Image splicing method based on feature point clustering four-fork division and local transformation matrixes

Abstract

The invention discloses an image splicing method based on feature point clustering four-fork division and local transformation matrixes, and belongs to the field of image processing and computer vision. SIFT (scale-invariant feature transformation) features of two images to be spliced are extracted and matched, one image is selected as a reference image, and matched feature points extracted from the other image are clustered by a k-means method. The non-reference image is subjected to layer-by-layer four-fork division according to clustering results, so that each subspace only contains one type of feature points. The transformation matrix corresponding to each subspace divided in the non-reference image is obtained by an MDLT (mobile data link terminal) method, each pixel value of each subspace divided in the non-reference image is subjected to coordinate transformation by the transformation matrix, and the non-reference image is spliced with the reference image in an aligned manner tofinally obtain a result map. By the aid of inherent properties and distribution conditions of the feature points, the number of the transformation matrixes is greatly decreased, a picture conformingto human eye vision cognition is still obtained, and the calculation efficiency of the splicing process is integrally improved.

Description

technical field [0001] The invention relates to the fields of image processing and computer vision, in particular to an image mosaic method based on feature point clustering quadrangle division and local transformation matrix. Background technique [0002] Image stitching technology refers to the stitching of two or more overlapping single-view images of the same scene to obtain a clear result image with a wide view angle including all source images. With the development of image processing technology, the research of image mosaic technology is deepening continuously, and its application is becoming more and more extensive. In the field of aviation, the pictures taken by space probes and satellites are sent back together for splicing to obtain more information; in medicine, the pictures obtained by CT, X-ray and other medical imaging technologies are spliced ​​to find more pathological information; In addition, many other technical fields, such as remote sensing technology,...

AI Technical Summary

Problems solved by technology

However, the transformation models they use are only suitable for flat scenes or changes in field of view caused by rotation only. When the data does not fully conform to the assumptions of the transformation model, the splicing results will have obvious ghosting and misalignment.

However, in actual use, it is difficult for the images to be stitched to fully meet the strict conditions of using the existing model. There are often depth differences between the two images. Using these existing software or tools for stitching will cause blurring, ghosting, or errors. Alignment and other phenomena, even with superior pre-processing and post-processing, it is difficult to completely remove the ghosting phenomenon caused by the inherent defects of the transformation model

[0006] In order to solve the problem of alignment accuracy, the existing research divides the image into distant view and near view and transforms them separately, but the trial scene is too limited; or the image is divided into tens of thousands of very fine grids and each grid in turn The corresponding transformation matrix is ​​used to transform the grid. Although the alignment accuracy is greatly improved, the calculation amount is too large and the processing efficiency is too low; there are also some studies that add constraints such as straight lines or shape corrections in the transformation process. Although for some extreme large Parallax scenes can continue to improve alignment accuracy but inevitably further increase computational complexity

Images