阈值自适应的SIFT全景图像拼接算法

Adaptive Threshold SIFT Panorama Image Matching Algorithm

针对SIFT(Scale Invariant Feature Transform)算法中阈值影响着图像匹配的成功率,提出了一种阈值自适应的匹配算法。该方法首先对SIFT算法中的阈值系统地研究,发现检测局部特征点的阈值α和图像匹配时最近距离与次近距离的比值的阈值β对图像是否能够成功匹配起着决定性的作用,然后利用控制α的大小来检测特征点,生成特征描述符。再利用广义紧互对原型的基础上,自动调整β的大小来控制匹配的对数,最后结合RANSAC和最小二乘法求出图像间的映射关系得到拼接后的图像。实验结果表明,该算法通过自动调整阈值和利用RANSAC剔除误匹配点,加快了图像的匹配速率,开发了全景图像拼接软件。

Aimed to the success rate of image matching which is affected by thresholds of the SIFT(Scale Invariant Feature Transform)algorithm, a thresholds of adaptive matching algorithm was proposed. Thresholds were first systematically investigated and found that the thresholds of local feature points( α)and the ratio of distances(closest/next closest)( β)had influence on successfully matching images. Then, controlled α was used to detect feature points and generate feature descriptors, and based on the generalized tight interaction of the prototype, automatically adjusted β to control matched pairs. Last, RANSAC and least-squares method derived mapping relations between images image stitching. Experiments show that the algorithm can automatically adjust the threshold so that accelerate the image matching rate, a panoramic image matching software have been developed.