Gray Scale Image Restoration Using Advanced Atmospheric Light Estimation

Visibility of scene is often degraded by different atmospheric phenomena which results in failure of many computer vision applications like outdoor object recognition systems, barrier detection systems, visual surveillance systems, intelligent transportation systems etc. Light coming out from the source and scattered by atmospheric particles towards the camera is the atmospheric light. In many existing image restoration algorithms the results depend on this light. So this paper introduces a image restoration method applied for fog removal which search atmospheric light in a superior way by dividing an image into blocks. Fog decreases slowly from infinite sky regions to the nearer camera regions horizontally downward direction. Consequently atmospheric light is estimated locally in each blocks from upper to lower. The global atmospheric light is the weighted average of all the local atmospheric light. The weight is calculated from histogram of each block. There are many algorithms which suffer from halo effects and edge hammering at the output. In this paper it is rectified by nonlinear filtering which is a pre-processing step. A new edge preserving transmission is also produced using this nonlinear filtering to reduce halo effects. Experimental results are verified and compared qualitatively and quantitatively with the existing haze removal methods. Comparison results shows a better performance in terms of saturated pixels and visual evaluation of different real time scenes.