A New Video Denoising Algorithm Based On Adaptive Polyview Fusion

With the progress and development of digital multimedia technology, digital video has become ubiquitous and indispensable in our everyday lives. However, video signals are subject to noise contaminations during acquisition and transmission. It not only reduces the clarity and quality of video signals, but also affects other video processes, such as compression, segmentation, object detection, recognition and tracking. So it is highly desirable to remove the noise to enhance video quality.The existing video denoising methods can be roughly divided into two categories:2-D methods and3-D methods. In the2-D methods, it has two approaches for denoising video signals:frame by frame using image denoising methods and adding information of adjacent frames. In the3-D methods, all available space-time information are used for denoising, resulting in better performance but the efficiency is lower than2-D methods. The recently proposed Polyview Fusion algorithm is a compromise between them. It applies2-D denoising methods to three views (front, top and side views) of the same noisy video signals and then merge the denoised results into one, so that the performance is improved.In general, denoised result in smooth content is better than that in texture content; and if there is no significant motion in the video signal, the top or side view is mainly consisted of smooth content. Based on it, a new adaptive fusion method is proposed. Three denoised results are transformed into top or side views. And then they are divided into smooth region and texture region by the variance of the same pixel at a time interval, i.e., its focus is on the same pixel’s difference in the time index, so it is same to top view and side view. In this paper, we use the top view. Next the different regions are given different weights based on local features to be merged into one. After transformation, it obtains the final front view result. Experimental results confirm the better performance of the proposed method for the video sequences without significant amount of rapid motion.