The Multi-source Feature Analysis And Extraction Of Digital Image Quality

The main way which human beings learn form is visual information that is received by human perceptional system, and the images are the main component among the visual information. With the development of the digital communication, multimedia and the network technology, digital video and image are one of the important informational intermedium in the modern world, and it is widely used in people's daily life. But in the procedure of image processing system, such as digital images coding, transmission and reconstruction, it usually caused fidelity distortion, such as blur effect, blocking affect, noise and so on. Those distortions effect the image quality seriously. So how to assess the image quality becomes to be a basic and challenge problem in the region of image processing. Through analyzing the three main distortions in the images—blur, blocking effect, noise, then extract the characteristics of the three type distortions in the processed images. At last, according to the human perceptional system, the paper designs the object image QA models of the three types distortions, and let the result of QA has a good correlation to the subject score.The main content of this paper is how to extract the characteristics of the three type distortions--blur, blocking effect, noise, accurately, and further designs the object image QA models according to the characteristics of the three type distortions. The main work and innovations are listed as follow:(1)The paper combines with the human perceptional system, analyzing and extracting the characteristics of the three type distortions accurately, and make sure the characteristics extracted are the general features of the images, and further build the perfect QA models.(2) Proposed a new reference free image QA approach aimed to blur distortion in digital image. The proposed metric extract the general feature—blur distance of the distorted image, and further using the feature builds the QA model.(3) Proposed a new reference free image QA approach base on the flat-region of the image aimed to blocking effect. The proposed metric make a good use of the human perceptional system, doing the feature extraction on the flat-region of the image, then builds the QA model using the two index of block strong and block rate.(4) Proposed a new reference free image QA using noise distortion. The gradient of the noised pixels in the eight directions are usually larger than other pixels, but to some extent, it should avoid the interference of the edge of the image. Because the edge pixels also have a big gradient, the proposed metric take the measure by wiping the maximum and the minimum gradient, it can effective make up for the shortage, and then build the image QA model using the remained six directions.(5) Proposed a multiple characteristic video quality assessment based on the information pooling method. This metric take a whole consideration the types of video distortion, the blur, blocking effect, and noise. Then using the pooling method build a high performance video quality assessment model.With the large amount of data test and experiment, show the proposed reference free image QA models'results has a good correlation to the subject score of DMOS. And it can effective process the distortions of blur, locking effect, noise.