Image snow removal algorithm based on snow model and deep learning fusion

Abstract

The invention relates to an image snow removal algorithm based on snow model and deep learning fusion. Aiming at the problem that the robustness of the visual system of the co-melting robot is influenced by snow weather, a simplified snow model is deduced according to the imaging process of snow, then an image deep snow removal network based on the model is designed, and the network is formed by connecting a snowflake detection sub-network and a snow removal sub-network in series. The snowflake detection sub-network adopts a residual learning network, and the network can accurately learn the difference between a snow image and a snow-free image. And the snow removing sub-network adopts a densely connected U network. On one hand, detail information of a background is reserved by using U-net, on the other hand, the snow removal accuracy is improved by using the characteristic that DenseNet multiplexes low-layer features to high-layer features, and after the detail information and the snow removal accuracy are combined, the contradiction between background detail loss and incomplete snow removal caused by excessive snow removal is relieved. Experiments prove that the deep snow removalnetwork based on the snow model can better detect and remove snowflakes in the image.

Description

technical field [0001] The invention relates to an image processing algorithm, in particular to an image snow removal algorithm based on snow model and deep learning fusion. Background technique [0002] Often, bad weather causes image degradation, which reduces the robustness of computer vision systems. At the same time, bad weather will also bring certain troubles to human vision. For example, people driving in rainy and snowy weather can easily cause traffic accidents. Bad weather conditions are generally divided into two categories: static (fog, haze) and dynamic (rain, snow). Static bad weather is usually caused by small particles floating in the air, which mainly reduce the clarity of the image. Dynamic bad weather is usually caused by large particles moving randomly, which partially occlude the image. Therefore, the restoration of images under static bad weather conditions is mainly to improve the clarity of the images, while the restoration of images under dynamic ...

AI Technical Summary

Problems solved by technology

Therefore, these prior-based rain and snow removal methods can only remove some narrow snow streaks and semi-transparent snowflakes, and they are usually unable to remove wide opaque snowflakes.

In recent years, there have been some algorithms based on deep learning to remove rain, and these methods can only remove translucent and narrow snowflakes

Images