Research On Video Noise Reduction And No-Reference Quality Assessment Based On PCA Noise Prediction

Video-image will be disturbed by noise in the process of formation and transmission,which will lead to distortion of the edge contour of the transmission picture.It is important to find an effective algorithm to restore the distorted image,and to use a scientific image quality assessment method to judge whether the image meets the human visual standards.In the process of video denoising,the video frames are regarded as continuous image sequences,which is combined with adaptive threshold motion detection to achieve denoising.Before detecting the moving area in the image,the principal component analysis(PCA)is used to estimate the image noise.Then,the PCA noise variance prediction method is combined with the improved inter-frame difference algorithm.According to the variance of predicted noise,the threshold of motion detection can be adjusted adaptively to avoid the regional "misjudgement" caused by the change of noise in transmission and improve the accuracy of motion detection.Two different denoising algorithms,spatial adaptive non-local mean(NL-means)filtering and improved time-domain weighted mean filtering,are adopted for moving and stationary regions respectively,which can improve the performance of denoising more effectively.In the process of evaluating the effect of noise reduction,this paper studies the application of universal no-reference image quality assessment.When color image is directly converted into gray image in traditional algorithm,there are some problems such as ignoring image color information and inaccurate prediction results.The HSV model of color image is applied to the no-reference image quality assessment algorithm.Using the inherent MSCN(Mean Subtracted Contrast Normalized)coefficients of the image in accordance with the characteristics of Gauss distribution,the MSCN distribution characteristics of color image in color saturation space and brightness space are extracted respectively,and the corresponding model is established by support vector regression machine.The training model was tested with the standard image quality database,and the quality score which was in good agreement with the visual effect of human eyes was obtained.The improved algorithm not only has a good coincidence with the visual effect of human eyes,but also has short operation time and strong practicability.