Research On Parallax Image Stitching Based On Super Pixel Segmentation

With the rapid development of industries such as virtual reality,artificial intelligence,video production,and mobile terminals,people’s demand for wide-view panoramic images is increasing.Image stitching technology is widely used in panoramic image shooting,video conferencing,security surveillance and other fields.This technology can effectively synthesize wide-view panoramic images to meet people’s daily needs and improve the quality of life.For now,some classic algorithms can better stitch the parallax images,but they cannot effectively solve the problems of image misalignment,distortion,and stitching in the stitched image.Therefore,this article combines the current market demand and application value to study the splicing method of parallax images.The experimental subjects in this article come from pictures taken in any scene on the handheld mobile terminal and public image collections,and are dedicated to solving the stitching problem caused by image parallax to obtain high-quality stitched panoramas.The main research content and innovations of this paper are as follows:1.Researched a feature matching algorithm based on mapping constraintsAiming at the problems of insufficient feature matching and mismatching in current feature matching algorithms,this paper studies a feature matching algorithm based on mapping constraints.First of all,the point feature algorithm is optimized to achieve feature retention while improving feature richness.Then,a linear feature matching algorithm based on LSD(Line Segment Detector,LSD)descriptor is introduced to extract and match line features to perfect the image feature types.Finally,the feature points are screened by combining the mapping relationship and threshold constraints to eliminate mismatched pairs.The experimental results show that the algorithm in this paper effectively solves the problems of insufficient feature matching and mismatching.2.Research on a local registration algorithm based on super pixel segmentationAiming at the phenomenon of image misalignment and distortion in the mosaic image during the image registration process,this paper studies a local registration algorithm based on super pixel segmentation.First,the super pixel segmentation algorithm is used to replace the traditional grid division model to perform protective segmentation on the image,obtain the center point of each image subset,and obtain an accurate local homography matrix.Then,the image is locally aligned using the mapping relationship to improve the accuracy of image alignment.The experimental results show that the proposed algorithm can effectively solve the problems of misalignment and deformation in the stitched image.3.Researched an image fusion algorithm based on cosine functionIn the process of acquiring images,due to the influence of factors such as shooting angle and light,there are certain brightness differences and chromatic aberrations between the images,leading to problems such as ghosting,chromatic aberrations,and stitching seams in the stitched images.Aiming at the above problems,this paper studies the image fusion algorithm,and adopts a cosine function-based fusion algorithm with smoother transition to effectively eliminate the ghost and chromatic aberration between images,so that the image has a better visual effect.The experimental results show that the algorithm used in this paper effectively solves the problems of ghosting,chromatic aberration,and stitching in the image.In summary,in view of the insufficient number of feature points matching,image misalignment,stitching seams and other problems in the image stitching algorithm,this article has taken some innovative and targeted measures.The algorithm in this paper has the characteristics of rich feature matching pairs,high image registration accuracy,and smooth image fusion,and it has obtained a wide field of view and high-quality panoramic image.The experimental results verify the effectiveness of the algorithm in this paper.