Mixed order l 0 Regularized Blur Kernel Estimation Method

Abstract

The present invention proposes a hybrid L 0 The regularized blur kernel estimation method is characterized by performing mixed-order processing on the intermediate clear image in the blur kernel estimation model. L 0 Regularization constraints use first-order constraints to protect image edges, and second-order constraints suppress the ringing effect caused by first-order constraints to restore a clear intermediate image; then add an improved adaptive adjustment factor to the blur kernel estimation model to make the middle image clearer. More significant edge information that is beneficial to blur kernel estimation is extracted from the image. The proposed fuzzy kernel estimation model can be solved based on the semi-quadratic variable splitting technique. Experiments conducted by the present invention on artificial blurred images and real blurred images prove that the proposed blur kernel estimation method is effective. Compared with the most representative methods in recent years, the restored image has better subjective visual effects and objective evaluation indicators. There is a significant improvement.

Description

technical field [0001] The present invention relates to an image processing method, in particular to a method based on the mixing order L 0 Regularized Blur Kernel Estimation Method. Background technique [0002] In the process of image acquisition, transmission, and storage, due to the influence of factors such as physical defects of the imaging equipment itself, changes in the external environment, and improper operation of the operator, it is inevitable that the image will be degraded to varying degrees. Not only seriously affect the visual effect of the image, but also greatly reduce the practical application value. As a result, image restoration technology emerged as the times require, and has been widely used in astronomical observation, medical imaging, video multimedia, criminal investigation and other fields. At present, many image restoration methods require more prior information, or have disadvantages such as poor effect and high algorithm complexity. For this...

AI Technical Summary

Problems solved by technology

However, when the image contains rich details or a large blur scale, the blur kernel estimated by the current method is generally not ideal, and the image restored from the inaccurate blur kernel will naturally produce different degrees of ringing effect

[0005] Through analysis, when the image details are rich or the blur scale is large, the existing processing techniques are: 1) impose a single first-order regularization constraint on the image, resulting in an accurate estimation of the blur kernel; 2) according to the magnitude of the image edge instead of If the scale is used to estimate the blur kernel, it is easy to misjudge the details of the image as the edge of the image, resulting in inaccurate blur kernel estimation

Images