Geometric-Pixel Guided Single-Pass Convolution Neural Network With Graph Cut for Image Dehazing

One of the major shortcomings existing image dehazing algorithms is in estimating scene transmittance, which has assumed many items algorithms. key assumption been pixel uniformity and smoothness. In this paper, we propose to solve problem using a combination single-pass CNN with graph cut It considers transmittance based on differential pixel-based variance, local global patches energy functions improve transmission map. The proposed algorithm was tested different images evaluated various evaluation metrics. Our results show more details when compared four benchmark enhancement methods. method one drawback: over-bright areas tend lose some features final image.