Research On Image Fusion Based On Regional Characteristics

Image fusion technology achieves an effective and comprehensive representation of a scene by extracting information from images acquired by multiple sensors and fusing the effective information into one image.Image fusion technique has been widely used in military,digital imaging and other fields.But the image fusion technique has still not achieved the desired results,the mechanism and imaging characteristics of different sensors imaging are quite different.In this thesis,three image fusion algorithms are proposed for multi-focus images and infrared and visible images based on regional characteristics,which are studied as follows.(1)In response to the edge diffusion phenomenon in the current spatial domain algorithm for focus region detection,an image fusion algorithm based on side-window filtering technique and majority filtering is proposed.In this thesis to study the edge characteristics of the focus region of multi-focus images,side-window guided filtering and mean filtering for focus region detection on the source image are used to avoid edge diffusion and to obtain a more accurate focus region.And the decision weight map is optimized by continuous filtering.The fused image is obtained by using the weighted fusion rule.The algorithm can better preserve the edge features of the focus region,and accurately extract the effective information of the source image to improve the quality of the fused image.Meanwhile,in order to extract more texture features within the focus region of the source image,an image fusion algorithm based on side-window filtering technique and PCANet(Principal Components Analysis Network,PCANet)is proposed,which combines the effective performance of PCANet in image deep feature extraction to fuse more regional features of the source images to the result image,making the fused image texture detail information expression more comprehensive and richer information.(2)An image fusion algorithm is proposed in this thesis based on partial differential model and gradient information measurement based on the contrast and gradient characteristics of the salient regions of infrared and visible images.The partial differential equation is used to decompose the infrared and visible images to obtain the approximate layer sub-map and the detail layer sub-map.In the approximate layer sub-map,we calculate the approximate layer fusion sub-map based on the difference of regional luminance features.In the detail layer sub-map,we use the gradient operator to extract regional gradient features to calculate the detail layer fusion sub-map,and finally the fused image is obtained through sub-images reconstruction.The algorithm can extract the significant target region information of the source image more accurately,and also has a certain suppression effect on the noise of the source image to improve the visual effect of the fused image.(3)An image fusion algorithm is proposed in this thesis based on latent low-rank representation and salient region detection with the luminance features and contour texture features of the salient regions of infrared and visible images.The source image is decomposed by latent low-rank representation to obtain the salient sub-map and low-rank sub-map,while the noise of the source images is removed to a certain extent.And the source image is detected with the luminance and contour texture features as the targets respectively to add more feature information for the fusion of the salient sub-map and low-rank sub-map.Finally,the fused image is obtained.The algorithm can better maintain the spatial continuity and transfer the effective information of the salient regions of the source images to the fused image more comprehensively,and also has the effect of removing noise and suppressing artifacts.The algorithms proposed in this thesis perform well in both subjective evaluation and objective indexes,showing better robustness and adaptability for subsequent practical applications.