基于属性形态学分析的图像显著极值检测及应用

Salient Image Extrema Detection and Applications Applying Attribute Based Morphological Analysis

由于图像噪声的存在,使得利用传统的极值检测算法通常会使要提取的显著极值淹没在大量的噪声极值中;同时由于先验知识的缺乏,采用普通滤波技术也往往不能很好的滤除噪声,反而会破坏图像的关键结构。本文提出了一种基于属性形态学分析的图像显著极值检测算法。该算法可以在不需要对图像进行滤波的前提下,从数学形态学的角度对图像极值的显著性进行计算和评估,从而能够较好地提取出显著的极值。在沉浸模拟算法的基础上,给出了基于属性形态学分析显著极值检测的快速算法实现,并将其成功应用在视觉注意选择和独立运动目标检测上。实践证明,该算法不仅具有较好的抗噪声特性,而且快速实用,具有广泛的应用价值。

Salient image extrema detection has been a critical problem due to the importance of extrema in the description of image significant structures. However, due to image noise, traditional extrema detection methods like non-maximum suppression cannot give ideal results. Filtering techniques are commonly used to reduce the negative effect of image noise, but can hardly produce satisfied results because the filter sizes cannot be properly determined beforehand. Accordingly, in this paper, a new salient image extrema detection algorithm is proposed by applying attribute based morphological analysis. During detection, the image extrema saliency is analyzed and valuated by its morphological attributes, and the salient extrema is effectively detected without the need of noise filtering. An efficient algorithm implementation is also provided based on flooding simulations and successfully applied to both cases of visual attention selection and independent moving target detection. It is proved that the algorithm is not only noise-resistant in salient extrema detection, but also applicable to a wide range of applications.