Infrared And Visible Image Fusion Method Based On Dual Domain Enhancement In Low Illumination Environment

Infrared and visible image fusion is an important research topic in the field of multi-sensor fusion.In the harsh environment,the visible image can not make a complete description of the scene information independently.With the help of infrared image,the amount of image information can be greatly improved.Especially in the low illumination environment,the visual effect of infrared image and visible image is poor and the contrast is low.After reading a lot of literature,this thesis finds that the previous algorithms directly fuse the source image with poor quality,resulting in problems of low contrast and loss of detail information in the fusion image.To this end,two infrared and visible image fusion algorithms based on dual domain enhancement are proposed in this thesis.The source image is preprocessed and enhanced before fusion to improve the quality of the source image:for the infrared image,the local adaptive processing algorithm based on Retinex decomposition is used to improve the contrast and edge detail information of the image;for the visible image,the low illumination image enhancement algorithm based on logarithmic image processing subtraction model is used to enhance the brightness and detail texture of the image.The specific work and innovations of this thesis are as follows:（1）An infrared and visible image fusion method based on filter decomposition is proposed.Firstly,the edge preserving filter is used to decompose the preprocessed infrared image and visible image into two different layers:base layer and detail layer,and then the saliency detection method based on spectral residual is used to extract the saliency part of the preprocessed image,which is used to calculate the weight,and the weight is used to sum the base layer to control the overall contrast of the fused image;for the detail layer,in order to retain more detail texture information,the summation fusion strategy is adopted.Finally,the final fused image is reconstructed by simple linear combination.Experimental results show that the proposed method can fuse more background details in low illumination environment,and the fused image has clearer and more natural visual effects.（2）An infrared and visible image fusion network based on dense nested connection is proposed,which can extract and retain a large amount of detailed feature information in the input image at multiple scales.The network consists of three parts:encoder,fusion layer,and decoder.Firstly,the preprocessed image is input into the encoder network composed of several densely connected convolutional modules to extract multi-scale depth features,and then the fusion strategy based on l₁norm and soft-max operation is used to fuse the features of each scale.Finally,the fused image is reconstructed by using the decoder network based on nested connection.From the perspective of subjective visual perception,the image fused by the algorithm has clear detail contour texture and high contrast.From the objective fusion index,the algorithm can extract the deep detail texture information of the fused image and has better fusion performance.