针对尺度变化不明显场景的拼接，提出一种基于特征不变描述的图像无缝拼接算法。利用Harris特征检测算子进行特征点提取，在此基础上对提取出的特征点采用SIFT描述子进行处理，使其具备旋转不变性，然后利用k-d树算法进行遍历搜索定位，并用RANSAC算法进行图像间单应性变换矩阵计算，最后利用加权平均的融合方法进行图像无缝平滑，得到无缝拼接图像。实验结果表明: 该算法在图像任意角度旋转的情况下，能有效地实现图像无缝拼接，并具有较强的鲁棒性。

A new approach for seamless image mosaic based on feature invariant description was proposed. Firstly, the interesting points from two images were extracted by Harris corner detector, and the interesting points were descript by SIFT descriptor. Then the corresponding points were obtained by k-d tree algorithm and RANSAC algorithm. At last, the images were stitched by the corresponding features. The experiments show that this algorithm is invariant to rotation, and it has good robustness to image mosaic.