Research On Image Quality Assessment Based On Contourlet Transform

Image is the important sources of information in the human perception and machine recognition, its quality play a decisive role in the adequacy and accuracy of information. However, quality becomes lower in image compression, reconstruction and transmission, so the image quality assessment have very important role in image processing. Image quality assessment method has become one of the very important issues.As the human visual system (HVS) is the ultimate image processing handler. Therefore, subjective image processing algorithm is the most reasonable image quality algorithm. The algorithm needs an objective evaluation and subjective evaluation of the results compared. But the subjective evaluation method is very time-consuming and requires a lot of professional players, also constraints by many objective conditions, such as observing environment and image display devices with different quality. Image quality assessment algorithms generally require embedding in other image processing algorithms, as part of image pre-processing. But the subjective evaluation method cannot be embedded into other algorithms, so now the focus of study is the objective evaluation.This paper analyzed multi-resolution characteristics of the human visual system and multi-directional multi-scale features of Contourlet transform. An image quality assessment based on Contourlet transform algorithm is proposed, it's more in line with human visual assessment of the results. Firstly, in the proposed algorithm, with Contourlet, the original image and distorted picture is decomposed by 4 levels, then analyzes each layers on structure similarity (SSIM), and finally get a normalized value of the objective image quality assessment. In this algorithm, the paper also studied quality evaluation method based on Contourlet transform and human visual system image, images are converted into Contourlet transform domain coefficient, then through some of the underlying characteristics of HVS modeling, distorted image and the absolute error between the reference image can be mapped to the human eye perceived the Just Noticeable Difference (JND) units. If the error is higher than the threshold of visual sensitivity, indicates its ability to be aware of the human eye, or can be ignored. Through the above treatment and observation the results are consistent with the subjective feelings of the evaluation results. For example, visual sensitivity function (CSF, contrast sensitive function) is applied on Contourlet transform domain to simulate the evaluation of image quality. Experiment results show that evaluation of this method is closer to human perception than Peak Signal-Noise Ratio.