Learning-Based Image Super-Resolution

Image Super-resolution predicts image's high-resolution part from its low-resolution one. It can be utilized for many computer vision tasks, such as face detection and recognition, low bit rate video transmission, image restoration, facial expression analysis and digital zooming. Learning-based image super-resolution is the most popular technique in the super-resolution area. By investigating the characteristic of learning-based super-resolution algorithm and the inherent property of face image, we propose a steerable-based face super-resolution algorithm. We explore the issues in several directions:Firstly, this paper introduces the basic concept, current status and existing problems of image super-resolution techniques. Based on the analysis of traditional interpolation algorithms and edge-adaptive image interpolation techniques, we implement a commercial system ImageZoom, which utilizes many interpolation algorithms to resize images obtained from digital camera.After doing research on the limits of reconstruction-based super-resolution, we find that the prior on the high-resolution images becomes more important as the factor by which the resolution is enhanced gets larger. Then we implement a system, which utilizes Markov network to learn the prior on the high-resolution images.Then, we propose a new learning-based super-resolution algorithm for face images. In the first step, steerable pyramid is used to capture low-level local features in face images, and then these features are combined with pyramid-like parent structure and locally best matching to predict the best prior, In the second step, the prior is integrated into Bayesian maximum a posteriori framework. Finally, steepest descent method is used to obtain the optimal high-resolution face image. The effectiveness of our approach is demonstrated by extensive experimental results with high-quality predicted face images.At last we discuss the potential research and application of learning-based super-resolution.