Research on Image Fusion Technology of Infrared and Visible Image Based on MST and CNN

Abstract

Target image detection technology based on multi-source data fusion has a wide range of applications in civil, national defense and other aspects, and has become a research hotspot. Due to the inherent characteristics of different types of imaging sensors, the image information collected by sensors with different imaging types is different and complementary. Infrared image and visible image are two common image data. Multi-source image fusion can fuse a variety of different types of image information collected in the above two different situations to become a fusion image that accurately describes the background or target object, and provide reliable data for computer vision work such as target detection in the later stage. Non subsampled shearlet transform (NSST) in MST multi-scale transform (MST) theory is used to decompose visible and infrared images respectively. On this basis, the model is applied to the color fusion of infrared and low light level night vision images, and a night vision image color fusion method based on NSST and pulse coupled neural networks (PCNN) is used. In this method, NSST is used to decompose infrared and low light level images to obtain their low-frequency and high-frequency components respectively; The Kirsh characteristic energy is constructed and fused by using the rule of maximum Kirsh characteristic energy; Then, the fused image, infrared image and low light level image are combined into YUV color space to get the pseudo color image, and the image is finally converted back to RGB color space to get the dyed pseudo color fused image. This method not only enhances the spatial details of the fused image, but also effectively highlights the target, and is superior to the fusion methods based on Laplace transform, wavelet transform, stationary wavelet transform, NSCT, target extraction and NSCT in terms of information entropy, spectral distortion and other indicators.