基于局部特征点配准的图像拼接算法

为解决尺度、视角、光照变化较大及存在噪声和模糊变化情况下的图像拼接问题, 提出了一种具有较强鲁棒性的图像拼接方法。首先, 根据Harris算法和SIFT算法各自的特点, 提出了一种自适应的Harris-SIFT特征点提取方法, 利用最邻近法完成图像间的特征点粗匹配; 然后, 应用随机抽样一致性(Random Sample Consensus, RANSAC)算法对粗匹配的特征点进行筛选, 同时估计出透视变换模型的变换矩阵, 并对相邻的两帧图像进行配准; 最后, 利用加权平均融合算法消除图像拼接处的缝合线, 实现图像的高质量拼接。实验结果表明, 该算法在提升SIFT算法鲁棒性的同时, 还增强了图像拼接的效果, 消除了图像亮度和色度差异的影响。

Image Mosaic Approach using Local Feature Points Registration

In order to solve the problems of scale, viewpoint and brightness changes, also with the noise and blurring changes in image mosaic, a novel image mosaic approach with stronger robustness was proposed. Firstly, according to the distinguishing feature of Harris algorithm and SIFT algorithm, an adaptive Harris-SIFT feature point extraction method was proposed, and most-adjacent method was used to realize coarse matching of points between images. Secondly, random sample consensus (RANSAC) algorithm was adopted to filter the coarse matching key-points, and transformation matrix under perspective collineation was estimated, at the same time image registration was executed between two adjacent images. Finally, weighted fusion algorithm was utilized to remove stitch line in the area of image mosaic, and high quality image mosaic was achieved. Experimental results demonstrate that the proposed approach can not only promote the robustness of SIFT algorithm, but also reinforce the effect of image mosaic and eliminate the impact of image brightness diversity and chrominance difference.