Non-uniform Motion Image Deblurring Based On Vector Field Fitting

When the image is acquired,the visual scene and the imaging plane have relative motion during the exposure time,so the acquired image will have a certain degree of motion blur.Motion blur results in reduced image quality,which seriously affects people's search for the outside world,access to information,and the development of scientific research.Most of the traditional motion deblurring studies use a simplified uniform motion model,so the reconstruction accuracy is limited.In this paper,the convolutional fuzzy nucleus achieves the restoration of non-uniform motion blurred images,and comprehensively considers the associated motion information among multiple continuous motion images,making it easy to obtain clear images when some inhomogeneous motion blurs are unavoidable.easy.First,this paper proposes a preprocessing method for non-uniform motion deblurring based on vector field fitting.The method selects the matching feature point positions on two adjacent motion pictures,obtains the limited motion vectors by matching the SIFT feature points,maps the motion vectors to the two-dimensional grid through bilinear interpolation,and then passes the weighted minimum two.Multiplication fits the grid data into a two-dimensional vector field.Experiments show that compared with the traditional optical flow method,the vector field estimates the non-uniform motion more accurately,which effectively improves the performance of subsequent tasks.Secondly,this paper divides the recovery process of blurred image into three steps.According to the generated non-uniform motion vector field,the fuzzy kernel of image is estimated.Then the defuzzification kernel is obtained through pseudo-inverse operation,and the blurred image is deconvoluted.blurry.The overall process needs to combine these three steps to obtain accurate fuzzy kernels and clear images by optimizing experimental parameters.