Research on image de-raining method based on high scale rain pattern image block training algorithm

In this paper, for the complex situation of rain streaks in rainy and snowy weather, as well as the phenomenon of poor image quality after removing rain in the existing image processing, we propose a picture de-raining algorithm to automatically detect the location of rain streaks and reduce the blurring effect of removing rain streaks as well as the covering effect. Based on the CNN deep learning framework, the rain streaks are considered as high-frequency noise, and low-pass bilateral filtering is utilized to separate the rain streak images. The rain streak location information of the binary rain map is used to select the rainfall percentage in the high-frequency rain layer, and the data is introduced into the CNN for training to optimize the rain removal ability of the CNN model. The rain fog model is used to remove the covering effect in the image, reduce the blurriness of the rain streak image, and improve the image quality. In the algorithm performance comparison, the CNN-based rain streak image de-raining algorithm for high percentage of rain streak images proposed in this paper is able to achieve the best performance value in terms of accuracy and speed. For the four concentrations in the SPA-data test set of no fog and light rain, no fog and medium rain, no rain and medium rain and medium fog, the image processing time is less than 0.1s.