摘要： 多聚焦图像存在聚焦区和离焦区,聚焦区通常吸引人的注意力,具有突出的视觉显著性。传统融合算法缺乏对聚焦区域的定位能力,对多聚焦图像融合的适应性普遍较差。为此,提出一种模拟人类视觉注意机制的多聚焦图像融合方法。利用谱残差算法计算源图像的显著度图,通过判断不同源图像相同位置上的像素显著性,选择显著度大的图像像素组成该源图像的聚焦区,显著度相等的像素构成边界带,使用腐蚀膨胀操作消除聚焦区内的孤立像素点,以每幅源图像的聚焦区域和梯度值较大的边界带像素作为融合图像的像素。实验结果表明,该方法能自主选择清晰像素,获得37 dB以上的高峰值信噪比,且基本无参数设置,在不同类型图像融合中均表现出较强的鲁棒性。

关键词: 视觉注意, 多聚焦图像, 图像融合, 显著度, 聚焦区, 显著图

Abstract: Multi-focus image is divided into focus area and defocus area.Focus area usually attracts attention with outstanding visual salience.Traditional fusion algorithms can not locate the focus areas automatically,resulting in low adaptability to multu-focus image fusion.This paper presents a fusion method for multi-focus image by simulating visual attention mechanism of human.The saliency maps of the source images can be calculated by using the spectrum residual algorithm.By which the saliency degree of the same pixel position in different images are determined,and selecting pixels with lager saliency degree as the focus area of the original image,the pixels with equal saliency degree as the edge area,eliminating the isolated pixel in the focus area by eroding and dilating method.The fusion image consists of focus areas of original images and pixels with lager gradient degree in the edge area.Experimental results show that the proposed method can choose the clearest pixels from the focus area,obtains over 37 dB Peak Signal to Noise Ratio(PSNR) and needs no parameter settings.It also exhibits strong robustness in different type of images.

Key words: visual attention, multi-focus image, image fusion, saliency degree, focus area, saliency map

中图分类号: 

• TP391