Implementation of image super-resolution on Rasberry Pi
Comfort Abiodun, Iyanda
MetadataShow full item record
Improving image quality has been a bottle neck of image technology for the past few decades. High resolution is a continuous and on-going need and it becomes essential to have the best quality of image in some applications such as in forensic department where in order to receive maximum possible information, image has to be enlarged in terms of size. Image degradation that is caused by blur (as a result of motion of the scene, wrong focus and atmospheric turbulence and point spread function), sensor noise (detector sensitivity, optical imperfections and environmental changes) and aliasing (as a result of under-sampling) is produced in the process of acquiring image through an image acquisition device (Camera). Image super-resolution technique is used to compensate these degradations factors by reconstructing a high resolution image from single or multiple low resolution images to facilitate better visual contents and scene recognitions. There are many applications in which high quality image is required such as remote sensing, satellite imaging, surveillance, medical imaging etc. In this dissertation, an overview of different approaches to super resolve an image is provided, how the techniques works along with their pros and cons accordingly, recent improvements carried out by different researchers is also included. Iterative backprojection (IBP) which is one of the methods of super resolution is implemented on a Raspberry Pi in order to analyse it performance and accuracy measuring it robustness to noise. The IBP approach is conducted by starting with an initial estimate of the SR image, and then it compares the projected LR results with the observed images and updates the HR estimate according to the errors that is predicted. MATLAB is used to design the algorithm and the same code is run on Raspberry Pi with the use of GNU Octave application. The performance of IBP is confirmed by calculating the MSE and PSNR respectively.