摘要
稀疏表示分类算法在有监督的图像识别上有广泛的应用.该分类算法的准确度与训练样本个数有很大的关联.通常训练样本越充分,则该算法分类准确率越高,然而遇到小样本问题时,该算法分类准确率会明显降低.针对小样本问题,提出使用基于图像边缘位移的方法,得到和原始训练图像样本高度相关的新样本,达到扩充训练样本容量的目的,进而提高算法的分类准确率.同时,对于带仿射约束的稀疏表示分类算法,也可以经过图像边缘位移方法来提高分类准确率.实验结果证明,所用方法能够取得较好的图像识别效果.
Sparse Representation-based Classifier(SRC) is widely used in supervised image recognition. However, the accuracy of SRC is tightly associated with the number of training samples. The more the training samples are,the more accurate the SRC is. However, when small sample problems occur,the accuracy of SRC decreases. As for the small sample problems, the paper puts forward a method based on image edge displacement to obtain new samples which are highly associated with the original training samples in order to expand the training sample capacity and therefore to increase the classification accuracy of SRC. Meanwhile, as for the SRC with affine constraints, the classification accuracy of SRC can be further improved by image edge displacement. The expe 'nments show that the above-mentioned method can help achieve much more accurate image recognition.
作者
廖亮
杨程凯
LIAO Liang YANG Chengkai(School of Electric and Information Engineering, Zhongyuan University of Technology, Zhengzhou 451191, China)
出处
《成都大学学报(自然科学版)》
2016年第4期355-357,共3页
Journal of Chengdu University(Natural Science Edition)
关键词
图像分类
稀疏表示
训练样本
仿射约束
image classification
sparse representation
training samples
affine constraints