Object-based Video compression using neural networks

This paper presents a new object-based video compression approach. It consists on predicting video objects motions throughout the scene. Neural networks are used to carry out the prediction step. A multi-step- ahead prediction is performed to predict the video objects trajectories over the sequence. In order to reduce video data, only the background of the video sequence is transmitted with the different detected video objects as well as their initial properties such as placement and dimensions. Experimental results show the effectiveness of the proposed approach in terms of the compression rates.