Block-Based Feature Adaptive Compressive Sensing for Video

This paper focuses on the problem of feature adaptive reconstruction of Compressive Sensing (CS) captured video. In CS, sparse signals can be recovered with high probability of success from very few random samples. Utilizing the temporal correlations between video frames, it is possible to exploit improved CS reconstruction algorithms. Features that relate to the changes between frames are one of the options to benefit reconstruction. However, to choose the optimal feature for every particular region in each frame is difficult, as the true images are unknown in a CS framework. In this paper, we propose two systems for block-based feature adaptive CS video reconstruction, i.e., a Cross Validation (CV) based system and a classification based system. The CV based system achieves the selection of the optimal feature by applying the techniques of CV to the results of extra reconstructions and the classification based system reduces complexity by classifying the CS samples directly, where the optimal feature for the particular class is employed for the reconstruction. Simulations demonstrate that both of our systems work appropriately and their performance is better than uniformly using any single feature for the whole video reconstruction.This work is supported by EPSRC Research Grant (EP/K033700/1); the Natural Science Foundation of China (61401018); Beijing Jiaotong University; the Fundamental Research Funds for the Central Universities (2014JBM149).This is the author accepted manuscript. The final version is available from IEEE via http://dx.doi.org/10.1109/CIT/IUCC/DASC/PICOM.2015.25