Development of Enhanced Image Dehazing Techniques for Security and Surveillance

Abstract

Image enhancement is very important element in forensic security and surveil- lance applications. Image enhancement is a technique which removes artifacts in image and improves the quality of information in image using different methods of image processing. Image enhancement highlights the relevant features of image and removes or reduces the irrelevant features. Image De- hazing is one of the techniques of Image enhancement which is frequently used in several image and video processing applications. Images of scenes, obtained from outdoor surveillance cameras, UAV’s, drones and satellites are sometimes degraded by the presence of different particles and the water droplets in the atmosphere. Haze, fog, smoke and clouds are such atmo- spheric phenomena due to atmospheric absorption and scattering. These processes effect the image quality and visibility by removing the haze from input image we can improve the visibility and extract more information. It also increases the color and contrast of the scenes. Image dehazing has evolved from multi-image dehazing to single image and many state of the art techniques including He, Meng, Bi etc. have been introduced. However, most of the schemes fail in dense fog conditions, resulting in halo and blocking artifacts or having high computational and time complexity. Quest to explore and effective and comprehensive image dehazing method, three schemes have been proposed. In first technique, I propose an efficient regularization method to remove haze from a single input image. Quadtree decomposition is used for accu- rate estimation of global atmospheric airlight. This method benefits much from an exploration on the inherent boundary constraint on the transmission function. This constraint, combined with a weighted L1norm and entropy based contextual regularization, is modeled into an optimization problem to estimate the unknown scene transmission. Experimental results on a variety of haze images demonstrate the effectiveness and efficiency of the proposed method. In second method, bright and dark channel priors dependent single image based dehazing in two color spaces is proposed. The hazy RGB is first converted into Y Cb Cr space. Dark channels and airlight for both color spaces are computed using multi-scale windows for accurate global value of airlight. Transmission maps are estimated using dark channel priors (DCPs) and airlight of intensity in Y Cb Cr and RGB components computed using three window sizes. Bright channel priors (BCPs) and threshold values are also computed for Y Cb Cr and RGB color spaces. The transmission maps are adjusted using DCPs and BCPs and refined using multiscale guided filter to obtain a superior dehazed image. Comparison (both visually and quali- tatively) on variety of images exhibits the significance of proposed technique over existing techniques. In third one, a refined single image dehazing technique based on contrast limited adaptive histogram equalization, luminance balancing, boundary con- straint and contextual regularization in Y Cb Cr color space is introduced. Contrast limited adaptive histogram equalization and luminance balancing is employed for improved dark channel. Accurate airlight is estimated using the balanced luminance channel. Transmission function is computed using Boundary control and contextual regularization. An effort has been made to improve the airlight estimation and transmis- sion function computation techniques for single image dehazing. Enhancing the ability to extract features like boundaries, colors and structural details from hazy images. This has been achieved through improvements in existing techniques and making use of both RGB and YCbCr color spaces.