一种自适应四叉树分块的图像拼接算法

An image stitching algorithm via adaptive quadtree segmentation

图像拼接技术是计算机图形学的研究热点,具有代表性的As-Projective-As-Possible(APAP)算法采用均匀分块法来实现图像拼接的局部映射,没有充分考虑图像不同区域之间的差异性.为了兼顾图像拼接的精度和效率,提出了一种基于特征点分布的图像映射算法,利用四叉树对图像分块来实现图像拼接的局部映射.首先,设定四叉树的最大深度,并根据图像匹配点的总数估计出每个图像块的匹配点阈值;然后,利用最大深度和匹配点阈值对图像做自适应四叉树分块;最后,对不同的图像块使用移动线性变换估计出单应性矩阵,依据估计的结果完成图像拼接.实验结果表明,提出的算法提高了图像拼接的精度,其速度远大于APAP算法,并且提升了拼接图像的质量.

Image stitching is a hot topic in computer graphics research, and the representative As-Projective-AsPossible（APAP） stitching algorithm segments image into uniform blocks to achieve local image warping, which does not take into full account the differences between varying image regions. In order to achieve the tradeoff between the accuracy and the efficiency of image stitching, an image warping algorithm based on the feature points distribution is proposed, which uses quadtree to segment image into blocks to achieve local image warping for image stitching. Firstly, the maximum depth of the quadtree is set, and the matching threshold of each image block is estimated according to the number of image matching points. Then, the maximum depth and the matching point threshold is used to adaptively segment the image into blocks by quadtree. Finally, homography matrices of different image blocks are estimated by Moving Direct Linear Transformation algorithm to complete the stitching. The experimental results show that the proposed algorithm notably improve the accuracy of image stitching, and the speed is much faster than the APAP algorithm, and the quality of the stitched image is ameliorated accordingly.