Research On Image Fusion Method Based On Image Segmentation And Feature Extraction

As an important branch of image processing,image fusion has been widely utilized in computer vision,clinical medicine and digital imaging.Multi-focus image fusion is the process of fusing useful information from multiple images in the same scene into one image.This thesis mainly focuses on multi-focus image fusion methods,and improved two multi-focus image fusion methods,and evaluates the two methods.This thesis mainly includes the following three aspects:1.A multi-focus image fusion algorithm combined visual saliency and wavelet transform is improved.Firstly,the saliency is improved and to get the multi-scale spectral residual model.Then the low-frequency coefficients and high-frequency coefficients of two source images are obtained by wavelet transform.The low-frequency coefficients are fused by multi-scale spectral residual model,and the high-frequency coefficients are fused by local average gradient criterion.Finally,the final fusion image is obtained by inverse wavelet transform of the high and low frequency coefficients fused by different fusion rules.2.A multi-focus image fusion algorithm combined convolutional neural networks(CNN)and algebraic multi-grid(AMG)was improved.Firstly,one of multiple source images is segmented by CNN and Watershed segmentation,and to get the segmentation results.The CNN segmentation results are more accurate than others,but the boundary of the region is multi-pixel boundary.There is over-segmentation phenomenon in Watershed segmentation which each region has a smaller area,so it is not accurate to judge the clarity through the small area.Therefore,the results of the CNN segmentation are utilized to guide the Watershed segmentation to merge the regions through combined the advantages of the two methods.AMG results are utilized for each region as a criterion for clarity to select clear regions and perform final image fusion.Compared with the traditional fusion methods,it can be concluded that the improved algorithm has certain advantages.3.The second method for improvement is improved.It is mainly improved from two aspects.The first one is to modify the criteria for judging image sharpness and change the mean square error to spatial frequency.The second is to segment the two source images,then merge the regions and finally fuse them.From the evaluation index value of the improved algorithm,the improved algorithm has some advantages.In view of the above research,the improved algorithm is verified to be better through utilized Lytro sequence images and some commonly used images.And some evaluation inicators utilized to evaluate it.Compared with other methods,the improved algorithm has certain advantages.