拉普拉斯稀疏编码的图像检索算法  

Image Retrieval Algorithm Based on Laplacian Sparse Coding

作　　者：王瑞霞[1] 彭国华[1] 郑红婵[1] 

机构地区：[1]西北工业大学理学院,西安710129

出　　处：《计算机科学》2014年第8期278-280,285,共4页Computer Science

基　　金：国家自然科学基金(61070233)资助

摘　　要：由于稀疏编码中过完备的码本和独立的编码处理,在编码过程中图像块相似性信息被丢失。为了保留块与块之间的相似性信息,提出了拉普拉斯稀疏编码的图像检索算法。给定编码特征的初始稀疏码,计算拉普拉斯矩阵,将相似性合并到稀疏目标函数,结合特征符号搜索算法和黄金分割线搜索算法,逐个更新每个稀疏编码系数。实验表明,拉普拉斯稀疏编码增强了稀疏编码的鲁棒性,与SPM模型算法相比,拉普拉斯稀疏编码的图像检索算法有较高的准确率。Due to the overcomplete codebook and the independent coding processing, the similarity of the image is lost between block and block to be encoded. To preserve such similarity information, we proposed the image retrieval algo- rithm based on Laplacian sparse coding. Given initial sparse coding and calculating the Laplacian matrix, similarity pre- serving term was incorporated into the objective of sparse coding. We used the feature-sign search algorithm and the golden section line search algorithm to update one by one each coefficient of sparse coding. The experiments show that Laplacian sparse coding can enhance the robustness of sparse coding. Compared with the improved SPM model, the new image retrieval algorithm better improves the retrieval accuracy.

关 键 词：稀疏编码 图像检索 码本 拉普拉斯矩阵 相似度矩阵 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]