CAJ | 학술논문

传统的稀疏编码方法在遇到大规模数据时，因计算复杂度高而出现异常。针对这种异常导致不能很好地进行特征提取的问题，提出正则化双阶线性稀疏编码DLRSC（ Double Linear Regularization Sparse Coding）方法。借助于广义多特征子空间框架来学习噪声和异常像素的结构特征，通过使用L1球理论，计算出唯一的近似解，并且利用滤波技巧避免了大规模数据的复杂计算，从而降低了时间及空间复杂度。最后，在ORL及Yale两大通用人脸数据库上的实验验证了所提的DLRSC方法的有效性，实验结果表明，相比其他几种最先进的稀疏编码方法，所提方法取得了更好的识别效果。
전통적희소편마방법재우도대규모수거시，인계산복잡도고이출현이상。침대저충이상도치불능흔호지진행특정제취적문제，제출정칙화쌍계선성희소편마DLRSC（ Double Linear Regularization Sparse Coding）방법。차조우엄의다특정자공간광가래학습조성화이상상소적결구특정，통과사용L1구이론，계산출유일적근사해，병차이용려파기교피면료대규모수거적복잡계산，종이강저료시간급공간복잡도。최후，재ORL급Yale량대통용인검수거고상적실험험증료소제적DLRSC방법적유효성，실험결과표명，상비기타궤충최선진적희소편마방법，소제방법취득료경호적식별효과。
Anomaly will occur in traditional sparse coding method when large-scale data is encountered due to high computational complexity, and this leads to the feature extraction not working well.In light of this problem, we propose the dual-stage linear regularised sparse coding ( DLRSC) method, it learns the structure features of noise and abnormal pixel by means of generalised multi-feature subspace framework, and calculates the only approximate solution by using the theory of L1 ball.Moreover, the filtering skill is adopted to avoid the complex computation of the large-scale data, thereby reduces the time and space complexity.Finally, the experiments on two universal database, the ORL and the Yale, verify the effectiveness of the mentioned DLRSC method, experimental results show that compared with other state-of-the-art sparse coding methods, the proposed method achieves better recognition results.