基于视觉显著性的运动图像去模糊研究  

Motion image deblurring based on visual saliency

作　　者：王维哲[1,2] 李娜[1] 

机构地区：[1]河南职业技术学院信息工程系,河南郑州450046 [2]河南省信息网络重点学科开放实验室,河南郑州450001

出　　处：《计算机工程与设计》2014年第8期2827-2830,2855,共5页Computer Engineering and Design

基　　金：国家863高技术研究发展计划基金项目(2008AA01A315)

摘　　要：为解决图像中运动物体的模糊问题,提出一个基于图像视觉显著性的快速去模糊机制。根据图像的显著图将其划分为视觉显著区域和非显著区域,使用改进的快速去模糊算法对非显著区域的模糊核进行预测;使用可以将显著区域运动物体的空间可变模糊转换为空间不可变模糊的补偿机制对显著区域进行补偿,把预测到的非显著区域模糊核应用到显著区域的去模糊处理中;使用反卷积算法恢复图像。实验结果表明,在保证整个图像具有统一的模糊度和不产生回铃现象的前提下,该机制具有较好的去模糊效果。Aiming at the blur of the fast moving object in video images, a new fast deblurring scheme base on the visual saliency was presented. Firstly, the image was divided into a significant part and a nonsignificant part through its saliency map, the blur kernel of the non-significant regional was estimated with an improved fast motion deblurring algorithm. Then, a compensation mechanism was proposed to convert the variable space blur of the significant part into the invariable space blur, and the estimated blur kernel mentioned before was applied to deblur the significant part. At last, the image was restored using the deconvolution algorithm. Experimental results show that the scheme can reduce the running time of the image deblurring process with the uniform blur and maintain the deblurring quality on the motion object.

关 键 词：运动图像 视觉显著性 空间可变模糊 去模糊 补偿 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]