Histogram Analysis and Pattern Recognition of Sky Region in Defogging Algorithm

Model-based image dehazing algorithms have theoretical defects of dealing with the sky region, and the existing methods of sky protection are with non-adaptive parameters or high computational complexity. Based on experiments we found that the sky area has obvious pattern characteristic, a separable peak distribution, in the histogram of image. Based on this discovery, a simple pattern recognition method is proposed to decide whether there is sky region in the image, and then to compute the confidence coefficient of sky area. In fog-removal applications, the sky-related parameter can be set adaptively. Comparison to the tolerance mechanism, a novel method of data correction is presented with better effect. More testing results based on an image database with1500 pictures show that more than 90% images can be exactly distinguished by the proposed method, which in turn verify that the ideas and methods in this article are reasonable and effective.