摘要
为解决数字图像处理中的质量评价问题,从自然图像统计和空域变换角度出发,提出一种无参质量评价方法.首先采用空域变换并结合梯度关系获得图像的统计特性分布;然后采用非对称广义高斯分布进行模拟,在充分考虑非对称因素的情况下求得特征参数,以反映分布特征;最后利用失真图像与原始图像统计分布的差异,将特征参数与参考标准直接进行K-L距计算获得评价值.实验结果证明,文中方法无需参考图像,适用于任意失真类型图像的质量评价,与同类方法相比,评价结果与主观DMOS值更具一致性;同时,该方法计算复杂度较低,对Live库中图像运算耗时少于130 ms,具有广泛的实际应用价值.
To solve the problem of image quality assessment, a novel no-reference image quality assessmentmethod is proposed based on natural scene statistics and spatial transformation. Firstly, statistical distribution ofimage by spatial transformation and grads are got. Secondly, asymmetric generalized Gaussian distribution modelis used to simulate the distribution. And by reflection on unsymmetrical character, parameters are calculated toshow distribution feature. Finally, diversity between distortion image and original image is compared andassessment value by K-L divergence is computed. Experiments reveal that this method can evaluate pictures withany kind of distortion without reference. The assessment results are better than other methods in accordance withsubjective DMOS. At the same time, the complexity of calculation of this method is comparatively lower, thetime for simple processing each of images in Live database is less than 130ms, which makes it great and wide value for engineering application.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2015年第2期249-255,共7页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(61108066)
国家"八六三"高技术研究发展计划(2010AA8080202)
中国科学院三期创新工程
长春光机所所内创新工程(Y10532B110)
关键词
图像质量评价
自然图像统计
非对称广义高斯分布
quality assessment
natural scene statistics
asymmetric generalized Gaussian distribution
