Single lens space change image restoration method based on Encoder-Decoder depth learning model

Abstract

The invention discloses a single lens space change image restoration method based on an Encoder-Decoder depth learning model. According to the method, firstly, an intelligible image and fuzzy image pair corresponding to a single lens is acquired, a single lens space change data set is constructed, an Encoder-Decoder depth learning model for end-to-end image restoration is constructed, the constructed data set is utilized to train the Encoder-Decoder depth learning model to acquire the trained Encoder-Decoder depth learning model corresponding to each space change image block, for a fuzzy imagenewly shot by the single lens, the fuzzy image is divided into blocks, each fuzzy image block is directly inputted to the corresponding trained Encoder-Decoder depth learning model, and lastly, the restoration result of each Encoder-Decoder depth learning model is spliced to form a final restoration image. The method is advantaged in that individually estimating PSF of the single lens is not needed, plenty of optimization iteration of the method in the prior art can be avoided, the image processing speed is fast, aberrations and chromatic aberration of the single lens can be further corrected, and the method is suitable for actual application.

Description

technical field [0001] The present invention relates to the field of image restoration, and specifically refers to a single-lens space-varying image restoration method based on an Encoder-Decoder deep learning model. Background technique [0002] Single-lens computational imaging is a new research direction in the field of computational photography. The front-end lens contains only one lens, and the image blur caused by lens aberration and chromatic aberration is eliminated by post-image restoration algorithms. Single-lens computational imaging has low cost and small size, and has important research value in the fields of image restoration and optical design. The difficulty of single-lens computational imaging is that the front-end optical imaging structure is too simple. There is only one lens. Since there is no other optical correction structure, the aberration and chromatic aberration of this lens will appear in the captured image, and it is affected by the spherical stru...

AI Technical Summary

Problems solved by technology

[0004] In order to overcome the shortcomings of the above situation, the present invention aims to provide a single-lens space-varying image restoration method based on the Encoder-Decoder (encoding-decoding) deep learning model, which does not need to perform PSF estimation alone, and avoids the complexity of the existing single-lens calculation imaging method The iterative optimization process of the single lens, and simultaneously solve the problem of aberration correction and chromatic aberration correction in the single lens computational imaging process, and realize the end-to-end fast image restoration of the single lens

Images