压缩感知稀疏识别用于多视角图像目标分类

Compressed Sensing Sparse Recognition for Target Classification from Multi-view Images

针对多视角条件下的图像目标分类问题,提出一种基于压缩感知特征的稀疏识别方法.该方法以原始图像的感知数据为特征描述,将测试样本与训练样本集的压缩感知特征纳入稀疏识别的框架,并通过求解一个l_1范数优化问题来获取分类结果.实验表明,该方法不仅有效利用了压缩感知特征的信息冗余性来保证稀疏识别的性能,而且无需进行预处理就能较好地实现多视角图像的目标分类.

Multi-view image target classification is usually difficult. To deal with the problem, we propose a sparse recognition （SR） method with compressed sensing （CS） features. Sensing data of the original image are used as corresponding features. Both the test sample and the training sample set are integrated into an SR framework with their CS features. Classification results can be obtained by solving an l1-norm optimization problem. Experiments show that excellent performance of SR can be obtained by using CS features that retain information redundancy of the original sample. Meanwhile, multi-view image target classification is robust without preprocessing.