Singular Value Decomposition technique is applied for the process of balancing improvements. Pixels that are to possess great resemblance to nearby pixels may be removed from storage as they are replicated. As part of color corre