Effect of downsampling on the quality and resolution of super-resolved images
More and more municipalities in the Philippines are installing CCTV cameras in their streets. But if the cameras are of low resolution or the cars are too far away then identity marks of the car cannot be read. If a video of the car was captured, super-resolution techniques can be applied to reconstruct a high-resolution image from the low-resolution video. In this paper, we examine the effect of downsampling on the quality and resolution of a super-resolved image by measuring their structural similarity index and modulation transfer function at varying resize factors. We simulated a CCTV footage by capturing movies of a remote controlled toy car with a signage attached to its side. To simulate low resolution, we resized video frames from a factor of 0.1 to 1 of their original dimensions. It was observed that text is recognizable at super-resolved images having SSIM index values of greater than or equal to 0.5 which corresponds to a scale factor of 0.41 (in SSIM = 0.5). At this scale factor, the resolution limit is at 7 pixels per line pair.