A partial reference quality evaluation method for a super-resolution reconstruction image based on wavelet transformation

Abstract

The invention relates to a partial reference quality evaluation method for a super-resolution reconstructed image based on wavelet transform. According to the technical scheme, the method comprises the following steps of firstly, carrying out blocking operation on a low-resolution image Io and a super-resolution reconstructed image Is to respectively obtain M image blocks; respectively carrying out wavelet transformation on the M image blocks; determining a threshold T, fitting the wavelet sub-band coefficient c'ji corresponding to the wavelet sub-band of the lth image block of the low-resolution image Io with the variance greater than the threshold T into a generalized Gaussian model; fitting the wavelet sub-band coefficient of the lth image block of the low-resolution image Io to a firstfitting parameter ol and a second fitting parameter ol of a generalized Gaussian model; correspondingly, obtaining a first fitting parameter sl and a second fitting parameter sl of the lth image block of the super-resolution reconstruction image Is; and finally obtaining an image quality evaluation value D. According to the method, the quality of the super-resolution reconstructed image Is can berapidly and accurately evaluated only by utilizing the information of the low-resolution image Io, and the robustness is high.

Description

technical field [0001] The invention belongs to the technical field of image quality evaluation. In particular, it relates to a partial reference quality evaluation method for super-resolution reconstructed images based on wavelet transform. Background technique [0002] Super-resolution reconstruction refers to the use of image reconstruction algorithms to process a low-resolution image to obtain a high-resolution image. The reconstructed images obtained by using different reconstruction algorithms for the same low-resolution image are different. To compare the pros and cons of reconstruction algorithms, it is necessary to provide a standard for super-resolution reconstructed images to evaluate the reconstructed image quality. [0003] To evaluate image quality objectively, the most widely used indicators for full-reference evaluation are mean square error (MSE), peak signal-to-noise ratio (PSNR) and image-based structural similarity (SSIM). The calculation of the first t...

AI Technical Summary

Problems solved by technology

This method can evaluate images with different types of distortion very well. However, only low-resolution images can be used as a reference for super-resolution reconstruction image quality evaluation, and the high-resolution original image is unknown, so it is not suitable for super-resolution reconstruction. Image part reference quality evaluation

Gao et al. (X.Gao, W.Lu, and D.Tao, Image quality assessment based on multiscale geometric analysis, IEEE Trans on Image Processing, Vol.18, No.7, 1409–1423, 2009.) proposed to use multiscale The method of geometric analysis is used to evaluate the quality of distorted images, including using Wavelet, Ridgelet, Curvelet or Contourlet transformation to extract the feature information of the image. Different transformation methods are suitable for images with different types of distortion. Although this method has high evaluation accuracy, it does not Can only target specific distortion types, thus reducing the generality of image quality evaluation

Lu Wen and other scholars (Lu Wen, Gao Xinbo, Wang Tisheng. A method for evaluating the quality of partial reference images based on wavelet analysis [J]. Journal of Electronics and Information, 2009, 31(2): 28-32.) proposed a method based on Wavelet analysis extracts features, and uses the visual characteristics of the human eye to obtain the quality evaluation measurement method of distorted images by counting the changes in the proportion of visual perception coefficients of images before and after distortion in each sub-band. There will be deviations in the proportion of the coefficients in each sub-band, and the quality of the reconstructed image cannot be accurately evaluated

Images