具有判别能力的低秩投影字典对学习

Discriminative Low-Rank Projection Dictionary Pair Learning

与传统字典学习算法相比,新的投影字典对学习(DPL)算法在字典学习过程中引入了投影字典,利用投影编码替代目标样本在字典上的稀疏编码,有效降低了模式识别算法的计算量,但是原始的DPL算法对遮挡和噪声干扰较为敏感。为了解决这一问题,提出了一种具有判别能力的低秩投影字典对学习(DLPL)算法,该算法在模型中增加了对字典的低秩约束并利用最小二乘估计法对标签样本投影编码的分类误差进行约束,待求字典和投影字典都具有封闭形式的解,通过交替优化方法进行快速求解。不同数据库中的实验结果表明,DLPL算法不仅可以改善字典在遮挡和噪声干扰下的性能,提高模式识别正确率,而且可有效缩短模型的训练时间和测试时间。

Compared with the conventional dictionary learning algorithm, the novel projection dictionary pair learning （DPL） algorithm introduces a projection dictionary in the dictionary learning process and utilizes the projection coding to replace the sparse coding in the dictionary for object samples. The computational cost of the pattern recognition algorithm can be effectively reduced. However, the original DPL algorithm is sensitive to occlusion and noise interference. To solve this problem, a discriminative low-rank projection dictionary pair learning （DLPL） algorithm is proposed. A low-rank constraint is added to the dictionary in the model and the least squares estimation method is used to constrain the classification error of the projection coding for the labeled samples. The unknown dictionary and the projection dictionary with closed solutions can be solved quickly by the alternative optimization method. Experimental results in different databases show that the DLPL algorithm can not only improve the performance of dictionary under occlusion and noise interference and raise the pattern recognition rate, but also effectively reduce the training time and the test time for the model.