Multi-focus Image Fusion Based On Super-resolution Reconstruction And Focused Area Detection

With rapid development of science and technology,image has become an important way to obtain information.Due to the limitation of sensor equipment and the interference of external factors(such as light intensity,temperature,etc.)during the process of image acquisition,the image quality is often poor and a single image cannot reflect all information in the scene.Therefore,image fusion methods are usually adopted for information complementation.The fused images are widely used in fields such as military,medical diagnosis,and security monitoring.As an important branch in the field of fusion,multi-focus image fusion has solved the problem of limited depth of field of optical devices.The focused images in different depths of field in the same scene can be merged to generate a clear image.Most traditional fusion algorithms process the source image directly in the spatial or transform domain by designing different fusion rules and focus detection methods.Although these algorithms can achieve image fusion and solve the problem of limited depth of optical devices to a certain extent,the poor quality of the source image often leads to bad selection of focus area and further causes artificial artifact or fuzzy boundary between focus and defocus areas in fused image.For the above problems,this paper deeply studies the image quality improvement and focus area detection and proposes two multi-focus image fusion algorithms based on super-resolution reconstruction and focus area detection,applies it to the color multi-focus image fusion.The specific research contents is as follows:(1)Multi-focus image fusion based on dual-channel convolutional neural network reconstruction and guided filteringTo improve the quality of the source image and the effect of fusion,this paper uses a dual-channel convolutional neural network based super-resolution algorithm for focus area detection and proposes a multi-focus image fusion algorithm combined with bilateral filtering and guided filtering.First,the source image is input into the dual-channel convolutional neural network to restore the details and structure of the image,which will improve the contrast of the image.Secondly,bilateral filtering and guided filtering are used to reduce the interference of noise on the fusion image and detect the focus area of the image respectively,which continuously refine the decision map.Finally,the fused image is obtained by weighted fusion of the source image according to the decision graph.Experimental results show that the algorithm in this paper has strong robustness,excellent retainment of image information and spatial consistency.(2)Multi-focus image fusion based on deep ResNet reconstruction and structural gradientIt is difficult to distinguish the boundary between the focus area and the defocus area for the traditional multi-focus image fusion algorithm which leads to the poor edge continuity in the fusion image.In order to solve the above problems,this paper proposes a multi-focus image fusion algorithm based on deep residual network reconstruction and structural gradient.First,the source image is input into the deep residual network for low-vision image reconstruction and the details of the image are restored through mapping of the network,and rolling guided filtering is adopted for edge preservation.Secondly,the feature detection of the focus area is detected by combining with structural gradient and the image fusion decision diagram is generated.Finally,the source image is weighted and fused according to the image fusion decision graph to obtain the final fusion image.Experimental results show that the algorithm is more accurate in the edge detection of the focused region.Compared with other algorithms,the recognition accuracy of the focus and defocus region of the decision graph is greatly improved and the detail texture features and edge structure of the source image can be well preserved.