Research Of Noise Imgae Quality Assessment Algorithms

The development of information technology and computer technology have greatly enriched the ability of people to access and share information. Image is more intuitive and convenient in information transmission than other types of information such as audio, text, etc. It is generally known that noise is inevitably introduced during the image generation and transmission process. Noise image quality assessment mainly includes two aspects: the distortion degree assessment of noise image and noise type identification and noise parameter estimation.The distortion degree assessment of image can determine the degree of image distortion, severe distortion image are no use, therefore the research on the image distortion assessment can determine whether the image can be further processing or be discarded. This paper makes a detailed introduction about the relevant knowledge on image quality assessment, including the human visual system, subjective quality assessment, objective quality assessment, full-reference assessment, no-reference assessment and the consistency of subjective and objective image quality assessment and so on. Many noise filtering algorithms require some priori knowledge of the image noise, such as noise type and noise intensity. When this prior knowledge obtained, the appropriate filter algorithms can be selected and appropriate filter can be designed, thus improving the visual quality of image. At present, most of these priori knowledge are obtained through observation of people’s subjective speculation, which has a certain blindness and lack of systematic and objective derivation. Therefore, the effective assessment of noise image quality is of great significance, is helpful to reduce the blindness and enhance the process result of the image processing system.This paper makes a brief introduction of the degree of noise image distortion evaluation and focuses on noise type identification and parameter estimation. On the basis of summarizing the predecessor’s results, two improved algorithms were proposed:1. Image noise type identification and parameter estimation based on wavelet transform. It makes use of the characteristics of wavelet transform that separate the high-frequency signal from low-frequency signal and noise belong to the high-frequency signal. Using the characteristics that different types of noise exert different impact on the high-frequency signal, this algorithm identified noise type according to the histogram of high-frequency coefficient. Noise intensity is estimated by using the effect of noise intensity on high-frequency energy. The difficulty lies in the image details, edge, texture and noise are all belong to high-frequency signal and mutual aliasing. Therefore, a reasonable selection of high-frequency component is of great importance for accurate estimation result. Experimental results show that the algorithm proposed has a higher success rate of identifying the noise type and a more precisely estimated result over the algorithm proposed by Zhang Q et al, improved about 3%～7%.2. Image noise intensity estimation based on the image block structural features degree. It divides the image into sub-block first, using five direction feature vectors to study the relationship of pixel sampling and combined with the proportion of edge points in image sub-block which obtained by the result of edge detection determines the simple smooth block. Combining the weight based on the sub-block structure feature degree with the image block variance, the final estimated result will be obtained by weighted sum. Because of the combination use of 360 degree structure feature degree and edge detection, the calculation amount increased a little and the precision is higher than the algorithm proposed by Wang Jing et al.Finally, the result simulation result shows that the algorithm proposed is valid and precise. This paper involves the selection of multi threshold and mainly discussed two typical noises, thus make it adaptive for most image and take other noises into account should be the direction of future research.