Blind deblurring of single image based on kernel estimation of texture image

A blind image deblurring method based on image texture autocorrelation is proposed to address the difficulties of unclear edges, and inaccurate main structure in the process of blind image deblurring. First, the blur-causing motion information is retrieved by examining texture autocorrelation in the frequency domain, and a high-quality blur kernel is directly created. Furthermore, to constrain the high-frequency components and refine the blur kernel, a regularization term of the image gradient to norm ratio is included to build a new deblurring model, which can improve the stability and accuracy of blur kernel estimation. Additionally, the deblurring models optimization problem is tackled by regularizing the blur kernel and latent image. Ultimately, deconvolution is used to obtain the reconstructed image. The findings of a comparative analysis of subjective and objective experiments with classical approaches reveal that the proposed method has greater visual effects in terms of overall detail of benchmark datasets, as well as actual blurred images such as nature and human faces.

I. Introduction

People have grown accustomed to utilizing portable imaging devices to record their daily lives since their popularity. However, factors such as target object movement and portable device shaking may ultimately result in blurry images. Blurred images may cause substantial losses to personal and property safety, as well as the national economy, as the demand for high-quality images of visual equipment grows, particularly in the domains of medical treatment, transportation, remote sensing, and automatic driving. As a result, research on deblurring algorithms based on visual equipment has emerged as a critical task and hotspot.