An index framework for large scale clothing image retrieval based on GMM-cluster tree

The current image retrieval researches are focused on the low-level features of clothing images, while the characters of the whole clothing dataset are ignored. The clothing images have many classes, styles and details, and the dataset is growing in an amazing speed, which brings great challenges to the traditional retrieval methods in accuracy and efficiency. To address this problem, a new index framework named GMM-cluster tree was designed, which could classify the clothing images and save them into corresponding tree branches according to their classes, styles and details through hierarchical clustering, so as to avoid the wrong clothing classification due to artificial designation of cluster numbers. The accuracy and efficiency of clothing image retrieval are tested respectively based on a small and a large dataset. The experiment results show that both accuracy and efficiency of the research can be improved through the automatic determination of cluster numbers and layer-by-layer classification of the GMM- clustering tree.