Abstract
Existing hash methods only compare the hash codes by the similarity of hash codes, which cannot fully express the relationship of between the hash codes corresponding to images. This work improves the quality of image retrieval by end-to-end representation learning. We propose a compact hash code learning method named DCCH (Deep Continuous Center Hashing), which learns hash representation with a well-specified loss function which adopts label information and spatial information. We train deep network model with proposed loss function to improve image hashing and video hashing. Experiments on image hashing and video hashing demonstrate that proposed DCCH method yields higher mean average precision than state-of-the-art hashing approaches.