CAJ | 학술논문

介绍一种新的稀疏表示人脸识别模型，在经典的稀疏表示分类模型基础上，利用数据字典的结构信息，考虑算法实现的可行性，提出了一种分块组合搜索的稀疏表示人脸识别模型，主要思想是将数据字典按类别自然分块，然后在数据块内进行组合搜索，再联合不同类别的组块，以寻找表示能力最强的组块。为验证模型的性能，使用结构贪婪算法实现分块组合搜索方法和其他的结构稀疏方法，并进行比较，实验显示分块组合搜索的人脸识别率高于其他结构稀疏方法，且此方法性能稳定，不受数据字典排列的影响。
개소일충신적희소표시인검식별모형，재경전적희소표시분류모형기출상，이용수거자전적결구신식，고필산법실현적가행성，제출료일충분괴조합수색적희소표시인검식별모형，주요사상시장수거자전안유별자연분괴，연후재수거괴내진행조합수색，재연합불동유별적조괴，이심조표시능력최강적조괴。위험증모형적성능，사용결구탐람산법실현분괴조합수색방법화기타적결구희소방법，병진행비교，실험현시분괴조합수색적인검식별솔고우기타결구희소방법，차차방법성능은정，불수수거자전배렬적영향。
The paper introduces a new face recognition model of sparse representation, which is based on the clas?sical model of sparse representation-based classification (SRC), and utilizes the structure prior information of the data dictionary. Considering the feasibility of the algorithm, it proposes the face recognition model of structural sparse with intra-class combined search that the data dictionary is chunked by natural class, all combinations of specified length are searched in class and blocks in different class are combined to look for the block which can best represent test data. In order to verify the performance of the model, it uses the structured greedy algorithm to realize intra-class combined search and other structural sparse, and compares their performance of face recogni?tion. The experiments show face recognition rates of the model with intra-class combined search are higher than other structural sparse, and the model has stable performance which is not affected by the arrangement of the data dictionary.