Search

Scene text in natural images carries additional semantic information to aid in image classification. Existing methods lack full consideration of the deep understanding of the text and the visual text relationship, which results in the difficult to judge the semantic accuracy and the relevance of the visual text. This paper proposes image classification based on Cross modal Knowledge Learning of Scene Text (CKLST) method. CKLST consists of three stages: cross‐modal scene text recognition, text semantic enhancement, and visual‐text feature alignment. In the first stage, multi‐attention is used to extract features layer by layer, and a self‐mask‐based iterative correction strategy is utilized to improve the scene text recognition accuracy. In the second stage, knowledge features are extracted using external knowledge and are fused with text features to enhance text semantic information. In the third stage, CKLST realizes visual‐text feature alignment across attention mechanisms with a similarity matrix, thus the correlation between images and text can be captured to improve the accuracy of the image classification tasks. On Con‐Text dataset, Crowd Activity dataset, Drink Bottle dataset, and Synth Text dataset, CKLST can perform significantly better than other baselines on fine‐grained image classification, with improvements of 3.54%, 5.37%, 3.28%, and 2.81% over the best baseline in mAP, respectively. [ABSTRACT FROM AUTHOR]

In order to improve recognition accuracy of clothing style and fully exploit the advantages of deep learning in extracting deep semantic features from global to local features of clothing images, this paper utilizes the target detection technology and deep residual network (ResNet) to extract comprehensive clothing features, which aims at focusing on clothing itself in the process of feature extraction procedure. Based on that, we propose a multideep feature fusion algorithm for clothing image style recognition. First, we use the improved target detection model to extract the global area, main part, and part areas of clothing, which constitute the image, so as to weaken the influence of the background and other interference factors. Then, the three parts were inputted, respectively, to improve ResNet for feature extraction, which has been trained beforehand. The ResNet model is improved by optimizing the convolution layer in the residual block and adjusting the order of the batch-normalized layer and the activation layer. Finally, the multicategory fusion features were obtained by combining the overall features of the clothing image from the global area, the main part, to the part areas. The experimental results show that the proposed algorithm eliminates the influence of interference factors, makes the recognition process focus on clothing itself, greatly improves the accuracy of the clothing style recognition, and is better than the traditional deep residual network-based methods. [ABSTRACT FROM AUTHOR]