Robust Hashing for Character Authentication and Retrieval Using Deep Features and Iterative Quantization

This paper proposes a hashing approach for character authentication and retrieval based on the combination of a convolutional neural network (CNN) and the iterative quantization (ITQ) algorithm. This hashing approach is made up of two steps: feature extraction and hash construction. The feature extraction step involves the reduction of high-dimensional data into low-dimensional discriminative features by applying a CNN model. While, the hash construction step quantizes continuous real valued features into discrete binary codes by applying ITQ. These two steps are combined together in this work to achieve two objectives: (i) a hash should have a good anti-collision (discriminative) capability for distinct characters. (ii) a hash should also be quite robust to the common image content-preserving operations. Experiments were conducted in order to analyze and identify the most proper parameters to achieve higher authentication and retrieval performances. The experimental results are performed on two public character datasets including MNIST and Font-Char74K. The results show that the proposed approach builds hashes quite discriminative for distinct characters, and is also quite robust to the common image content-preserving operations.