Effective image dehazing using multi-objective optimlzation

Abstract

Environmental issues like snow, rain and haze badly effect visibility of image. The degraded quality of images ineptly effects performance of automated surveillance and tracking systems. Images enhancement and dehazing has an extended range of appli­ cations, such as consumer electronics and surveillance systems, autonomous robotics Systems and military tracking. Our atmosphere contains large amounts of particles that interfere with all outdoor photography. They make scenes appear foggy or hazy, which means that the visibility and contrast of objects is reduced and affects how easily or difficultly we can recognize objects in images. Recently, however, developments in computer vision have shown that it is possible to improve exterior photos and that haze can be removed. Many computer vision applications can benefit from haze-free im­ ages. These techniques are based on theories from meteorology and other disciplines. Unfortunately, these techniques are very expensive when it comes to complexity. A new measurement method is introduced for the evaluation. In this work, Dark Channel Prior (DCP) is common to reduce the haze from the image and then applying expo­ nential intensity transformation using genetic algorithm. Calculating the expon value using genetic algorithm shows the improvements in the results. This work shows that it is not only fast processing but also give the better statistical results. An optimal com­ bination of these two techniques is generally considered to lead to the desired solution. These results allow new applications such as high-resolution, high-frequency exterior surveillance, on-board camera applications on vehicles and many others to use for haze removal. Various computer vision based techniques are used to enhance quality of de­ graded images. Different physical modelling approaches are promising to produce the better results in terms of visuals, color fidelity and contrast improvement are computa­ tionally expensive. Computer vision based techniques are computationally efficient but sometimes result in weakening the important information. The aim of present thesis is to carry out literature background and analytical analysis of various methodology to present an effective approach for image dehazing while taking care of complexity and statistical result performance.