A Fast Image Stitching Algorithm Based On Expanded Super Pixel

As the most intuitive form of presenting information,people can intuitively obtain the desired information from images.Stitched images can provide people with better information and richer visual experience visually.As an important branch of computer vision,image stitching has never lost researchers' enthusiasm for image splicing.However,the traditional single-mapping stitching algorithm is not strong in image alignment,and some images stitched in some scenes may suffer from ghosting and dislocation problems,which greatly reduce people's visual experience.Recently,the emerging multi-map stitching algorithm has strong ability to align images,but due to the excessive number of aligned cell grids,the program runs too slowly.Therefore,on the premise of ensuring image stitching,this paper focuses on how to improve the speed of image stitching.The reason why the stitching speed of multi-map stitching algorithm is too slow is that the image is divided into dense grids and the number of alignment units is too much.On the premise of ensuring the quality of stitching images,the number of alignment units should be reduced in order to improve the speed of stitching.Based on the above considerations,this paper proposes an image stitching algorithm based on super pixel.The algorithm in this paper still uses the idea of multi-mapping stitching algorithm.Firstly,the image is divided into super pixels,and the super pixels are taken as alignment units.Secondly the weight matrix of the super pixel is calculated according to the coordinates of the super-pixel's center point.Thirdly according to the weight matrix of super pixel,the mapping transforma t io n matrix of super pixel is calculated.In this paper,10 groups of experimental scenes were divided into super pixels.The experimental results showed that the number of aligned cells was concentrated in the range of 500 to 1000,and the number of aligned cel s was reduced by an order of magnitude compared with the number of 10,000 grid aligned cel s assembled by multiple maps.Meanwhile experiments show that,the grid as alignment unit in the image will be a certain overall regional split open,and the algorithm of this paper will super pixels as alignment unit,avoids the image of a certain overall regional split open.Because an entire region in the image should be transformed by using a single mapping stitching,the errors generated in the mapping stitching process can be reduced.After the multi-mapping transformation of the super pixel,the stitched image needs to be interpolated.In order to solve the pixel value loss between super pixels,the super pixels are inflated to fill the problem of pixel value loss between super pixels.In order to solve the problem of missing pixel value inside the super pixel,the weighted average algorithm of 4 neighborhoods is used to fill the problem of missing pixel value inside the super pixel.In this paper,10 groups of experimental scenes were stitched.The experimental results show that the proposed algorithm not only has a strong ability to align images,but also can obta in good stitching image quality.At the same time,it also improves the speed of image stitching,which is an order of magnitude higher than the grid-based multi-map stitchig algorithm.In general,our algorithm in this paper has the best performance.