Image retrieval based on deep Tamura feature descriptor

Various levels of visual features have different effects in image retrieval, and deep features can express higher-level features or semantic information. Tamura texture feature belongs to the handcrafted feature, and it can represent texture features corresponding to human visual perception. However, exploiting deep Tamura texture within deep learning models and aggregating them into a more discriminative representation remains challenging. To address this problem, we proposed a novel image retrieval approach named deep Tamura feature descriptor (DTFD). The main highlights are as follows: (1) We exploit Tamura texture features within the deep feature maps to provide deep Tamura features. It can improve the discriminative power of deep features. (2) We propose a new spatial layout optimization method to provide the optimized deep Tamura feature maps. It can highlight the target objects and reduce the negative effects of background noise, thereby improving retrieval performance. (3) We combine the advantages of Tamura texture features and deep features to provide a more effective yet compact representation. Comparative experimental results on five famous benchmark datasets demonstrate that deep Tamura feature descriptor provides the satisfying results and outperforms some existing state-of-the-art methods.