LAUSR

Due to the fast growth of image data on the web, it is necessary to ensure the content security of uploaded images. One of the fundamental problems behind this need is retrieving relevant images from the large-scale databases. Recently, hashing/binary coding algorithms have proved to be effective for large-scale visual information retrieval. Most existing hashing methods usually seek single linear projections to map each sample into a binary vector. In this paper, a supervised deep hashing method is proposed, which seeks multiple non-linear transformations to generate more discriminative binary codes with short bits. We implement a deep Convolutional Neural Network to achieve end-to-end hashing. A loss function is elaborately devised to preserve the similarity relationship between images, meanwhile minimize the quantization error and make hash bits distribute evenly. Extensive experimental comparisons with state-of-the-art hashing algorithms are conducted on CIFAR-10 and NUS-WIDE, the MAP reaches to 87.67% and 77.48% with 48 bits respectively. It shows that the proposed method achieves very competitive results with the state-of-the-arts.