Research On Multi-focus Image Fusion Algorithm Combining Filter Operator And Bi-scale Decomposition

In recent years,image fusion plays an important role in information fusion,computer vision and machine learning.In some machine vision tasks,due to the limitations of imaging sensor equipment,the camera lens can only focus on the objects with high contrast and clarity in a certain depth of field or a certain distance in the natural scene,otherwise the target will become blurred.Therefore,in order to effectively evaluate the blur of the input image,identify as many targets and details as possible as well as high-definition regions,and combine these effective data to generate more informative images,In this paper,we propose two algorithms for multi-focus image fusion using focus region detection strategy:(1)Multi-focus image fusion based on guided filter banks and gradient covariance matrixThe algorithm based on gauss Laplace filter and differential source images,Realize the separation of high frequency information of multi-source focused image,and use the double dimension analysis method on multi-source image processing complementary properties,as well as the focus measure method based on structure in the advantage of high frequency component processing,complete the image preprocessing in linear mixed double scale domain,to obtain the initial decision diagram;Then,the initial decision graph was refined step by step based on consistency test to generate the fusion decision graph.Finally,a conventional per-pixel weighted average method is used to obtain the fused image.Experimental results show that,compared with other focusing strategies,the proposed focusing region detection method has higher robustness and focusing region recognition ability for different noises.(2)Multi-focus image fusion based on guide filter bank and NSMLIn this algorithm,the salient region of the target image is obtained by means of the mean filtering of the source image and the digital subtraction technology,and the corresponding rough focusing region and refined focusing region are obtained by using the improved Laplacian operator for the double-scale decomposition of the subtraction image.Then,the initial decision graph is generated by the pixel-level linear mixed rules,and the final decision graph is obtained by refining the initial decision graph by the consistency test method.Finally,synthesize the results to reconstruct a new fusion image.The experimental results show that the improved algorithm is more robust to the noise than the fusion image generated by the control fusion algorithm,and has a better ability to identify the small and medium-sized defocus or focus area of the fusion image,and the identified edge information is clearer and smoothed.