Research On Infrared And Visible Image Fusion Method Based On Deep Network And Color Transfer

Image fusion is an important part in the area of digital image processing.Infrared and visible image fusion has become a main research orientation in the field of image fusion because of its wide application in production life and military surveillance.Infrared images capture thermal radiation,which can work around the clock and are not affected by the weather.Infrared images have high contrast.Visible images capture reflected light,which is consistent with the visual properties of the human eye.Visible images are susceptible to weather conditions and are characterized by rich detail information and high spatial resolution.The fusion of infrared and visible images can achieve the complementary advantages of the two images and obtain an image with both significant contrast information and rich texture details.However,most of the complementary information in the source images to be fused is not easy to be explored,which leads to the problem that the fusion results are not high in contrast or rich in texture details.Moreover,it is easier for the human eyes to obtain semantic information from color images than from grayscale images,which makes it significant to explore an effective method for fusing infrared and visible false color images.Based on the above problems,the main innovative work in this thesis is as follows:(1)A fusion method of infrared and visible images combining image enhancement and generative adversarial networks is proposed.Firstly,the infrared image is enhanced by bilateral filtering and local weighted scatter plot smoothing strategy.The visible image is enhanced by multi-scale Retinex.Then feature information in the enhanced image is fully extracted by combining the powerful feature extraction ability of the generative adversarial network.Finally,the fused image is decoded and output.Compared with existing algorithms,the method proposed has advantages in both visual perception and objective evaluation indicators.(2)An adaptive contrast enhancement-based fusion method of infrared and visible false color images is proposed.Firstly,the grayscale fused image is obtained by using the image fusion method based on target segmentation and convolutional neural network.Then,the grayscale fused image is inserted into the Y channel of YCb Cr color space to obtain the YCb Cr source color images.The selected natural light color image is transformed to the YCb Cr color space.Finally,the false color image is obtained by matching the color statistics of the YCb Cr source color image with our designed adaptive contrast-enhanced color transfer model.The proposed method not only improves the contrast of the fused image effectively,but also makes the color distribution of the fused image more natural and consistent with the visual observation of human eyes.The simulation results indicate that the algorithm proposed in this thesis has advantages in both subjective visual perception and objective evaluation indicators.