Visible light and infrared image fusion method based on local processing convolution dictionary learning

Abstract

The invention discloses a visible light and infrared image fusion method based on local processing convolution dictionary learning. The method comprises the following steps of decomposing a source image; performing low-pass component fusion; detail component fusion; image reconstruction. The method is advantaged in that a problem of global convolution sparse coding is solved, and the defect of fuzzy fused images is overcome; experimental results show that the method is suitable for industrial production, the method provided by the invention is superior to other comparison methods in subjectiveand objective evaluation, and compared with a deep learning fusion algorithm, the objective evaluation indexes QAB / F, Qe and Qp are averagely improved by 5.18%, 5.02% and 4.77% respectively, the global information of the source image is better expressed, and local details are reserved.

Description

technical field [0001] The invention relates to the technical field of image fusion, in particular to a visible light and infrared image fusion method based on local processing convolution dictionary learning. Background technique [0002] According to its own imaging principle, thermal infrared images can distinguish the background from the monitoring target, which is very effective for all-weather and night monitoring, but the obtained visible light images are low in definition and blurred, and their recognition is very low; on the contrary, visible The image can provide texture details with high spatial resolution and clarity, but it cannot detect the target well at night or with occluders. Therefore, through fusion technology, it can not only achieve high resolution of visible light, but also have thermal radiation information of thermal infrared images. [0003] In recent years, infrared and visible light fusion algorithms have developed rapidly, and scholars at home a...

AI Technical Summary

Problems solved by technology

However, several popular fusion algorithms mentioned above still have certain problems: methods based on multi-scale transformation rely too much on predefined transformations, corresponding decomposition and reconstruction levels, yet there is still no standard for evaluating these transformations and levels, blind Choosing a transformation can easily reduce the fusion performance; it is still difficult to construct a suitable over-complete dictionary with a good representation of the target data based on traditional sparse domain methods; fusion methods based on deep learning need to design appropriate neural networks and adjust the corresponding parameters

Images