Using GPU-accelerated Genetic Algorithm for non-linear motion deblurring in a single image

One of the key problems of restoring a degraded image from motion blur is the estimation of the unknown nonlinear blur filter from a single input blurred image. Many blind deconvolution methods typically assume frequency-domain constraints on images, simplified parametric forms for the motion path during camera shake or use multiple input images with specific characteristics. The paper proposes an algorithm for removing non-linear motion blur from a single input blurred image using Genetic Algorithms, by finding proper parameters and goal function. Also recent research in natural image statistics is exploited, which shows that photographs of natural scenes typically obey heavy-tailed distribution. Then a Graphics Processing Unit-Accelerated version of the Genetic Algorithm is presented, that achieved a huge speedup in the running time. The accelerated algorithm works 12.6× faster than the standard Genetic Algorithm. Experiments on a wide data set of standard images degraded with different kernels of different sizes demonstrate the efficiency of the proposed approach compared to other algorithms.