基于L_0稀疏约束的近似稀疏解人耳识别

Human ear recognition approximating sparse solution based on L_0 sparse constraint

传统的人耳识别算法在人耳图像遮挡、噪声和人耳多姿态变化中表现出低识别率,近年来稀疏表示在模式识别领域中取得了很好的成果。决定稀疏分类器识别精确度的因素主要是稀疏解的稀疏度,而稀疏度的估计就是稀疏向量中非0元素的估计,即向量L_0范数。因此在人耳稀疏分类算法的研究中引入L_0范数稀疏约束。采取基于SRC(sparse representation-based classification)稀疏模型,选取对人耳姿态变化具有强鲁棒性的特征逼近过完备字典,然后使用OMP(orthogonal matching pursuit)算法直接解L_0问题,并加入稀疏约束,从优化稀疏解的角度对人耳稀疏分类算法进行改进,提高了人耳识别效率。

The traditional human ear recognition algorithm showed low recognition rate on ear image block, noise and multi- profile ear. In recent years, sparse representation have made great achievements in the field of pattern recognition. However, sparse degree of solution is the main factors to decide sparse classifier recognition accuracy. Sparse estimation was the estima- tion of the non-zero elements in the sparse vector, namely vector L0 norm. Therefore, this paper introduced L0 norm sparse con- straint into the study of human ear sparse classification algorithm. In summary, this paper took the characteristic of the human ear which had a strong robust posture change of approach over complete dictionary, then used OMP algorithm for L0 questions and joined the sparse constraint from the perspective of the optimized sparse to improve the human ear sparse classification algorithm. It improves the ear recognition efficiency.