自然图像中感兴趣目标检测新技术.doc

自然图像中感兴趣目标检测新技术 XXX1,2，XXX1，XXX1,2 （1 上海大学机电工程与自动化学院，上海 200072； 2 上海电力学院电子科学与技术系，上海 200072） 摘要：针对基于显著图目标检测方法不能象人类视觉一样精确地得到感兴趣目标的位置以及同一感兴趣目标上检测出多个感兴趣区域的问题，本文提出一种视觉注意机制和模糊SVM相结合的算法。首先根据显著图和角点分布从图像中获得包括单个目标的视觉窗口，并在窗口中采用FSVM算法分割目标和背景。对多目标的图像可根据显著度分布信息自动选择其他目标，解决视觉转移过程中同一目标内出现多个感兴趣区域的问题。实验结果表明该方法符合生物的视觉注意机制，分割效果令人满意。 关键词：感兴趣目标；显著图；模糊支持向量机；视觉注意; 特征提取 中图法分类号：TP391.4 文献标识码：A Novel Technology of Detecting Regions of Interest in a Natural Image XXX1,2，XXX1，XXX1,2 (1. School of Mechatronics Engineering and Automation, Shanghai University, Shanghai 2. Dept. of Electronic Science and Technology, Shanghai University of Electric Power, Shanghai, 200090, China） Abstract: In general, saliency map based the region of interesting(ROI) detection often has the problems of not able to locate object of interesting accurately as human visual attention should be and that many interesting object can be detected on the same ROI. A new technique for detecting regions of interest in a natural image by using visual attention model and Fuzzy SVM（FSVM）is proposed in this paper. Firstly a visual window including single object is created according to the visual attention and edge information based on distribution of corner points from an image, by using FSVM based on affinity among samples ,we can extract single object in the visual window. Another region can be estimated and selected automatically according to the distributing information of saliency value. Experimental results coincide with human visual attention mechanism and demonstrate the effectiveness of the proposed approach. Keywords: object-of-Interest; saliency map ;fuzzy SVM ;visual attention ; feature extraction 1概述 传统的感兴趣区域检测算法，主要是通过用户人工指定或用图像分割算法估计图像中的感兴趣对象或区域。人工指定感兴趣区存在主观性问题，图像分割算法得到的感兴趣区从本质上说，是通过图像低层视觉特征对用户实际兴趣的一种估计，无法完成区域兴趣的度量，也难以实现对感兴趣区的客观描述。Koch 等人[1]提出一个基于视觉生理和心理物理实验结果，模拟生物体视觉注意机制的选择性注意模型，得到图像中最容易引起注意部分的方法，其显著性的大小一般用一幅灰度图像（saliency map）来表示。显著值越大对应的部分用户关注度越高，但仅仅基于这种显著图的区域检测方法通常不能象人类视觉一样精确地得