基于视觉标准模型和复数小波的自然图像识别

摘要

根据双树复数小波具有近似平移不变性和良好方向选择性的特点,提出了一种基于双树复数小波的目标识别方法,获得了有限平移不变和尺度不变的特征,解决了Serre视觉标准模型中Gabor小波计算复杂的问题.用Caltech101图像库的图像进行测试,获得了良好的识别效果.结合Walther提出的自上而下注意机制,在保证较高识别率的同时大大提高了运算速度.实验结果表明,该方法在识别准确率上可与Serre标准模型相比,且运算速度明显提高.%This paper proposed an algorithm based on the dual-tree complex wavelet, which had approximate shift invariance and the good directional selectivity.The algorithm here could extract features with shift-invariance and seale-invariance and could solve the problem that computed complicatedly in visual standard model proposed by Serre.When tested on Caltechl0l database,the algorithm could achieve high recognition rate.Combined with up-bottom attention mechanism, the method could keep the recognition rate, and decreased the computation cost greatly.Experimental results show that the proposed algorithm can comparable to standard model but at a lower computation cost.