摘要： 针对图像质量的局部评价融合问题,基于结构相似度(SSIM)评价算法,研究各种类型滤波器加权因子、滤波器窗口尺度及滤波器标准差参数对SSIM算法性能的影响。基于参数优化的高斯滤波器加权因子,改进峰值信噪比(PSNR)和奇异值分解(M_SVD)2种图像质量客观评价算法 ,通过优化的高斯权重对图像子块进行局部评价,在融合局部评价结果的基础上获得总体评价结果。实验结果表明,改进PSNR算法的预测斯皮尔曼等级相关系数、皮尔逊相关系数和均方根误差评价指标分别提高了3.78%,2.40%和2.02%,改进M_SVD算法的对应指标分别提升了 1.78%,0.67%和4.99%,具有较高的评价稳定性和实时性。

关键词: 计算机视觉, 图像质量评价, 图像滤波器, 高斯加权, 结构相似度

Abstract: Aiming at the local assessment fusion problem of image quality,the effect of different filters along with their weighting factor,window size and standard deviation on the performance of Structural Similarity(SSIM) algorithm is studied in the paper.Two algorithms named Peak Signal to Noise Ratio(PSNR) and Mean Singular Value Decomposition(M_SVD) for objective image quality assessment are improved based on optimized parameters of Gaussian filter.Overall evaluation results are achieved by assessing local blocks with Gaussian weight and fusing local assessment.Experimental results show that the Spearman Rankorder Correlation Coefficient(SROCC),Pearson Correlation Coefficient(CC) and Root Mean Square Error (RMSE) of the improved PSNR algorithm are enhanced by 3.78%,2.40% and 2.02% respectively.The above indexes are improved by 1.78%,0.67% and 4.99% respectively for the M_SVD algorithm.The two improved algorithms hold higher assessment tability and real-time performance.

Key words: computer vision, image quality assessment, image filter, Gaussian weighting, Structural Similarity(SSIM)

中图分类号: 

• TP391.41