Quality Assessment Of Remote Sensing Image Based On Structural Similarity

At present,a large number of image quality assessment algorithms have emerged and achieved good evaluation results in the field of image quality assessment.Remote sensing image is a kind of high resolution and high altitude shooting natural image,which is different from common natural image.Therefore,it is necessary to combine with the characteristics of remote sensing image to study the quality assessment method,and those characteristics are more in line with human vision.Since the current quality assessment of remote sensing image cannot acc.urately reflect the visual perception of human,a full reference and a no reference quality assessment are proposed in this paper,which are more suitable for remote sensing image application scene.The human eye is more sensitive to the edge area and its contrast on both sides and the texture area itself contains most of the structural information of the whole image.However,SSIM treats all pixels equally and doesn't conform to the human visual characteristics.Based on this,this paper presents a full reference quality evaluation algorithm based on edge and texture feathers(ET_SSIM).Firstly,non-subsampled contourlet transform is used in this thesis to perform multi-scale decomposition.Secondly,the edge area is detected by K-means and thinned by Mathematical morphology.And texture region is extracted.Finally,we use the edge area to calculate the contrast component and used texture area to calculate the structural similarity component.Because of the lack of subjective assessment database of remote sensing image,a remote sensing image database is established to verify the accuracy of the method.The experimental results show that the linear correlation between ET_SSIM and MSE,PSNR,SSIM,GSSIM has improved by 22,8%,6.7%,1.3%,0.7%respectively.On one hand,In order to get a more general method of non-reference remote sensing image assessment,First of all,the reference image is constructed by "re-descending",then the edge texture region and the smooth region are evaluated separately.ET_SSIM assess the edge texture regions are assessed by ET_SSIM,smooth areas are assessed by SSIM,Last,According to different emphasis on the edge texture and smooth area of the human eye,weighted sum to get the final result VS SIM.On the other hand,in the same SSIM value of different subjective feelings of remote sensing image set,self-built database and LIVE database experiments show that the monotonicity and consistency of VSSIM is slightly lower than ET_SSIM.But as a non-reference method,its accuracy is significantly better than SSIM.And the linearity of VSSIM method is 1.3%higher than the current mainstream non-reference method BLIINDS-?.