一种时间优化的全参考图像质量评价算法

摘要

Aiming at the problem that the algorithm based on structural similarity has a disadvantage of high computational complexity,a novel optimization algorithm is presented.The computation is accelerated by using integral images for the mean computation in luminance term; For the computation of variance in both contrast term and structure term,the gradient magnitudes of the image is adopted to substitute the variance and lower the complexity.Finally an integer template of 8 × 8 is employed to approximate Gaussian window.Experimental results show that the proposed algorithm has nearly the same prediction accuracy as SSIM,but its running time is only a quarter of SSIM's running time.%针对已有的基于结构相似性的图像质量评价算法的计算复杂的不足,提出一种新的优化算法.针对亮度项中的均值计算,利用积分图像进行的加速计算;针对对比度项和结构项中的方差项计算,利用梯度幅值进行简化计算;最后采用8×8整数窗口逼近多尺度高斯窗口.实验表明,提出算法的评价能力和SSIM相当,运行时间只是SSIM的四分之一.