COMBINED ITERATIVE BACK PROJECTED-MAXIMUM A POSTERIORI TECHNIQUE FOR RECONSTRUCTING LOW RESOLUTION SURVEILLANCE VIDEOS


Terrorism, ongoing destruction of public facilities, attacks on human and the recurring global
security insurgences have posed new opportunities for the sporadic influx and deployment of
Video Surveillance Systems. Consequentially, videos captured by these systems are typically of
low resolution. This challenge greatly accounts for the failure of most existing video-based face
recognition systems. However, most existing promising solutions to this challenge are
computationally very expensive and inefficient for restoring continuous variation region and
suppressing blocky artifacts within reasonable time bound. In this paper, a hybrid Iterative Back
Projection – Maximum a Posteriori Technique (IBP-MAP) for Pixel domain super resolution
reconstruction of low quality surveillance video feeds was developed. L1, a sparse prior and
Simultaneous Auto Regression, a non-sparse prior of Bayesian MAP were combined with
Iterative back projection, a spatial domain method to realize appreciable denoising, edge and
non-edge preserving properties for super-resolved high resolution video frames in a
computationally-efficient manner. The performance of the developedhybrid IBP-MAP technique
was evaluated using Peak Signal to Noise Ratio (PSNR) and Improvement in Signal to Noise
Ratio (ISNR). Results obtained using the hybrid IBP-MAP technique show significant
improvements over existing techniques in terms of quantitative and visual qualities of the
video sequences in a time-efficient manner.