基于心理学的图像多分辨率统计特征提取

Extracting image features based on psychologic evidence

从采集的图像数据中提取特征,这是图像的模式识别中最重要的问题之一,解决的方法通常依赖于对具体应用的经验知识,对同样的任务,人的视觉系统却以一种普适的方法去完成。本文从人的心理学特性出发,导出信号特征的全局性质,然后讨论由局域信号构成总体特征的统计方法,进而借助于信号的Lebesgue积分表示与测不准原理提出一种普适性的特征表示方法MSR。实验结果表明,对图像信号MSR是适合于并行计算的有效逼近。

Extracting the features from the image data is one of the most important issues in image pattern recognition, the approaches dealing with the issue usually depends on how well one know of a specific object; but a man with his vision system deals with the same issue in a more general or universal way. In this paper, the global property based on the psychologic evidence of vision system is discussed at first and then the statistics method that make local original signal into global feature, afterwards to represent the feature a universal scheme MSR (multiresolution statistics representation) is given by Lebesgue integrate of signal and uncertainty principle, finally the experimental evidences support that MSR is an efficient approximation and also suitable for parallel calculation..