盲复原图像振铃效应评价  

No-reference Restoration Image Ringing Metrics

作　　者：左博新[1,2] 明德烈[2] 田金文[2] 

机构地区：[1]中国地质大学(武汉)计算机学院,武汉430074 [2]华中科技大学图像识别与人工智能研究所,武汉430074

出　　处：《中国图象图形学报》2010年第8期1244-1253,共10页Journal of Image and Graphics

基　　金：中国地质大学(武汉)中央高校基本科研业务费专项(CUGL09030)

摘　　要：线性空不变盲复原算法通常会在较为明显的边缘处产生振铃效应。振铃现象主要受噪声,复原算法种类,以及复原算法参数的选择的影响。提出了一种盲复原图像振铃效应评价方法。方法根据振铃效应的不同类型,使用Gabor滤波器,共生向量等方法分别对其进行评价,最后提出整体的振铃效应评价方法。实验结果表明,该方法可以有效地评价不同复原算法和不同复原参数下的复原图像中的振铃效应,评价结果符合主观评价结果。Linear space-invariant image restoration algorithms often introduce ringing artifacts near sharp intensity transitions. It is shown that these artifacts are attributable to the noise, the different of the restoration parameters and the restoration algorithms. This paper presents a blind deconvolution image ringing metric. According to different ringing artifact types, two ringing metrics are presented to measure the ringing artifacts in restoration images by using Gabor filter and Co-occurrence vector etc. Finally, a whole ringing metric is proposed. The experimental results validate the proposed method perform well over a wide range of restoration image ringing levels and restoration algorithms. And the proposed model is in good agreement with MOS.

关 键 词：图像质量评价 复原 振铃效应 边缘 测度 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置]