| 本文已被：浏览 3485次 下载 6001次 |
码上扫一扫！ |
|
|
| 视觉词汇的主成分线性编码方法 |
|
艾浩军1,2, 张敏1,2, 方禹1,2, 赵梦蕾1,2, 李泰舟1,2, 王红霞3
|
|
1.武汉大学 计算机学院, 湖北 武汉 430072;2.武汉大学 国家多媒体软件工程技术研究中心, 湖北 武汉 430072;3.武汉理工大学 计算机科学与技术学院, 湖北 武汉 430063
|
|
| 摘要: |
| 针对视觉物体分类中视觉词汇局部线性编码缺少显著性检验和共线性分析的问题，提出了主成分线性编码方法，选择与特征点具有最强线性相关性的K近邻视觉单词，采用主成分多元线性回归方法以解决视觉单词的共线性问题，从而减小编码系数的偏差和不稳定，提高视觉物体分类的精度.依据图像量化结果的稀疏性是影响分类精度的重要因素，进一步对主成分线性编码得到的量化结果做稀疏性分析并进行能量正则化处理，提高分类效率.实验结果表明，与已有方法相比，平均分类正确率提高了1%以上. |
| 关键词: 视觉词袋 共线性 主成分回归 特征点归并 能量正则化 |
| DOI： |
| 分类号: |
| 基金项目:国家科技支撑计划（2012BAH35B03） |
|
| Principal Component Linear Coding for Visual Words |
|
AI Hao-Jun1,2, ZHANG Min1,2, FANG Yu1,2, ZHAO Meng-Lei1,2, LI Tai-Zhou1,2, WANG Hong-Xia3
|
|
1.School of Computer Science, Wuhan University, Wuhan 430072, China;2.National Engineering Research Center for Multimedia Software, Wuhan University, Wuhan 430072, China;3.School of Computer Science and Technology, Wuhan University of Technology, Wuhan 430063, China
|
| Abstract: |
| By means of significant test and co-linearity analysis, this paper proposes principal component linear encoding which selects the K-nearest neighbor visual word with the strongest linear correlation. The multiple linear regression method based on principal component is used to solve weak and instable coding caused by the visual words' co-linearity problem, improving the accuracy of the visual object classification effectively. Recognizing that the scarcity of the image quantify plays an important roles in the classification accuracy, the study analyzes the scarcity of the quantitative results obtained by the principal component linear encoding and then processes it with energy regularization to improve the classification efficiency further. The experimental results demonstrate that this method increases the recognition rate average over 1% than existing algorithms. |
| Key words: bag of visual words co-linearity principal component regression feature points merging energy regularization |
