Based On Feature Image Stitching Technique

Image stitching technology is a comprehensive research field, which involves inimage processing, computer vision and computer graphics. We often need use theclear panoramic images in our daily life and work but due to the limitation of thehardware devices, we only get the local image. At the same time the devices got thepanoramic images are very expensive so the technology of image stitching emerges asthe times require. Image stitching technique has been widely used in remote sensingimage, medical image, video monitoring and computer vision etc.This paper mainly studies the image stitching technique based on feature, includingimage registration and image fusion. Image registration includes: feature extraction,feature description, feature matching and calculation transformation matrix. Thegradient domain technology is studied in the fusion of image, mainly includes:extracting optimal stitching line, detecting and matching the feature points in theneighborhood of the stitching line, propagation deformation and reconstructed image.In this paper, the main work and innovation points as follow:We deeply study of the Harris-Laplace algorithm in the terms of feature extraction,and improve the faults that the same local area easily detected in multiple featurepoints. Firstly grouping multi-scale harris feature point; secondly delete the notobvious feature points according the response value of each feature point to make surethat the number of feature point in every group is not more than four; finally, throughthe local extremum feature point to keep only one point in each group as the featurepoint in the local structure. Using the Sift operator to describe the detected featurepoints, through lots of experiments show that this algorithm can effectively improvethe matching accuracy and reduce the matching time.In the terms of image fusion mainly uses poisson edit method, and puts forwardfour points to improve: firstly growing the stitching line avoids scaning the all pixelpoints in the overlap, which not only can reduce the time of getting stitching line butalso can make the stitching line as far as possible through the smooth area of theoverlap; secondly detects the feature points in the neighborhood of the stitching lineand puts forward a method speeding up the feature matching; thirdly use threedifferent ways to process the deformation vector that can makes the imagedeformation more accurate; the last solve poisson equation introducing chebyshev accelerated overrelaxtion iteration method to improve the speed of image stitching.Experiments show that this fusion method can achieve good vision effect in structureand color.