Infrared and visible light image fusion method based on multi-scale analysis and VGG-19

Abstract

The invention discloses an infrared and visible light image fusion algorithm based on multi-scale analysis and VGG-19, belongs to the field of image processing, and mainly solves the problems of detail loss, low contrast and edge blurring in the current fusion method. Comprising the following steps: step 1, performing Laplacian pyramid decomposition on a source image to obtain sub-bands; 2, for the Nth layer of sub-band, extracting image features by adopting a neural network (VGG-19)-based method, and performing feature fusion by adopting a maximum selection mode; 3, fusing the first layer of sub-bands by adopting a method based on convolutional sparse representation (CSR); fusing the second layer to the Nth layer of sub-bands by adopting a'maximum selection 'strategy; and step 4, carrying out Laplace inverse transformation on the fused sub-bands to obtain a fused image. The invention provides a new fusion framework, and the obtained fusion image has superiority in subjective and objective evaluation compared with other classic fusion algorithms.

Description

technical field [0001] The invention relates to the field of digital image processing, in particular to an infrared and visible light image fusion method based on multi-scale analysis and neural network (VGG-19). Background technique [0002] Infrared and visible light image fusion, as an image enhancement technology, aims to fuse images from different sensors, generate fused images with complementary information and greatly reduce the redundant information of images. According to the sensor principle, visible light sensor captures reflected light and infrared sensor captures thermal radiation. Therefore, in the case of sufficient illumination, visible light images contain rich detailed information; in extreme environments (such as fog, night, etc.), infrared images contain rich structural information, which can capture more useful image information. The fused image will contain more complementary information, detail information and structural feature information. Therefor...

AI Technical Summary

Problems solved by technology

For the space domain method, it is simple and efficient, but it can rarely take into account the difference between infrared and visible light images, resulting in the loss of some image detail information; for the transform domain method, it can solve the problem of image detail preservation, but due to the fusion strategy At present, many methods only use the "maximum selection" or "average" method, resulting in a decrease in the contrast of the fused image; for neural network methods, this type of method can extract deep image structure information and detailed information, but involves training. The process is complicated and the running speed is slow

Images