A fast image retrieval algorithm with automatically extracted discriminant features

摘要: Fisher's discriminant analysis is very powerful for classification but it does not perform well when the number of classes large samples in each class small. We propose to resolve this problem by dynamically grouping at different levels a tree. recast as regression so that (class labels output) and (numerical values are unified. The proposed HDR tree automatically forms clusters input space guided desired output, which produces spaces. These spaces organized coarse-to-fine structure A unified size-dependent negative-log-likelihood handle both under-sample situations (where cluster smaller than dimensionality space) over-sample where can reach near-optimal performance. For fast computation, has logarithmic retrieval time complexity. been tested with synthetic data, face image databases, publicly available data sets use manually selected features.