基于邻域特征搜索与局部单应性变换的图像拼接算法

打开文本图片集

关键词：图像拼接;特征匹配;单应性矩阵；局部变换;线性加权；羽化融合；图像配准;计算效率

中图分类号：TP183 文献标志码：A doi： 10.12415/j.issn.1671-7872.25077

Abstract： Image stitching effectively extends the field of view of a single lens by integrating images with overlapping regions to generate wide-angle panoramas.The typical as-projective-as-posible （APAP） image stitching algorithm achieves high alignment accuracy but sufers from low computational efficiency.To addressthis issue， an image stitching algorithmbased on neighborhood feature search and local homography transformation was proposed to balance accuracyand eficiency.The source image was first divided uniformly into multiple rectangular grids，and feature pairs within the neighborhood of each grid center were obtained using a neighborhood feature search strategy.The local homography matrix for each grid was thencalculated via the least squares method，and a smooth transition was achieved through linear weighting of globaland local transformations.Subsequently，the alignment and overlayof the source and target images were realized through perspective transformation，and seam artifacts were eliminated using a distance transform-based feathering fusion method.To validate the efectiveness of he proposed algorithm，computer simulations and multi-scene image stitching experiments were conducted. The results demonstrate thatthe proposed algorithm outperforms global homography transformation （GHT），adaptive as-naturalas-possible （AANAP），seam-guided local alignment and stitching （SLAS），and APAP algorithms in terms of noise resistance. Compared with APAP， the root mean square error is reduced by 8.9% ，while the peak signal-to-noise ratio and structural similarity index are improved by 11.4% and 12.5% ，respectively，witha 29.4% reduction in computational time. This study simplifies the computation of the local homography matrix， which significantly improvescomputational efficiency while maintaining stitching quality，thereby offering a new direction for image stitching technology.

Keywords：image stitching；feature matching；homography matrix；local transformation；linear weighting; feathering; image registration; computational efficiency

随着国家在数字经济、智能制造及智慧城市建设等战略方向的深入推进，地理测绘、医学成像、智能驾驶和虚拟现实等前沿领域对高精度、高效率的图像处理技术变得日益迫切[1]。（剩余15382字）