A new algorithm for compressing massive region-of-interest location information in videos

Region-of-Interest (ROI) location information in videos is of increasing importance in many applications including user interest analysis and user experience improvement. Although ROI-based coding has been studied by many researchers, most of the works only focus on using the ROI information for improving coding efficiency while the ROI location information itself is seldom coded or transmitted. Though some methods can recover the ROI locations at the decoder by some parameters such as the quantization parameter, they will fail to work when the number of ROIs becomes large or the ROIs become overlapping. In this paper, we propose a new algorithm to compress this massive ROI location information in videos. The proposed algorithm introduces the region position information extracted from the reconstructed frame as the reference to reduce the ROI location data. Furthermore, the temporal correlations among ROIs in neighboring frames are also utilized for compressing the ROI locations. By suitably integrating the extracted region position as well as the temporal correlation, the proposed algorithm can reduce the data by about 20% for videos with 30 ROI regions. Experimental results demonstrate the effectiveness of the proposed algorithm.