The Algorithm Of Fast Image Mosaic Based On Stitching Seam Elimination And Panorama Alignment

With the rapid development of computer vision,people's vision requirement for images are getting higher,including the high quality and informative panoramic images.Image mosaic technology combines several ordered images with repeating region in the same scene into a panorama,which has great research value and application value.Its application prospects in remote sensing,UAV aerial photography and other fields are very wide.Image mosaic technology is generally divided into three parts: image registration,image fusion and image alignment.Aiming at the phenomenon of uneven color transition,image tilt distortion and low stitching efficiency,the main research of this thesis is how to improve image quality and stitching efficiency.A fast image mosaic algorithm based on stitching seam elimination and panorama alignment was proposed in this thesis.Firstly,the SIFT feature algorithm was improved to extract image features and shorten the image mosaic time.Secondly,the adaptive update mechanism was used to reduce the uneven color transition of stitching seams in different image backgrounds and improve image quality.Finally,an adaptive straightening model was established for panorama to get a completely new panorama.The main work of this thesis includes:1.The classic SIFT algorithm results in a large amount of computation due to the feature extraction of the entire image,and it was easy to cause subsequent matching errors and computational redundancy.The SIFT algorithm was improved in this thesis.Firstly,the feature points were extracted only in the specified image area and the image detection range was limited,so the algorithm time was shortened.Then the bidirectional KNN algorithm was used for image registration,at the same time,the RANSAC algorithm was used to eliminate the mismatching features.So the time efficiency was effectively improved while ensuring the accuracy of feature extraction.2.For the classic image fusion algorithm to directly fuse multiple ordered images after image registration,it would produce very obvious stitching seams,chromatic aberration and artifacts.An image fusion algorithm for optimal stitching adaptive elimination was proposed in this thesis.Firstly,the adaptive energy formula was proposed to find the optimal stitching seam of the stitched image.Then using multi-resolution image fusion for stitching seam region to effectively eliminate artifacts and stitching seam.The proposed algorithm improved the fusion efficiency and image quality.3.Aiming at the inevitable distortion error in image acquisition and image mosaic,an adaptive fitting quadrilateral straightening model for the stitched image was proposed in this thesis.Firstly,the panorama was divided into several same regions.Then the adaptive image alignment was established for each region.The proposed model can retain the panorama information while reducing the distortion of the panorama,and a high-quality panorama can be obtained.Finally,the research work is summarized in this thesis,and the future research direction is prospected.