Research on NO-reference Image Quality Assessment Based on HVS
|School||Jiangsu University of Science and Technology|
|Course||Signal and Information Processing|
|Keywords||Human visual sensitivity No reference Image Quality Assessment Support Vector Regression Neural Network|
Image quality evaluation of image processing plays an important role in guiding. Difficult to obtain in the case of the original image, the image of the non-reference evaluation method has been widespread attention. Image human services determines its nature the evaluation results must be consistent with human visual system characteristics, therefore, image quality assessment is to design efficient algorithms to derive consistent with human subjective perception of the evaluation value. JPEG image is still current network and database of the most widely used and most popular image formats. Accordingly, this thesis based on human visual system without reference JPEG image quality evaluation method for a more in-depth research, research main contents are: 1. Studying the current image quality evaluation method based on the focus on a non-reference picture quality evaluation methods, pointed out the influential non-reference evaluation. (2) the characteristics of the human visual system is summarized as masking effect and visual sensitivity two characteristics. For masking effect, in the DCT domain using different mathematical models to extract texture edge masking effect masking effect and brightness characteristics, as a measure masking effect characteristic indicators. For human visual sensitivity by using the different filter extraction operator that best reflect the human visual sensitivity to amplitude and length of the edges, and brightness of the background activity of these four characteristics. Experimental results show that, using the extracted feature mathematical model showed good discrimination. 3 masking effect is proposed based on the non-reference image quality evaluation method. Discrete detected using a DCT block extraction best reflect the value of blockiness masking effect, a synthesis method by Minkowski masking effect may reflect the evaluation index, to achieve the function of image quality evaluation. Experimental comparison shows that the evaluation can better capture the human visual attention mechanism, better reflect the average subjective evaluation of image quality values. 4 presents a new vision based on human sensitivity to non-reference image quality evaluation method. Using support vector regression neural network approach to find and evaluate image quality characteristics of human visual sensitivity and the average value of subjective evaluation functional relationship between the use of edge magnitude and length, background activity and brightness and other visual sensitivity characteristics, to achieve consistent with human visual characteristics without reference image quality assessment function. Experimental results show that support vector regression neural network self-learning ability to automatically add new characteristics of the sample, with excellent generalization capability and universality, evaluation results of the obtained image with a higher average value of subjective assessment of the consistency, adequate embodies the human visual characteristics in the image quality assessment role.