Advanced Multiple Linear Regression Based Dark Channel Prior Applied on Dehazing Image and Generating Synthetic Haze

Haze removal is an extremely challenging task, and object detection in the hazy environment has recently gained much attention due to popularity of autonomous driving traffic surveillance. In this work, authors propose a multiple linear regression haze model based on widely adopted dehazing algorithm named Dark Channel Prior. Training with synthetic dataset, proposed can reduce unanticipated deviations generated from rough estimations transmission map atmospheric light To increase accuracy environment, further present build COCO training dataset by generating artificial MS dataset. The experimental results demonstrate that obtains higher image quality shares more similarity ground truth images than most conventional pixel-based algorithms neural network haze-removal models. also evaluate mean average precision Mask R-CNN when preprocessing test removing model. It turns out both approaches significantly outperform existing models over images.