An image enhancement algorithm of video surveillance scene based on deep learning

Target enhancement is the most important task in a video surveillance system. In order to improve accuracy and efficiency of target enhancement, better deal with subsequent recognition, tracking, behaviour understanding other processing targets, deep learning-based image algorithm for scenes proposed. First, super-resolution reconstruction carried out through method based on hybrid convolutional network sharpness image. Then, reconstructed scene image, watershed morphology region merging used realize Deep learning algorithms can iterative calculations. Experimental results show that after daytime, night noisy scenes, maximum difference rate less than 0.5%, cross-linking degree close 1, average time 1.3 s. It clarity scene.