具有Fisher判据约束的非负稀疏编码模型

Non-negative Sparse Coding Model with Fisher Discriminant Ceritior Constraint

在标准非负稀疏编码(NNSC)的基础上,引入Fisher线性判据约束,提出一种改进NNSC模型。该模型能够提高稀疏系数的空间可分性和特征分类能力。通过测试掌纹自然图像可知,提取的图像特征具有方向性、空间性和选择性,利用掌纹特征基可实现图像重构,采用距离分类器可得到较好的识别效果。仿真结果验证了该模型在可视神经元建模、图像特征提取和模式分类中的有效性。

On the basis of the standard Non-negative Sparse Coding（NNSC）, the Fisher Linear Discriminant（FLD） constraint is introduced and a new modified NNSC model is proposed. This model can promote the spatial separability of sparse coefficients and benefits to enforce the classification capability. Using palmprint images to test, the image features extracted by this model have orientation, spatiality and selectivity. The image reconstruction work can be implemented successfully by using these features extracted. Utilizing the distance classifier to test image features, test results show that the model can improve the feature recognition efficiency. Simulation results show that the model is efficient in the visual neuron modeling, image feature extraction and pattern classification.