Research On The Fusion Method Of Visible And Infrared Images

The resolution and contrast of the visible image are high and rich in detail.However,when shooting with heavy fog and haze or in the case of poor illumination conditions,the target of the obtained image would become blurred,and the target and background cannot be distinguished.While using the infrared image can just make up for this disadvantage of the visible image.The infrared image is achieved by the principle of thermal radiation.Therefore,it can work at night or in poor illumination conditions.The target and background of the image is separated and the target is highlighted.However,the clarity and the resolution of infrared images are low.And the details of the images are blurred.By fusing infrared and visible images,the advantages of these two imaging methods can be combined,so that the fused image can not only have the target information of infrared,but also have the advantages of high resolution,high contrast and rich details of visible images.In this paper,the advantages and disadvantages of the previous visible and infrared image fusion methods are studied and summarized.Two fusion methods are proposed and tested on the public data set TNO＿Image＿Fusion＿Dataset.And the quality of the image is evaluated with different quality evaluation indicators.The main research contents are as follow:(1)The fusion theory and fusion rules of infrared and visible images are described in detail.And the existing fusion methods are summarized and analyzed.The principles and characteristics of the evaluation indexes for various fusion images are discussed.(2)In order to obtain the fusion image with more detailed information,higher contrast and resolution.An infrared and visible image fusion method based on Gabor filter and Sigmoid function enhancement is proposed in this paper.Firstly,the source infrared image is normalized,and then the normalized infrared image is mapped by Sigmoid function to obtain the Sigmoid enhancement coefficient matrix.Secondly,the Sigmoid enhancement coefficient matrix is used to enhance the source visible image,and the enhanced visible light image is obtained.Moreover,the source infrared image,the source visible image and the enhanced visible image are decomposed by Gabor to obtain their respective base layer and detail layer.Then,different fusion rules are used to fuse the enhanced visible with the source visible image to obtain the fused visible image.Finally,the fused visible image and infrared image are fused by different fusion strategies,and the quality of the fused image is evaluated by subjective and objective evaluation methods.(3)In order to obtain fused images with high structural similarity and high peak signal-to-noise ratio,an improved algorithm based on deep convolutional generative adversarial network(DCGAN)is proposed.Firstly,the infrared and visible images are prefused and fed into the generator to get a fusion image with the same size.Furthermore,the source image pair and the fused image are sent to the discriminator to judge the true or false.Finally,under the constraints of their respective loss functions,the generator and discriminator optimize their internal parameters to achieve the best results.