CAJ | 학술논문

在以稀疏编码思想为基础的人脸识别问题中，稀疏表征分类（ SRC）与协同表征分类（ CRC）近年来受到广泛关注。这2种模型分别代表了以l1和l22种范数作为约束项的最优化问题，两者在不同条件下各具优势。基于稀疏表示的人脸识别问题可归结为统计学中线性回归模型的实际应用，从线性回归的角度出发，结合Elastic Net回归模型思想，提出了一种基于稀疏与协同联合表征的人脸识别模型（ S＿CRC）。该模型是SRC和CRC的凸结合，这种结合方式使得所得到的线性表示系数同时受到l1和l22种范数的约束，具有更强的判别性，从而更加有利于分类。最后，通过在AR和Extended Yale B人脸库中的实验，论证了该方法的有效性。
재이희소편마사상위기출적인검식별문제중，희소표정분류（ SRC）여협동표정분류（ CRC）근년래수도엄범관주。저2충모형분별대표료이l1화l22충범수작위약속항적최우화문제，량자재불동조건하각구우세。기우희소표시적인검식별문제가귀결위통계학중선성회귀모형적실제응용，종선성회귀적각도출발，결합Elastic Net회귀모형사상，제출료일충기우희소여협동연합표정적인검식별모형（ S＿CRC）。해모형시SRC화CRC적철결합，저충결합방식사득소득도적선성표시계수동시수도l1화l22충범수적약속，구유경강적판별성，종이경가유리우분류。최후，통과재AR화Extended Yale B인검고중적실험，론증료해방법적유효성。
In the sparse coding based face recognition problem, sparse representation based classification (SRC) and collaborative representation based classification (CRC) had been successfully applied in recent years.These two models respectively represent the optimization problems of l1 or l2 norm characterization of coding coefficient.Both of them had their own advantages in different conditions.The sparse representation based face recognition problem could be attributed to the application of linear regression problems in statistics. Based on analysis from the respective of linear regression, combining with the thinking of Elastic Net model, it was proposed a method named Sparse joint Collaborative Representation Based Classification ( S_CRC) , which was a convex combination of SRC and CRC.This kind of combination made the linear representation coeffi-cient contain more discriminating information thus more conducive to classification.Finally, the feasibility of the proposed method was verified on AR and Extended Yale B face database.