Spatial Super Resolution Based Image Reconstruction
using IBP and Evolutionary Method

Abstract

Spatial image resolution explains about the pixel spacing in a digital image. As a
result more the number of pixels more detailed visibility of information contained in the
image. Increasing the sensor elements per unit area in camera is the solution to high
resolution images but at higher cost, since increase in camera resolution increases the cost of
the camera. Another limitation to this is hardware part of the camera i.e. we cannot minimize
the sensor element size beyond the optimum limit. Therefore an imaging system with
inadequate sensor array will generate low resolution image which causes pixelization effect
in them.
To resolve the above problem software level techniques are adopted. Interpolation of
a 2D signal improves the size of the image with additional row and column values. As
interpolation averages pixel intensity with neighboring pixels and assigns the result to new
pixels in HR image, the HR image loses edge information because of blurring or aliasing. To
improve image resolution while avoiding aliasing effect we are using Super Resolution
Image Reconstruction technique. As the application of image reconstruction in Computer
Tomography (CT) is capturing multiple 2D images with known depth information can create
3D picture of an infected tissue, here we are using multiple number of low resolution images
with subpixel shift that provide non-redundant information about the scene and generate a
HR image with more high frequency details.
This reconstruction problem uses iterative back projection technique along with
evolutionary techniques for optimization. In our image observation model, three lacunae of
LR images have been considered, i.e. relative motion between scene and camera, sensor blur
and down-sampling during image acquisition. In the inverse model we use the above
calculated parameters in back projection. Iteratively the reverse model is solved to find the
HR image; this method is also called as gradient based method. As the gradient remains
constant over here, the solution it provides may not be the best and hence the requirement of
optimization technique. The nature inspired optimization algorithm used here is Cuckoo
optimization algorithm with Lèvy flights. i