Machine learning and computer vision have found their way into various applications in our day-to-day lives. They are used in tasks ranging from navigation, multimedia to security among others. Convolutional Neural Networks showed a great increase in performance measure compared to feature extraction methods. In spite of their improving performances, they have been burdened with immense computational tasks.
In order to reduce the computation efforts from finding robust features, we develop approaches to video analysis in compressed-domain. Furthermore we investigate the use of statistical learning methods for video compression and evaluate several types of deep architectures for this task.