仿射不变的多尺度自卷积熵提取方法

Affine Invariant Feature Extraction Based on Multi-scale Auto-convolution Entropy

摘要: 为从在不同视角获取的同一场景图像中提取更加独特的不变特征,提出一种图像仿射不变特征的提取方法.首先基于多尺度自卷积变换(MSA)构造了一组新的变换量——多尺度自卷积熵(MSAE);并证明了该熵具有仿射不变性;最后利用最小距离分类器分别对视点变换图像,以及加噪声、加部分遮挡视点变换图像进行分类识别实验.实验结果表明,MSAE特征能够获得更高的正确识别率.

Abstract: A novel distinctive feature,called multi-scale auto-convolution entropy (MSAE),is derived based on multi-scale auto-convolution,and it is proved to be affine invariant. The MSAE is used for classification using the minimum distance classifier. The images with changing viewpoint corrupted with Gaussian noise,and with occlusion were tested,and a higher recognition accuracy is achieved.