CBIR Of Specific Object Based On Multi-feature Fusion And Re-ranking

Content based image retrieval(CBIR) has become a hot topic in the field of computer science these years. There is a dramatic growth in the number of publications on the related problems, while various applications boom up in fields of medical science and online sales.In this thesis, we investigate multiple features suitable for content based image retrieval for specific objects, and propose a retrieval system based on multi-feature fusion and re-ranking in the case of human face images. The system extracts visual features with diversified information of color, texture and shape; selects the most discriminative subset by methods of cluster and information gain; and get the images of most similarity in appearance and expression by feature fusion and re-ranking.Specific type of object exhibits slight difference between images. In this thesis, we investigate the effect of several features on expressing these subtle differences, and offer a method of fusing them together. The growth in feature dimension will lead to the degeneration of the retrieval efficiency. To this end, we implement a fusion and re-ranking method taking both efficiency and accuracy into consideration.A large number of experiments based on a dataset collected on the Internet demonstrate the good performance of our method in Mean Average Precision(mAP).