Optimized Discriminative Features for CBIR

In this paper,a method merging 2-class linear discriminant analysis is proposed to capture low-dimensional optimal discriminative features to improve the searching speed and precision of content-base image retrieval systems.First,a multi-class problem is translated to multiple 2-class problems with linear discriminant analysis to estimate a discriminant vector for each.Second,all the discriminant vectors are merged into a discriminant transformation matrix,by which image visual features are transformed into discriminant features.Finally,the discriminant features are employed to gain high precision of image retrieval and classification.The dimensionality of the discriminant features corresponds to the number of classes involved.The experiments,in which our proposed method is compared with various feature optimizing methods,show that the proposed approach improves the performance of image retrieval and classification dramatically.