Automatic Image Annotation using Conceptual Hierarchical Convolutional Neural Network

This paper explores the application of visual object detection by conceptual hierarchical convolutional neural network in automatic image annotation. To solve the image annotation due to the similarity of the object features cannot be classified perfectly, hypernyms tree with the semantic concept is used to enrich the recall rate. The development of CNN-based with conceptual object detect techniques provides another source about ISA relation with complementary concepts to ensure that the annotations are correct. The research method is divided into the two parts. The first part mainly introduces the multimodal learning neural network, which consist of visual model and hypernyms tree model. The visual model mainly carries out image content annotation using the R-CNN. We classify according to the features of the objects and propose a hierarchical labeling method. In order to make the system more effective in learning the image content, the system can detect the objects and understand the relationship between the objects. PASCAL VOC dataset is used to evaluate the proposed approach. According to the observation about the experimental results, the proposed approach outperforms traditional R-CNN and DPM approaches.