Sequence image stripe noise elimination method

Abstract

The invention discloses a sequence image stripe noise elimination method. The method comprises the following steps: constructing a sequence image space-time correlation information extraction module by utilizing three-dimensional convolution; constructing an image two-dimensional reconstruction module by utilizing two-dimensional convolution; forming an initial convolutional neural network according to the sequence image space-time correlation information extraction module and the image two-dimensional reconstruction module; creating a sequence image training set; training the initial convolutional neural network by using the sequence image training set to obtain a trained convolutional neural network; and sending a sequence image containing the stripe noise into the trained convolutionalneural network for processing to obtain a denoised image. According to the sequence image stripe noise elimination method, a sequence image space-time correlation information extraction module is added into an original convolutional neural network, and noise can be better predicted by learning inter-frame similar information.

Description

technical field [0001] The invention belongs to the technical field of digital image processing, and in particular relates to a method for eliminating band noise in sequence images. Background technique [0002] Due to the limitations of the manufacturing process and the influence of the environment, the image sensor will produce stripe-like fixed pattern noise during the imaging process, which seriously affects the image signal-to-noise ratio. Especially in infrared or hyperspectral imaging systems, image details are largely lost due to band noise pollution, which brings great difficulties to subsequent target detection and recognition. In view of the long cycle and large investment in the improvement of the manufacturing process, the image restoration technology based on signal processing is of great value in eliminating image band noise and restoring the original scene information. [0003] Traditional image denoising algorithms use image prior models for denoising, such...

AI Technical Summary

Problems solved by technology

In 2018, Xiao et al. proposed a ten-layer residual network (ICSRN) to remove the band noise in the image. The network obtains more feature representations by splicing the network layers. ICSRN's band noise The removal has achieved a certain effect, but it still needs to be improved in terms of detail restoration

At present, the field of image denoising has made great progress, and various convolutional neural network models have been continuously proposed, but there are generally problems such as serious noise residue and poor detail restoration effect.

Images