基于多层判别式字典学习算法的传统服饰图像分类

Traditional Clothing Image Classification Algorithm Based on Multi-Layer Discriminant Dictionary Learning

现有的多层判别式字典学习算法中大多采用交替方向乘子法实现字典的更新,在图像分类方面的应用较为成熟。然而,当图像内容较为丰富且含有多个标签时,现有方法在多标签分类上的表现不佳。对此,可采用递归最小二乘法与去相关增强重建系数算法构成的二层判别式字典学习结构,更适用于图像多标签分类。通过多层判别式字典学习对数据进行多次稀疏分解,在最后一层判别式中用线性分类器对稀疏分解得到的特征向量进行分类,采用4个多标签分类指标对分类效果进行评判时,发现One-error,Coverage,Ranking-loss三个分类指标越小,Average-precision分类指标越大,算法的性能越优。实验结果表明,在明清服饰纹样数据集上使用多层字典学习算法的分类精度达到了82.17%,在同类算法中的性能最优。

Multi-layer discriminant dictionary learning has achieved remarkable results in the field of image classification.However,the existing multi-layer discriminant dictionary learning algorithms mostly use the alternating direction multiplier method to update the dictionary.When the image content is rich and contains multiple tags,they perform poorly in multi-tag classification.The two-layer discriminant dictionary learning structure composed of the recursive least square method and decorrelation enhancement reconstruction coefficient algorithm is more suitable for image multi-label classification.The data is sparsely decomposed many times through multi-layer discriminant dictionary learning,and the feature vectors obtained by sparse decomposition are classified by the linear classifier in the last layer.The experimental results on the dress pattern data set of the Ming and Qing dynasties verify the superiority of this algorithm.Compared with the latest existing algorithm,the classification accuracy of the proposed algorithm reaches 82.17%,which achieves the best effect in similar algorithms.