Image is an important way of perceiving information and communicating. Image quality is closely related to the adequacy and accuracy of the obtained information. However, images have the problem of quality decreasing during acquisition, storage, transmission and reconstruction process. So image quality assessment would play an important role across the chain of image processing, as preprocessing or result analysis. For human being is the ultimate receiver of the image, the subjective method is the most accurate way of evaluation. The results of the subjective evaluation method are also used for assessing the performance of the objective evaluation method. But the subjective evaluation method is conditioned by factors such as the experimental environment, display equipment and the psychological emotion experimenters. The subjective method is very time-consuming and requires a lot of professional players, difficult to be automatic. Image quality assessment algorithm is generally embedded in image processing system, as part of the image preprocessing. But the subjective evaluation method cannot realize online working, so the focus of the study is the objective evaluation method. In this paper, the ringing artifacts produced during image deblurring and the multi-scale and multi-direction characteristics of the Contourlet transform are analyzed. Combined with the characteristics of relative error in frequency domain to distinguish the ringing artifacts, an image quality assessment algorithm based on Contourlet transform is proposed. Firstly, the original image and the distorted image are decomposed to three levels by Contourlet transform. Then the error in each layer is calculated. Finally, weighted by the correlation coefficient between the error statistics in each layer and subjective score, the index of overall image quality is computed. Through the above algorithm, we can get the evaluation results consistent with the observer's subjective feelings. Experiments show that, this method is closer to human perception than other algorithms like the Peak Signal-to-Noise Ratio.
修改评论