Improved Deep Hashing with Scalable Interblock for Tourist Image Retrieval

Figure 1

The proposed deep hash retrieval framework consists of the subnetwork hash layer and the improved triplet loss function. subnetwork is a multilayer convolution neural network. This hash layer comprises the divide-encode layer and the concat layer, and the length of the concat layer represents the hash code length. The parameter transferring process transfers the pretrained parameters on Place-2 to the subnetwork as the initial parameter to train the hash mapping function.