视觉注意机制在大视场目标快速定位中的应用

Fast approach to object locating in large field of view based on vi-sual attention mechanism and log_polar transform

视觉心理学研究表明人类在看一个场景时,往往会在很短时间内找到几个显著区,然后再细看显著区域的内容,这样可以使得人类可以快速分析复杂图像。算法首先模拟人类视觉系统特点,根据图像的底层信息如对比度、方向、亮度等提取图像中几个最需要关注的显著区域,然后按照显著性由强到弱的顺序分别在每个显著区域利用具有尺度旋转不变性的对数极坐标变换方法进行目标的匹配定位。该方法在没有牺牲定位准确度的前提下,大幅减小了运算复杂度。实验表明该算法定位速度快而且准确。

Biological and visual perception studies show that visual attention mechanism enables human to select several salient regions likely to contain interesting objects in the scene in very short time and then analyze the image very quickly. The algorithm in this paper firstly imitates the characters of the human vision system and uses the bottom-up visual attention algorithm to select several salient locations of the large field image based on the image＇s low-level features such as contrast, orientation,color, texture and intensity.And then the locating of the object based on the scale and rotation invariant log-polar transform is carried out at the salient regions in order of decreasing saliency.The proposed algorithm can be used to quickly detect the object with large scale and rotation variation in large field image.It also can dramatically decrease the computation complexity without the lost of the locating precision.The experimental results show the algorithm is rather fast and precise.