Semi-Supervised Orthogonal Discriminant Analysis with Biased Discriminant Euclidean Algorithm

For retrieving images from the database, content based image retrieval is a widely used technique. In order to improve the performance of content based image retrieval, relevance feedback has been widely used. Here in this method we proposed a technique called biased discriminant Euclidean algorithm, in which it considers both intraclass geometry and interclass discrimination. Labeled and unlabeled samples can be used in order to get the relevant images. To enhance the relevance feedback performance for the unlabelled samples we integrate a concept called Semi-Supervised Orthogonal Discriminant Analysis via Label Propagation with semi-supervised biased discriminant Euclidean embedding (SBDEE). It propagates label information from labeled data to unlabeled data according to the distribution of labeled and unlabeled data. Thus the distribution of the unlabeled data can be effectively explored to learn a better subspace. The scatter matrices based on soft label learned by label propagation are defined to perform the discriminant analysis, which gives us a general framework to extend many variants of supervised discriminant analysis to the semi-supervised ones.