采用 SIFT算法匹配视频图像帧前，必须首先提取特征点。如果输入图像的大小和特性变化，特征点的灰度阈值必须随之重新设置，以避免过大的计算量和配准失败。提出了一种视频图像的特征点自适应提取算法。该算法能够将前一帧的视频图像的参数反馈到当前帧，自动设置适当的特征点灰度阈值，使得从当前帧提取的关键点的数量接近预期值。实验结果表明，当输入图像改变时，采用自适应设置阈值方法，从视频帧提取的特征点的数量始终保持在预期值。该方法可以通过 SIFT算法自适应地配准数字视频图像，使特征点数量保持稳定，避免配准失败，减小计算量。

Before matching the video frames in Scale-Invariant Feature Transform（SIFT）algorithm, the key-points must be extracted firstly. If the size and characteristic of input images are changed, gray threshold of key-points must be reinstalled, to avoid extremely computation cost or failure in registration. In this paper, a self-adaptive SIFT key-points extraction algorithm for video images is developed. The algorithm can set appropriate gray threshold of key-points automatically by feeding parameter of previous frame back to present frame to make the number of key-points extracting from present frame close to the expected value. The experiments show that, when the input image is changed, the key-points number of the video frame always keep near the expected value by setting the threshold self-adaptively. The method makes it possible for digital video images to be registered self-adaptively by SIFT algorithm and the number of feature points remains stable so that the computation costs can be reduced while avoiding registration failure.