目的 统筹图像配准和缝合线生成过程实现小重叠度、大视角影像自然拼接。方法 主要涉及两方面内容：a)将单应性映射引入TPS(Thin Plate Spline)变换并建立二次配准模型，通过在同一变换框架下进行全局单应性及基于径向基函数的局部映射调整，实现拼接影像透视特性保持、减少局部变形并有效满足缝合线控制点无误差配准、匹配生成计算需要；b)从控制点像素配准精度优于非控制点像素这一事实出发，对基准影像控制点集进行三角构网并从中搜索出初始缝合线S0，进而结合二次配准模型参数、动态规划匹配过程获得控制点更密集且配准误差总体最小的最终缝合线S1，有效抑制图像融合时鬼影现象的产生且计算实现上更简单。结果 对网上两组标准影像进行拼接测试并与3种文献方法对比，本文方法匹配控制点训练集、测试集的配准精度均最高，拼接影像局部变形失真最小；多路、不同场景采集视频影像拼接透视特性保持良好，视觉效果自然。结论 本文方法影像拼接时无先验知识要求、参数可线性求解，整体视觉效果流畅、重叠区域与非重叠区域过渡平滑，具有较好的应用价值
Stitching of parametric-free images based on coordinated image registration and seam-line generation
gaojiongli,wujun,liuqichang,xugang(Guilin University Of Electronic Technology)
Objective This paper presents a novel approach of stitching parametric-free images by comprehensively considering the process of image registration and stitching. Method It mainly involves two aspects: a) the twice registration framework based on modified TPS transformation is established to meet the needs of error-free control point registration and automatic generation of reliable seam-line. In this step, the global homograph mapping and local mapping adjustment based on radial basis function are ingeniously incorporated into the uniform spatial transformation framework, which thus maintain perspective geometry and reduce local distortion in stitching image as possible; b) reliable seam-line is generated based on the fact that the registration accuracy of control point pixels is better than that of non-control point pixels in the process of image registration. In this step, the initial seam-line is primarily searched out from triangulated control points in the reference image and then, refined by adding more control points through dynamic programming matching process and remapping non-control point pixels with the registration parameters estimated in advance. Due to very small registration error of pixels on seam-line, troublesome ghost phenomenon in stitching images can be effectively suppressed with computationally simpler image fusion operation. Result The experiments on two sets of standard images from Internet are presented and compared with the state-of-the-art methods in literature. The registration accuracy of training set and test set of matching control points is the highest in this paper, and the local distortion of mosaic images is the smallest. Meantime, the splicing test is also performed on the multiple video acquiring at two different scenes and its visual mosaic effect is natural. Conclusion proposed method has a natural transition between image overlapping and non-overlapping areas and achieve better visual effect in the stitching image. Meanwhile, proposed method has no prior knowledge requirement about images and its parameters can be solved linearly, which turn out to be valuable for application.