一种基于视觉特性的PSNR图像评价方法  

Image quality assessment using PSNR based on visual feature

作　　者：丰明坤[1,2] 赵生妹[1] 邢超[1] 

机构地区：[1]南京邮电大学信号处理与传输研究院,江苏南京210003 [2]浙江科技学院信息与电子工程学院,浙江杭州310023

出　　处：《南京邮电大学学报（自然科学版）》2015年第4期33-38,共6页Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition

基　　金：国家自然科学基金(61271238)资助项目

摘　　要：针对峰值信噪比(PSNR)图像评价算法性能较低的问题,提出了一种基于视觉变换域的(VPSNR)改进方法。VPSNR基于二维log-Gabor变换获取图像的多通道特征并计算该特征的PSNR值;然后依据log-Gabor变换的尺度效应和方向效应融合各通道PSNR值。进一步结合改进的中央凹视觉方向和色彩敏感特性,提出了彩色图像质量评价(Vc PSNR)算法。仿真实验表明,VPSNR算法的评价性能较PSNR算法有大幅的提升,主、客观评价结果的相关系数(SROCC)提高了4个百分点,达到了结构相似(SSIM)算法的水平;拟合曲线的均方根误差(RMSE)和相关系数(CC)指标则超过了SSIM算法的水平。Vc PSNR算法性能较VPSNR有新的提升。A method based on visual PSNR （VPSNR） for image quality assessment is proposed to improve the performance of PSNR. A multi-channel visual feature of distorted and reference images is extracted with log-Gabor transformation and compared with PSNR algorithm. Every channel assessment is averaged by the visual wight with different scales and orientations. Furthermore,the color VPSNR（VcPSNR） algo-rithm for color image quality assessment is proposed considering the visual fovea and chronic modulation. Experimental results show that the assessment performance of VPSNR is better than that of PSNR with im- proved spearman rank-order correlation coefficient（SROCC） by 4% ,it is equal to that of structural similarity（SSIM） algorithm. Correlation coefficient（CC） and root mean square error（RMSE） of the VPSNR are superior to that of SSIM. The performance of VcPSNR is improved compared with VPSNR.

关 键 词：log—Gabor小波 图像质量评价 峰值信噪比 视觉特征 

分 类 号：TN911.73[电子电信—通信与信息系统]