An Improved Image Restoration Algorithm for Overcast Based on MSR

Abstract

According to degradation model of cloudy images, the paper uses MSR (multi-scale retinex) algorithm for restoration. But MSR algorithm can’t effectively restore the details and color of images in processing cloudy images, so the paper proposes a new MSR improvement algorithm to process cloudy images. Three scales of traditional MSR is changed into four scales, which means that a Gaussian function with middle scale keeping the details and color of images is added to. And information fusion strategy based on wavelet transform domain is substituted for linear weighted strategy of multi-scale reflection images for MSR algorithm. The basic idea of fusion is that: Firstly, the images to be fused are for two layers of wavelet decomposition. Then high-frequency component are taken the absolute value maximum to stress the details in image. And local energy method is used for low frequency component to adjust the background and color, which realizes the effect of fidelity. Lastly, subjective observation and objective evaluation in the paper indicates that the algorithm of the paper has better effect on details restoration and color fidelity than the traditional MSR algorithm for restoring cloudy images.