Simple lens PSF (Point Spread Function) mean value fusion method based on different fuzzy kernel priors

Abstract

The invention discloses a simple lens PSF (Point Spread Function) mean value fusion method based on different fuzzy kernel priors, and relates to the technical field of image restoration. The method comprises the following steps that: utilizing a simple lens camera to obtain a fuzzy image; converting a simple lens PSF estimation problem into a blind convolution image restoration problem; adoptingN types of different simple lens fuzzy kernel priors to estimate N pieces of PSF of the simple lens; and carrying out mean value fusion on the N pieces of PSF to obtain a final PSF. The PSF estimatedwith the method can simultaneously own characteristics embodied by different fuzzy kernel priors. Compared with the PSF estimated by single fuzzy kernel prior, the fused PSF is more accurate and is favorable for the subsequent processing of simple lens imaging.

Description

technical field [0001] The invention relates to the field of digital image processing, in particular to a simple lens PSF mean fusion method based on different fuzzy kernel priors. Background technique [0002] In recent years, with the development of computational photography technology, simple lens combined with later image restoration algorithm is gradually becoming a new research direction in the field of camera design and image processing. A simple lens consists of one or several lenses. Due to the different refractive indices of the spherical lens for different wavelengths of light during the imaging process, the image directly taken by the simple lens is blurred due to the influence of lens aberration and dispersion. In order to obtain a clear restoration For images, accurately estimating the blur kernel of a simple lens is the key. The blur kernel of a simple lens is also called a point spread function (Point Spread Function), and the PSF contains blur information s...

AI Technical Summary

Problems solved by technology

However, the main problem with this method is that generally only one fuzzy kernel prior is added to the objective function, and multiple fuzzy kernel priors will increase the difficulty of solving the objective function, which may cause the objective function to not be a convex optimization problem or have no solution.

A fuzzy kernel prior is only a manifestation of a certain aspect of the fuzzy kernel, and the estimated PSF will reflect more of the characteristics corresponding to the prior aspect of the fuzzy kernel, while other characteristics of the real fuzzy kernel will be weakened or not reflected. This will affect the estimation accuracy of PSF and thus the final image restoration quality

[0004] In Chinese Patent Application No. ZL.2015100547840, a single-lens computational imaging method based on combined blur kernel structure prior is introduced. This method adopts different blur kernel priors in the central area and edge area of ​​the image, because the The blur kernel characteristics of each area are different, but only one blur kernel prior is used for each area, which cannot reflect all the characteristics of a simple lens blur kernel. The estimated PSF and the final restored image quality need to be improved.

Images