Infrared and visible light image fusion method combining potential low-rank representation and convolutional neural network

Abstract

The invention relates to the technical field of computer information, and discloses an infrared and visible light image fusion method combining potential low-rank representation and a convolutional neural network, and the method comprises the steps: carrying out the preprocessing of a visible light image and an infrared image, decomposing the image into a significant part and a low-rank part through the potential low-rank representation, and fusing the salient part and the low-rank part by using two full convolution models, and finally adding the fused salient part and the fused low-rank part to obtain a final fused image. The fusion result obtained through fusion is rich in detail information, and compared with an original low-illumination image, the scene target is clearer, and the contrast ratio of the image is improved.

Description

technical field [0001] The invention relates to the fields of computer, image fusion and intelligent monitoring, in particular to an infrared and visible light image fusion method combined with latent low-rank representation and convolutional neural network. Background technique [0002] Image fusion is to extract the most meaningful parts from the images acquired by different sensors, and then combine them into an image. The fused image contains more scene description information, which is more convenient for subsequent applications. Infrared and visible light image fusion has unique advantages. The infrared sensor captures the thermal radiation of the object on the image, so the infrared image is not easily affected by the complex environment and can better distinguish the object from the background, but this also results in insufficient detail information and low contrast. The visible light sensor captures the reflected light of the environment, and the imaging is easily...

AI Technical Summary

Problems solved by technology

[0004] In order to obtain better fusion performance, the method based on multi-scale transformation tends to design more and more complex fusion strategies, which leads to difficulties in the implementation of the method and the designed rules cannot ensure that the characteristics of the decomposed image are fully utilized. The method based on neural network will design more complex network structure and loss function, which will make it very difficult to design a model suitable for fusion

Images