Image correction method based on deep learning

Abstract

The invention discloses an image correction method based on deep learning, and the method comprises the steps: (1), photographing images of the same field of view as much as possible through employing image collection equipment with color difference and good image collection equipment, and taking the images as a color difference image and a reference image; (2) solving offset of two times of shooting by using a template matching algorithm, cutting the two images according to the offset, and further dividing into a training set and a test set; (3) constructing an image correction model which comprises a weight prediction network and n learnable 3D lookup tables; (4) inputting an image with chromatic aberration into the network, comparing the corrected image with a reference image, and calculating a loss function; training by taking loss function minimization as a target, and updating network parameters; and (5) after model training is completed, image correction application is carried out. According to the method, the operation steps are simple, a large number of parameters and design algorithms do not need to be manually set, and the time for manually processing the image is effectively shortened while the good effect is ensured.

Description

technical field [0001] The invention belongs to the field of image processing, and in particular relates to an image correction method based on deep learning. Background technique [0002] Usually, when an imaging system is used to capture an image, the image quality is poor due to the chromatic aberration of the lens used, or the mismatch between the hardware of the imaging part and the hardware of the image acquisition part, resulting in the problem of chromatic aberration in the captured image. Specifically, the degradation of image quality caused by chromatic aberration is a serious problem. The chromatic aberration is mainly due to the different refractive indices of light of different wavelengths passing through the lens, resulting in different focal points of light of different wavelengths, so that the image plane produces The positional deviation of different colors, when the lens is simplified due to the spatial position, or when a lens with a higher magnification a...

AI Technical Summary

Problems solved by technology

[0006] The above two technical solutions use traditional image processing methods, which require a large amount of manual parameter adjustment

And the principle is mainly based on the processing of image edges. For some complex images with many edges and textures, the algorithm takes a long time

And the color difference of the image with more textures is not obvious and has alternate coverage, which will increase the amount of parameter adjustment of the algorithm

Images