关键词: 图像去雨雪, 局部特征约束, 基追踪去噪（BPDN）, 二次规划, 雨掩模

Abstract:

When video image data is collected, it is often affected by weather and other factors, which makes it difficult to further study the information in the video image data. In recent years, more and more researchers have paid more attention to video image data preprocessing about rain and snow weather which is one of the most difficult research topics. In order to solve the problem that the traditional Basis Pursuit Denoising（BPDN） algorithm has not considered the local features of rain and snow images, an image deraining and desnowing algorithm based on improved BPDN is proposed. In order to improve the rain streaks recognition rate and thus the rain streaks removal effect of an image, the local feature constraint theory is introduced into the image deraining and desnowing by adding the local information of the training image data to the refined rain map when solving the sparse coefficient optimization problem. The obtained optimization problem by introducing the local information of the training image data can be viewed as a quadratic programming problem, which provides a theoretical base for using BPDN algorithm. The deraining and desnowing experimental results on synthetic and real images show that the improved BPDN algorithm is better than the traditional BPDN algorithm in algorithm convergence, refined rain map recognition and rain streaks removal effect of an image.

Key words: image deraining and desnowing, local feature constraint, basis pursuit denoising, quadratic programming, rain mask