Deep Unsupervised Image Hashing by Maximizing Bit Entropy

Unsupervised hashing is important for indexing huge image or video collections without having expensive annotations available. Hashing aims to learn short binary codes compact storage and efficient semantic retrieval. We propose an unsupervised deep layer called Bi-Half Net that maximizes entropy of the codes. Entropy maximal when both possible values bit are uniformly (half-half) distributed. To maximize entropy, we do not add a term loss function as this difficult optimize tune. Instead, design new parameter-free network explicitly force continuous features approximate optimal half-half distribution. This shown minimize penalized Wasserstein distance between learned Experimental results on datasets FLICKR25K, NUS-WIDE, CIFAR-10, MS COCO, MNIST UCF-101 HMDB-51 show our approach leads compares favorably current state-of-the-art.