Research On Non-reference Image Quality Evaluation Method Based On Structural Similarity

The image quality degradation will bring to us in the process of the digital image acquisition, compression, storage and transmission. With the rapid development of image processing technology, image quality assessment (IQA) method becomes to an important research direction in the field of image processing. Image quality assessment method mainly divided into the people's subjective evaluation and objective evaluation by machines. Based on the amount of information required, objective IQAs can be classified into full reference, no-reference, and reduced-reference assessments.In the actual applications, because the original image couldn't to be obtained in advance, no reference image quality assessment method is very important at present.This paper mainly studied the objective full reference assessment SSIM as well as RMS-contrast, sharpness, Fisher and MFGS of no reference assessment method. Based on the above IQAs, this paper puts forward a no reference image quality assessment method named Median filtering Structure Similarity (MFSS). We use a median filter to construct the reference image, and calculate the structure similarity between the original image and the reference image. At last take the calculated result as the original image quality measure.Using Gaussian white noise, additive pink Gaussian noise and Gaussian blur three types of distorted images of the CSIQ image database, this paper made experiment to compare the RMS-contrast assessment, sharpness assessment, Fisher assessment, MFGS assessment and MFSS assessment. The experimental results showed that:the MFSS and MFGS assessment has good evaluation results for the Gaussian white noise image, the evaluation effect of Pink Gaussian noise is next, but cannot evaluate Gaussian blur image correctly. For the Gaussian white noise images, calculated the correlation coefficient (CC) between the subjective evaluation result value (DMOS) and the MFSS as well as MFGS, the Correlation coefficient between the MFSS and DMOS is 8.4% higher than the correlation coefficient between the MFGS and DMOS, at the same time, the distinction of different level Gaussian white noise pollution is 10% higher than MFGS.