Study On Video Image Noise Reduction

In recent years, digital video has been developing fast and widely. In digital video systems, the acquisition, encoding, transmission and decoding of digital video inevitably introduce various kinds of noise, which will seriously affect not only the subjective visual quality of video but also the digital video post-processing such as coding, object recognition and tracking. Therefore, along with a wide range of digital video applications, there is an urgent need for efficient noise reduction algorithm.The temporal filter has the advantage over spatial filter in the protection of edge and detail in the video image noise reduction, but the motion estimation is required to make better use of time domain correlation. The present temporal filter based on the movement estimate is limited by computation complexity and vulnerable to serious block effects when block match is inaccurate.After the systematic introduction of the theory and development of video the movement estimation is studied. A video image 3D (combined with spatial and temporal filters) noise reduction system based on the movement estimate is proposed. First, current frame and the reference frame are down-sampled; the block is taken as the unit to carry motion estimate of current frame, and to search for the match block in the former and next frame. Then, movement intensity between current block and match block is detected. The threshold of movement intensity detection is obtained from noise standard deviation estimate unit to determine the movement intensity accurately. If the movement is small, a temporal filter will be used. If the movement is large, a spatial filter will be used. Besides, the noise standard deviation is also used as the parameter of air zone filter.The 3D video noise reduction algorithm is combined separately with the three-step search method based on down-sample and the diamond search method based on down-sample. Moreover, the simulation results of a time domain noise reduction algorithm are compared. A number of test sequences with different noise level are tested under the algorithm proposed and the time domain noise reduction algorithm. The objective quality, subjective quality and the running rate is compared and analyzed. Experimental results show that the algorithm proposed can effectively suppress the noise, protect image detail, and avoid the degradation of video image, such as block effect. Its computation speed is faster and better satisfies the time requirement of video processing.