一种基于稀疏编码和字典学习改进的单幅图像去雨方法

An Improved Method for Single-Image-Based Rain Streaks Removal via Image Decomposition

在日常生活中,恶劣天气例如雨、雾等时常出现,对我们的生活造成一定的困扰。同时也一定程度地影响着我们日常使用行车记录设备甚至户外摄像头的使用。随着电子设备的快速发展,电子设备的性能快速提升并且价格逐步下降,因而行车记录设备和户外摄像头的普及率逐渐提高。由于恶劣天气导致此类设备获取的带雨线图像的可利用价值降低的问题变得日益突出。提出一种基于经典的稀疏编码和字典学习的单幅去雨图像的改进方法,加入基于二阶导数的Laplace变换和非局部均值滤波器加强了预处理过程。从而弥补原方法在预处理过程中,分离所得到低频部分的图像缺少足够的纹理细节。经过实验证明,该改进能够明显地使得图像的滤波过程中获得更多的细节部分。

In daily life,bad weather conditions such as rain,fog usually occurs,causing some trouble to our life.Meanwhile,it damages how car driv ing recorders and outdoor cameras work in our daily life,to some degree.With the fast development of digital equipment,its performance gets outstanding improvement and price declines slowly.Therefore car driving recorders and outdoor cameras has become more and more common.The value of images with rain steak captured by car driving recorders and outdoor cameras has decreased,attracting more and more attention.Based on such situation,we proposed an improved method based on traditional single-image rain steak removal,in which we used Laplace transform with the Introduction of Second Derivative and both Non-local means filter and bfilter to enforce preprocessing.Then in the final image,the low frequency image will get more detailed content.With the demonstration,the final image get more detailed content.