A Google Colab Based Online Platform for Rapid Estimation of Real Blur in Single-Image Blind Deblurring
Ihsan Ali, Aftab Khan, Muhammad Hasnain Waleed · 2020
Referenceless or Blind Image Deblurring (BID) is a challenging task in image restoration, because of the absence of previous information about the blurring process. In BID, the Point Spread Function (PSF) that causes blur is unknown and finding it is a challenging task to do. In contrast, reference-based image deblurring techniques are better in terms of achieving deblurred images, as in such cases we know the cause of blur and many filters and methods get the job done easily. In real life imaging, the presence of blur is caused due to many reasons including motion, defocus, atmospheric turbulence and noise in capturing device. Removing such blur to produce sharp images is very important in many cases as of today's technological driven era demands for images of unspoiled quality. This paper presents a method to eliminate blur in the fastest way. The scheme is devised to use a Genetic Algorithm (GA) on Google Colab to find PSF of the blur kernel. This work tries to estimate the parameters of blurring using quality score minimizing. Once the PSF is estimated, the Wiener filter is used for deblurring in Google Colab. This scheme presents the implementation of the algorithm on Google Colab to use GPU for achieving parallelism and reduce implementation time. Validation of the work has been carried out on different images and the system achieved marked improvements over existing approaches. The work presented is simple and efficient which requires no former information about the original sharp image or deblurring process.