Image Stitching Research Based On Improved APAP Image Registration And Optimal Seam Blending

As an information carrier,image is widely used in remote sensing measurement,emergency disaster response,security monitoring,medical imaging,military reconnaissance and so on.However,due to hardware limitations,the captured images are confined to a limited field of view,and distortions often arise when employing panoramic or fish-eye cameras.Therefore,to observe larger areas,it necessitates the utilization of multiple angles,shots,and image stitching technology to generate seamless,natural,and clear panoramic images.Feature registration involves aligning corresponding features in overlapping areas using multiple images,whereas image blending eliminates chromatic aberration and variations in lighting caused by differences in shooting angles and distances.This paper presents the following research contributions:(1)In terms of feature point extraction,this paper focuses on analyzing the advantages and disadvantages of ORB feature point extraction algorithm.Particularly,it addresses the issue of uneven distribution of feature points extracted by the traditional ORB algorithm.To alleviate the problem of concentrated feature points in texture-rich areas,an ORB feature point extraction algorithm based on adaptive thresholding is proposed.The algorithm divides the image into regions for feature point extraction.Simulation results demonstrate that this algorithm improves the registration accuracy of images by enhancing the extraction of image feature points.(2)In terms of image registration,the APAP image registration method is enhanced by introducing a feature point clustering technique.This approach clusters the feature points in the image.In cases where the feature points within a region do not belong to the same class,the image is divided using a quad-tree method until each subregion contains feature points of the same class.Subsequently,registration is performed based on the subregions.The precise division of subregions leads to improved registration accuracy.Simulation experiments validate the effectiveness of the enhanced APAP image registration algorithm in improving the accuracy of image registration.(3)In order to realize the blending of images,this paper analyzes and improves the parameter components within the energy function of the optimal seam search algorithm based on graph cutting.For calculating the color difference term,saturation(S)and luminance(V)in the HSV color space are utilized.In addition,the texture difference term employs entropy based on the GLCM to measure the texture difference between two pixels.The improved energy function enables detection of differences between two images,where the minimum energy function value corresponds to the seam line.Feathering and blending of the two sides of the image occur based on the optimal seam line.Each pixel value is weighted according to the position and distance relationship of the pixels,resulting in a smoother and more natural fused image.Subsequent simulation experiments validate the optimization of blending results achieved by the improved energy function.This paper comprehensively investigates key aspects of image stitching,image registration,and image blending,while addressing their respective limitations.Simulation results demonstrate the optimization effects of the proposed improved methods on image stitching,thereby offering valuable insights for future advancements in image stitching technology.