The Removal Of Rain And Snow From Video Images Based On Statistical Learning Of Spatiotemporal Property

With the development of technology, computer surveillance systems are widely used in our daily life. However, such systems are designed under normal weather conditions which don’t consider the impact of bad weather such as rain and snow. In case of rain, the raindrops will produce rivulet in videos. While in case of snow, the snowflake will generate salt noise in images. The noise from rain and snow will degrade the video images and reduce even loss the system function. Therefore, it is necessary to denoise video images impacted by the bad weather and to ensure the stability of surveillance systems by enhancing the video input in preprocessing stage.The thesis mainly summarizes the physical properties and dynamic model, studies and analysis the existing methods about the removal of rain and snow from images, and presents an algorithm based on the statistical learning of spatial-temporal property. In static scene, it uses the Kalman Filter Algorithm to remove the raindrops and snowflakes directly.from video images, While in dynamic scene, it use the model of spatial-temporal property of rain and snow pixels to distinguish the rain or snow pixels and moving pixels, fix the rain and snow pixels obtained by the classification and achieve the purpose of removing rain and snow.At last, a video enhancement algorithm for the removal of rain and snow from video images is designed and implemented. Large numbers of experiments show that the system can effectively remove the rain and snow from video images and it can also meet.the real-time requirements.