An Automated System to Accelerate Image Reconstruction Using GPU

Abstract

Whenever an image is distorted numbers of algorithms are
used to reduce the distortions in the image. Couple of times
algorithms are used to delete the distorted objects from the
digital images. One of the algorithms used to do this ROI
(Region-of-interest) approach. In this approach the salient
object is reconstructed but background image remains blur
also these approaches are more time consuming. To overcome
this issue, we propose method called HEMS (Hierarchical
Exemplar-Based Matching Synthesis) using GPU, in which
once salient object regions are encoded only quantized color
feature and local descriptor of background are kept achieving
bit rate reduction. So to reconstruct the background image
there are 3 steps: Firstly, search image from database, which
have relevant images, limitating search to feasible number of
patches. Secondly, patches are matched by color features to
select appropriate candidates. Finally, distorted optimized
image synthesis makes it possible to automatically choose
most suitable texture sample and reconstruct the image. At last
the algorithm is given to GPU which automatically reduces
the time for reconstructing the images.