Research On Multifocus Image Fusion Algorithm Based On Neural Computing

Through some fusion rules,a fusion image is synthesized by multiple images that have been registered or not completely registered,which is image fusion.The image obtained by image fusion contains more information than single source images,and it can describe the scene more accurately and comprehensively.The fusion image is not only more suitable for human visual observation,but also more conducive to subsequent processing of computer vision.Therefore,image fusion,as an important technology,is widely used in many fields,such as medicine,remote sensing,military,computer vision,public security and so on.Because the depth of the shot is limited,it is difficult to get a clear(or focused)image of all the objects in the scene when shooting a scene with two or more targets.That is,when one of the objects in the scene is focused,the other objects will present different degrees of fuzzy.In order to solve the above problems,multiple images from the same scene and focus on different locations need to be fused to obtain an image that all the objects in this scene are clear.The fusion image contains the complementary and redundant information of the source images,and it has the information that is not available in the single source image.This is what we call the multi-focus image fusion.Through the multi-focus image fusion technology,all the objects in the same scene can be clearly presented in one image.This not only increases the efficiency of image information,but also provides the convenience and lay a solid foundation for the further processing of the image,such as target detection and recognition,image segmentation,edge detection and so on.In the paper,some problems of neural computing in multi-focus image fusion algorithm are studied,and the improved algorithm is proposed.The main innovations of the paper are as follows:1.At present,the image fusion algorithms based on image characteristics mostly use a single index,and the indices are usually used to reflect the sharpness of the image itself.However,only using a single index to weigh the clarity of the image cannot fully reflect its characteristics.Most image fusion algorithms do not consider human visual characteristics when selecting indices,and one of the purposes of image fusion is to facilitate visual observation of human eyes.Hence,a multi-focus image fusion method based on the extreme learning machine(ELM)and human vision system is proposed in the paper.The algorithm takes full advantage of the visual features and extracts three visual features,texture,gradient and local contrast,of the source image as the input of ELM,and then the initial fusion image is obtained.Initial fusion decision map is obtained by judging the focus area of the source images.Here,root mean square error(RMSE)which is general used to judge the focus area is adopted,and mainly calculate the RMSE between source images and the initial fused image.The final fusion decision map is obtained by optimizing the initial fusion decision map,and the fusion image is acquired by weighting the source image and initial fused image according to the fusion decision map.Comparing the method with the traditional fusion algorithm and the recently popular fusion algorithm,the experimental results show that the fusion algorithm can obtain better fusion results,and computational efficiency of this algorithm is also high.2.The defect of many image fusion algorithms based on pulse coupled neural network(PCNN)is that the parameter setting is fixed.Therefore,a novel multi-focus image fusion method based on adaptive PCNN and combined with robust sparse representation(RSR)is proposed,in which RSR can well separate the focus region and non-focus region of multi-focus images.In the algorithm,the source image is represented by RSR,and the sparse coefficient matrix and residual matrix are obtained.The residual matrix can better distinguish the focus area from the non-focus area,and it contains more details information of the source image.Therefore,the spatial frequency of the residual matrix is extracted as the input of PCNN.Unlike most of the previous PCNN fusion algorithms,we first calculate the salience map of the source image and use its pixel values as the corresponding link strength in PCNN.Through PCNN,we get the ignition map.The initial decision map is obtained by using max rule on ignition times of the source image ignition map.Then optimize the initial decision map,and the final decision map is obtained.The fused image is obtained by weighting the source images according to the decision map.The simulated experiment is done on the proposed method and the proposed method is compared with some popular methods.In terms of both subjective vision and objective quality evaluation,the proposed method is superior to most of the existing image fusion algorithms.