Image de-blurring Model based on Machine Learning

Due to the influence of haze in winter, outdoor images usually lose contrast and fidelity. In view fact that most de-fog algorithms are not effective for with large sky areas, an improved dark channel a priori method is proposed. First all, region segmented according image gradient information, on basis segmentation, atmospheric light value reasonably estimated by setting discriminant formula combined high brightness smoothness reference pixels. Secondly, different values, piecewise linear function used dynamically modify adjustable parameters solve local shadow caused excessive defog. Then, transmittance bright model prior fused, edge optimized guided filtering. Finally, scattering model, defog obtained compensation stretching. The experimental results show can effectively improve distortion, enhance details, especially maintaining visual authenticity region.