Video Retrieval Based on Image Queries Using THOG for Augmented Reality Environments
Joolekha Bibi Joolee, Young-Koo Lee · 2018
Augmented reality includes the virtual object and real life together. On the other hand, video retrieval is inevitable to bring out similar videos from the large-scale video dataset. Database videos cover a temporal component, while the query images do not, so we need to design a retrieval system that compares the images to videos in augmented reality. In this paper, we propose a novel approach for video retrieval in augmented reality based on image queries. At first, we extract the key frames from the videos. Secondly, we employ a novel frame based feature extraction method, namely Ternary Histogram of Oriented Gradient (THOG). Thirdly, we utilize the Double-Bit Quantization based hashing to perform the nearest neighbor search efficiently, which is responsible to generate the candidate list of videos. Finally, the similarity measure is performed to re-rank the list of videos from the candidate list. Experimental result shows our proposed method can produce competitive accuracy.