Research On Image Fusion Algorithm With Multi-scale Transform Domain And Visual Perception

Image fusion technology aims to combine multiple source images captured by different sensors in the same scene or the same sensor at different moments into one high quality image.As the imaging information of a single sensor is relatively simple,it cannot cater the application requirements of the real scenario.However,the application of multiple sensor fusion technology can provide more accurate and reliable evaluation and decision-making for human visual perception,computer vision,machine vision and other tasks.In view of the current heavy medical image reading tasks in hospitals and the increasing demand of consumers for taking photos with consumption-grade equipment,this paper makes an in-depth study on medical image fusion and multi-exposure image fusion,and discusses some problems that may encounter in the practical scene application.Specifically,this thesis proposes a robust multi-mode medical image fusion algorithm and a multi-exposure image fusion algorithm suitable for static scenes,separately,which are summarized as follows:In order to improve the efficiency and accuracy of doctors or medical imaging robots,a multi-modal medical image fusion algorithm with parameter adaptive pulse coupled neural network in nonsubsampled shearlet transform domain.The parameter adaptive pulse coupled neural network is used to fuse the high-frequency sub-bands coefficients,and all its parameters can be adaptively calculated through the input sub-bands,which can overcome the difficulty of setting parameters in the traditional pulse coupled neural network.In addition,it can converge quickly,which greatly opens up the possibility that the proposed algorithm can be used in practical clinical applications.We present a novel low-frequency sub-band fusion strategy that simultaneously consider two crucial factors in image fusion,namely,energy preservation and detail extraction.To this end,two new activity measures named weighted local energy and weighted sum of eight-neighborhood-based modified Laplacian are defined,respectively.In order to verify the effectiveness of the proposed method,more than 80 sets of source images were comprehensively verified on the four types of medical image fusion problems.Nine of the most representative medical image fusion methods are compared,several of which are recently proposed.The experimental results demonstrate that the proposed method has reached state-of-the-art level in terms of visual quality and objective evaluation.Considering the difference and correlation between the exposure sequence images and the correlation between the three channels of color images,two visual perception metrics are redefined,and a human visual perception multi-exposure image fusion algorithm is proposed.First,the defects of the traditional saturation definition are analyzed and we found that it is particularly sensitive to the gray content of the image.Therefore it is not used to design the weight function.Then,adaptive two-scale well-exposure is defined in consideration of the difference and correlation between the sequences of the exposed images.Because of the correlation between the three channels of color image,the traditional two-dimensional gradient is generalized,and the color image gradient is presented to better reflect the change of illumination.Finally,considering the efficiency of multi-scale transform,the Laplacian pyramid framework is used for fusion.The proposed method and the current five best methods are evaluated on the multi-exposure image benchmark dataset(21 sets of exposure sequences).The experimental results demonstrate that the proposed method can achieve state-of-the-art performance on both subjective perception and objective measure,and the efficiency of the proposed algorithm is also satisfactory.