Research And Implementation Of The Image Quality Assessment Method

In recent years, as the picture processing technology development, processing and compression performance has improved, The needs of digital multimedia technology is more and more widely. How to accurately evaluate the effectiveness of these processing technology has been a new subject in this field. The image quality is compared various image processing algorithm to optimize the system performance and quality parameters of the important indicator, so in the image collection, coding compression, network transmission areas to establish an effective image quality evaluation system is of great significance. The image quality evaluation can be divided into two categories of objective quality evaluation and subjective quality evaluation. Subjective quality evaluation is the most reliable method of quality evaluation, but because of its inherent limitations, making the objective quality assessment of the emergence and development. Objective quality assessment methods can be divided into full reference, reduced reference and no reference to three categories.This thesis focuses on the study of image quality assessment based on structural similarity. First, we analysis of the original SSIM method, it is a top-down process, simulate the whole function of HVS. But a simple linear modeling of SSIM is difficult to model the high-level visual processing of the image structural information. It is not satisfied with human visual characteristics in the evaluation of the type of Gauss blur, white noise distortion images. This paper presents the WMSSSIM algorithm based on weighted multiscale structure similarity. It decomposes the image into multi-layer with the low-pass filtering so that the model can simulate the human visual characteristics more accurately, and can contain more objective conditions, and have better evaluation results. By using the LIVE2 image database of 5 kinds of distortion type, 982 images as the experimental material, we can confirme that WMSSSIM method is more accurate and effective than the original SSIM method.In addition, we study the no reference image quality evaluation, focus on the distortion type of blur image quality evaluation effectively, we introduce a kind of no reference image quality evaluation method based on blur perceptron. The more blurred an image is, the more slowly the trasition is between neighboring pixels, the difference of the neighboring pixels is smaller.So the basic idea of blur image evaluation method is that blur the original image first, then analysis the change of the neighboring pixels. The experiments prove that this method is better than the method of several classic early :Mean-gradient, Variance, Entropy. We though the method is good enough, we find some shortcomings of it and propose an improved algorithms based on it: a blur metric based on neighboring variation of second difference matrix. The new method is that Compute the absolute difference matrix of original image first, and then compute the second absolute difference matrix of the absolute difference matrix, compute the second absolute difference matrix as final assessment metric at last. By using 174 images with Gassin blur distortion of LIVE2 image database as the experimental material, we can confirm that the improved methods are better in veracity, consistency and monotonousness.