稀疏编码在动态纹理识别中的应用.doc

中图法分类号：TP391.4 文献标识码：A 文章编号：1006-8961(200 ) - - 论文引用格式： 基于稀疏编码的动态纹理识别 刘洋1，李一波2，姬晓飞2，王杨扬1 1. 南京航空航天大学,南京 210016； 2. 沈阳航空航天大学，沈阳 110136 摘 要：目的：线性动态系统有效地捕捉了动态纹理在时间和空间的转移信息。然而，线性动态系统属于非欧式空间模型，无法直接应用传统的稀疏编码进行分类识别，本文提出了一种基于稀疏编码线性动态系统的求解方法并应用于动态纹理识别。方法：本文提出的新算法是基于约束凸优化公式，将稀疏编码和控制论中相似性变换结合，优化学习模型参数，解决应用稀疏编码进行分类识别的问题，实现有效的动态纹理识别。结果：在公开的动态纹理图像数据库UCLA上进行实验并与其它方法进行比较，实验表明本文提出的算法对动态纹理及遮挡情况具有更好的识别率。结论：本文提出了一种新的动态纹理识别方法，线性动态系统描述动态纹理，稀疏编码进行分类识别。实验结果表明，本文提出的方法具有更好的性能，识别率可达97%，且对遮挡具有更好的鲁棒性。 关键词 ：稀疏编码；遮挡；动态纹理识别；相似性变换 Dynamic Texture Recognition Based on Sparse Coding Yang Liu1，Yibo Li2，Xiaofei Ji2，Yangyang Wang1 1. Nanjing University of Aeronautics and Astronautics, Nanjing 210016,China； 2. Shenyang Aerospace University，Shenyang 110136, China Abstract ： Objective: Linear Dynamical Systemescription for dynamic texture can capture the transition of appearance and motion effectively. However, the LDS model does not belong to Euclidean space, making it impossible to apply traditional sparse coding techniques for classification and recognition. A novel approach based on sparse coding and LDS is proposed to be applied in dynamic texture recognition. Method: The proposed algorithm employs a principled convex optimization formulation that allows both a sparse representation code and a linear transformation matrix to be jointly inferred. Model parameters are optimized and learned to realize good texture recognition. Result: Experiments are conducted on publicly available dynamic texture databases UCLA, and comparison with other methods is made. Results show that the proposed algorithm outperforms earlier approaches, including robustness to occlusion. Conclusion: A novel method for dynamic texture recognition is proposed in this paper. Linear dynamic system is used to describe the dynamic texture, and sparse coding is used as classifier. Experimental results show that the proposed method has better performance, for the recognition rate 97% and better robustness to occlus