SIFT算法大致有四个步骤:

1，尺度空间极值检测。在尺度空间通过高斯微分函数来检测潜在的对于尺度和旋转不变的兴趣点。

2，关键点定位。在兴趣点位置上，确定关键点的位置和尺度。3，方向确定。基于图像局部的梯度方向，给每个关键点分配方向。4，关键点描述。在每个关键点的领域内测量图像局部的梯度。最终用一个特征向量来表达。-SIFT algorithm roughly four steps:

1, the scale space extremum detection. In the scale space to detect potential points of interest for the scale and rotation invariant differential by a Gaussian function.

2, the key point positioning. In point of interest position, determine the location and scale of key points. 3, to determine the direction. Based on local gradient direction, the direction assigned to each key. 4, the key point descr iption. Local image gradient measurements within each key field. With a final feature vector to express.