自适应对比度阈值SIFT算法研究

Adaptive Contrast Threshold SIFT Algorithm

SIFT(Scale Invariant Feature Transform)算法剔除掉对比度小于给定阈值的候选特征点,认为这些点是不稳定的,但是并没有普遍适用的阈值;固定对比度阈值SIFT算法提取的特征点数目,随着图像对比度的降低而急剧减少,并且整幅图像采用一个固定的阈值,会造成特征点的分布不均匀,无法满足图像高精度匹配的需求;因此需要根据图像人工调整对比度阈值;但是人工调整阈值不能够实现图像的自动匹配,满足不了无法进行人工干预的场合。因此为了提高基于SIFT图像匹配算法的精确性和自动化水平,提出了一种根据特征点局部邻域内的灰度信息,确定对比度阈值的方法,用于改进SIFT算法,并将改进后的算法用于图像匹配;实验结果表明,改进后的SIFT算法能够根据特征点邻域内的灰度分布情况,自动计算对比度阈值,能够很好地适应图像对比度的变化,明显增强了SIFT算法对于低对比度图像匹配的鲁棒性。

SIFT keypoints are rejected if the contrast is less than given threshold. These keypoints are considered as unstable keypoints. But there is no universal contrast threshold. Fixed contrast threshold SIFT algorithim extract less keypoints as the image contrast become lower and the keypoints location is not equal. Therefore SIFT algorithm cannot satisfy image matching with low image contrast. The threshold has to be adjusted for differient images manually and therefore cannot achieve automatic image matching. In order to improve the robustness and automation of SIFT based image matching algorithms, SIFT algorithm based on adaptive contrast threshold is proposed. The experimental results show that the proposed method is more robust to image contrast variation and can compute contrast threshold automatically.