| With the development of network, electronics and multimedia technology, digital image and video has become one of the most popular media in social life, and plays an important role in various fields. However, noises are often introduced during the capturing, transmission and storage process of digital images and videos. In order to ensure good visual experience and minimize the impact on subsequent image and video processing, choosing the proper method to suppress noise and recover original information is particularly important. Based on the above, the thesis focuses on the study of image and video de-noising algorithm.State of art of image and video de-noising algorithms are introduced and analyzed. For image de-noising algorithms, various common algorithms are analyzed and compared, which are classified into space domain and transform domain. For video de-noising algorithms, various representative algorithms are explained and compared, which are preliminarily classified into time domain and time-space domain and further subdivided according to motion estimation.To eliminate the Gaussian noise of videos, a time-space bilateral filtering algorithm is introduced, which belongs to the Local Filter. The algorithm uses the correlation between frames of video sequences, and introduces the time domain similarity factor, representing the similarity of pixels in different frames, for the operation of video sequences de-noising, together with the space proximity factor and gray similarity factor. Simulation results on Matlab demonstrate good effect of video de-noising is achieved in subjective and objective evaluation system.Then, to eliminate the Gaussian noise of images, a self-adaptive non-local de-noising algorithm is proposed, which belongs to Non-Local Filters. The algorithm uses improved Sobel operator for division of smooth region and edge region of images, switch-based gray similarity algorithm for smooth region filtering and Non-Local Means method for edge region filtering. Experimental results show the algorithm significantly reduces the complexity of traditional Non-Local algorithm and has a good de-noising effect. A detailed analysis is conducted based on evaluation system containing such aspects as subjectivity, objectivity and complexity.Finally, time-space domain bilateral filtering algorithm and self-adaptive non-local de-noising algorithm are implemented on OpenCV, which has achieved a good processing effect. |
