Quality Estimation Model of Monochrome Still Picture Based on Distortion Factors and Texture Features

In this paper, we investigate the improvement of the estimation accuracy of the objective quality metric method named “the Picture Quality Scale (PQS)”. To eliminate the dependence of the picture content of the PQS, we newly employ the texture features as the independent variance and combine with the distortion factors. For constructing new PQS, we apply the nonlinear combination model between the Mean Opinion Score (MOS) and factors, which is based on the logistic function. Then, to discuss the improvement of the PQS, we carry out the subjective estimation experiment for 144 pictures and obtain the MOS. Finally, the proposed PQSnew closely approximate well the MOS, with a correlation coefficient of more than 0.98. Introduction The evaluation of picture quality is indispensable in picture coding. Subjective assessment tests are widely used to evaluate the picture quality of coded picture. However, careful subjective assessments are experimentally difficult and the results obtained may vary depending on the test conditions. Instead of subjective assessment tests, several estimation methods have been already proposed. The PQS which has already proposed as the objective quality metric method, is widely used for comparing other evaluation methods. In this paper, we investigate the improvement of the estimation accuracy of the PQS. For improving the performance of the approximation between the MOS and the obtained PQS, two key technology is employed here. At first, to eliminate the dependence of the picture content of the PQS, we newly employ the grobal texture features of image as some regression factors. Then, for constructing the new PQS, we apply the nonlinear combination model between the MOS and selected several factors, which is based on the logistic function. 419 Picture Quality Scale (PQS) The block diagram of the conventional PQS applied for monochrome still picture, is shown in Fig. 1. The distortion factors are defined as the function of the error which calculates between original and decoded picture. Before calculating the distortion factors, the visual weighting characteristic is applied to the error. In this model, the PQS is directly calculated from the distortion factors, by using the principal component analysis and multiple regression analysis. As the distortion factors, we employ 5 kinds of factors. Figure 1. Conventional PQS system. 1. ITU-R television noise weighting standard 2. weighted mean square error for visual perception 3. end of block disturbances 4. correlated errors • •