Research On Multi-focus Image Fusion Based On Region Detection

Due to the limitation of image acquisition equipment,it is difficult to get all information of the fixed scene or target from a single image,which brings difficulties to feature extraction,target recognition and other work.In order to solve the problem that a single image can not fully express the original information,image fusion technology can be used to obtain a fused image which is more consistent with the characteristics of human and machine vision according to the information complementarity of multiple input images.In recent years,scholars have proposed many multi-focus image fusion algorithms based on spatial domain in order to obtain more accurate focused and defocused regions.Although these algorithms have achieved better results,the determination of boundary is still a difficult point in the research.Therefore,different focus measures are uesd to realize region detection,and different boundary processing methods are also carried out to obtain better results.The details and results of this thesis are as follows:(1)A method based on morphological component analysis(MCA)and focus region detection is proposed.In order to solve the problem that MCA can not extract image features effectively,non-subsampled shearlet transform(NSST)is used to extract more details of the decomposed sub-images and realize the focus measure.Compared with the traditional MCA algorithm,the secondary processing of weighted map is added.The boundary map is extracted by setting thresholds for two weight maps.Then,an improved fusion method combining sobel operator is proposed to reduce the time consumption.In this method,focus measure is achieved by gradient transformation of the texture sub-images.The fusion results obtained by the proposed methods achieve better visual effect and get higher objective indicators.(2)A method for image fusion based on improved structure salience and non-subsampled contourlet transform(NSCT)is studied.In order to solve the problem that many spatial domain algorithms can not correctly identify small areas,a focus measure method based on improved structural saliency and guided filtering is proposed,which effectively improves the accuracy of small area detection.In addition,since the boundary area is actually a mixture of clear pixels and unclear pixels,morphological transformation is performed on the decision map to obtain the boundary,and NSCT is used for image decomposition and reconstruction to improve the spatial structure information of the fusion results.Therefore,the fusion results preserve more details and realize better performance.(3)A novel image fusion method based on gradient and generalized random walks(GRW)is proposed.Based on the detail enhancement of guided filtering,an improved maximum symmetric surround(MSS)method combined with gradient is proposed to detect the focus region.In view of the problem that the current algorithms are easy to misjudge the boundary of the focus region,three focus measure methods are used to get the initial decision maps.Then the XOR method is used to get the uncertain boundary,and the generalized random walk algorithm is used to achieve the accurate fusion of the focus region.This method avoids the boundary effect of the fused images and gets better fusion results.