一种基于频带选择的视觉注意计算方法  

A Computational Approach for Visual Attention Based on Frequency Band Selection

作　　者：吕琦[1] 王斌[1,2] 张立明[1] 

机构地区：[1]复旦大学电子工程系,上海200433 [2]复旦大学电磁波信息科学教育部重点实验室,上海200433

出　　处：《复旦学报（自然科学版）》2014年第3期407-417,共11页Journal of Fudan University：Natural Science

基　　金：国家自然科学基金(61071134);高等学校博士学科点专项科研基金(20110071110018)资助项目

摘　　要：视觉注意力的研究对计算机视觉的发展有很大的帮助.然而,绝大多数的现有模型仅能处理某一类型的数据或只针对特定的应用.事实上,不论场景中物体的尺寸大小如何,人类都能够正确无误地察觉到.本文基于生物学在视网膜上非经典感受野的发现,从频域的角度出发,提出一种新的自底向上的计算模型.显著图可以通过模拟扩展的感受野的思想而得到,主要由两步组成:首先利用Gabor小波将输入图像分解成一系列特征图,它们对应于整个频率域的不同频带;其次,挑选出其中的一些作为最优频带以生成显著图,以此来模拟感受野的调节.实验结果表明,本文所提出的算法无论是对于心理学图像还是包含了任意尺寸目标的自然图像,在显著性的预测上都优于现有其他算法.除此之外,感受野模型以及Gabor小波的生物特性使得模型具有一定的鲁棒性.Visual attention models contribute a lot to the development of computer vision. However, most of attention models only deal with some kinds of images or aim at specific applications. In fact, human beings have the ability to correctly perceive attentive focuses of objects with arbitrary size within any image patterns or scenes. This paper proposes a new bottom-up computational model from the perspective of frequency domain based on the biological discovery of non-Classical Receptive Field in retina. A saliency map can be obtained according to the idea of Extended Classical Receptive Field （ECRF）. The model is composed of two major steps.. Firstly decompose the input image into a series of feature maps representing different frequency bands that cover the whole frequency domain from the extreme low to the very high by utilizing Gabor wavelet. Secondly, select some maps as optimal bands, simulating the modulation of receptive field, to generate the saliency map. Experimental results show that the proposed algorithm outperforms other state-of-the-art algorithms at human fixation prediction for both psychological patterns and natural images containing salient objects with arbitrary sizes. Beyond that, biological plausibility of ECRF and Gabor wavelet makes our approach more robust.

关 键 词：视觉注意 显著图 非经典感受野 GABOR小波 二维熵 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]