计算机与现代化

• 图像处理 • 上一篇    下一篇

基于分布结构约束稀疏表示的图像分类方法

  

  1. 陕西交通职业技术学院经济管理系,陕西西安710018
  • 收稿日期:2015-04-07 出版日期:2015-07-23 发布日期:2015-07-28
  • 作者简介:范引娣(1979-),女,陕西礼泉人,陕西交通职业技术学院经济管理系讲师,硕士,研究方向:智能图像处理。

Image Classification Method Based on Distribution Structure Constrain Sparse Representation

  1. Dept. of Economic Management, Shaanxi College of Communication Technology, Xi’an 710018, China
  • Received:2015-04-07 Online:2015-07-23 Published:2015-07-28

摘要:

为了解决稀疏表示结构信息缺失的问题,从而更加准确地进行图像分类,本文提出一种新的基于结构约束的稀疏表示的图像分类方法。在对图像进行降采样的前提下,提取方向梯度直方图特征后
的训练样本上构建稀疏线性编码模型,通过样本间的分布结构信息约束和1范数最优化求解测试样本的稀疏系数x,利用稀疏系数均值法进行目标的分类识别。基于COREL图像库进行仿真验证,实验证明
,基于结构约束稀疏表示的图像分类方法能够获得很好的识别性能,与非结构约束稀疏表示相比本文方法显著提高了图像分类的准确率。

关键词:  , 结构约束, 稀疏表示, 图像分类

Abstract:

To solve the structure information loss issue on sparse representation for accurate image classification, a new method based on structure constrain sparse
representation was proposed. The training samples after downsampling and extracting histogram of orientated gradient (Hog) were utilized to construct sparse linear coding model.
The sparse coefficients were solved on the training samples by distribution structure information constrain and 1minimization, and image was classified by sparse coefficient
mean.  Experimental results with COREL dataset demonstrated that the proposed method can obtain the good recognition performance. Comparing with nonstructure constrain sparse
representation, the proposed method greatly improves the accuracy of image classification.

Key words: distribution structure constrain, sparse representation, image classification

中图分类号: