Hashing-based Batik Image Retrieval using Progressive Multi-Stage Training

This paper presents an effective way to generate binary feature descriptor for batik image retrieval. In this approach, the deep convolutional networks performs the embedding process of Batik image into vector space by exploiting the contrastive learning framework. This embedded vector is then projected into the Hamming space. Then, the batik image retrieval task is performed under the hashing approach. Progressive multi stage training with scrambled image is adopted to maximize the learning capability and capacity of the network. This training enhances the embedding vector representation and distribution. It can produce a more meaningful feature descriptor for the Batik image retrieval task. As documented in the experimental Section, the proposed method is superior compared to the other schemes in the Batik image retrieval system.