Opencv中ORB算法如何保证尺度不变形

   在ORB算法论文中，并没有对尺度空间如何构建进行描述，但是Opencv内部构建了尺度空间，尺度空间层为8。

   当然这都是小问题，不知道你们有没有发现一个问题：若A图像尺度空间第一层上有一点a，与B图像尺度空间第三层上的一点b为同一匹配点，这时该如何匹配？

   或者说，该如何画出来？这两层所在尺度不同，进行筛选内点时该如何操作？

   先把Opencv内部的KeyPoint类贴出来：

class CV_EXPORTS_W_SIMPLE KeyPoint
{
public:
    //! the default constructor
    CV_WRAP KeyPoint() : pt(0,0), size(0), angle(-1), response(0), octave(0), class_id(-1) {}
    //! the full constructor
    KeyPoint(Point2f _pt, float _size, float _angle=-1,
            float _response=0, int _octave=0, int _class_id=-1)
            : pt(_pt), size(_size), angle(_angle),
            response(_response), octave(_octave), class_id(_class_id) {}
    //! another form of the full constructor
    CV_WRAP KeyPoint(float x, float y, float _size, float _angle=-1,
            float _response=0, int _octave=0, int _class_id=-1)
            : pt(x, y), size(_size), angle(_angle),
            response(_response), octave(_octave), class_id(_class_id) {}

    size_t hash() const;

    //! converts vector of keypoints to vector of points
    static void convert(const vector<KeyPoint>& keypoints,
                        CV_OUT vector<Point2f>& points2f,
                        const vector<int>& keypointIndexes=vector<int>());
    //! converts vector of points to the vector of keypoints, where each keypoint is assigned the same size and the same orientation
    static void convert(const vector<Point2f>& points2f,
                        CV_OUT vector<KeyPoint>& keypoints,
                        float size=1, float response=1, int octave=0, int class_id=-1);

    //! computes overlap for pair of keypoints;
    //! overlap is a ratio between area of keypoint regions intersection and
    //! area of keypoint regions union (now keypoint region is circle)
    static float overlap(const KeyPoint& kp1, const KeyPoint& kp2);

    CV_PROP_RW Point2f pt; //!< coordinates of the keypoints
    CV_PROP_RW float size; //!< diameter of the meaningful keypoint neighborhood
    CV_PROP_RW float angle; //!< computed orientation of the keypoint (-1 if not applicable);
                            //!< it's in [0,360) degrees and measured relative to
                            //!< image coordinate system, ie in clockwise.
    CV_PROP_RW float response; //!< the response by which the most strong keypoints have been selected. Can be used for the further sorting or subsampling
    CV_PROP_RW int octave; //!< octave (pyramid layer) from which the keypoint has been extracted
    CV_PROP_RW int class_id; //!< object class (if the keypoints need to be clustered by an object they belong to)
};

  这里，KeyPoint类中，对特征点所在的角度、位置、层数，都标出来了。
  但是，有没有发现使用drawMatches()函数和drawKeypoints()函数中，并没有对特征点所在层进行一个输入。

  例如在drawKeypoints()函数中Point center( cvRound(p.pt.x * draw_multiplier), cvRound(p.pt.y * draw_multiplier) )，直接把KeyPoint类转换成为了Point类。

  这是如何对齐的呢？

  重点来了！！！

  这是因为，在Opencv内部把不同尺度空间层上的点，进行一个处理，都乘上了降采样因子。这里的KeyPoint类上的octave只表明了该点是在该层上提取的。

  ORB中有getScale函数:

static inline float getScale(int level, int firstLevel, double scaleFactor)
{
    return (float)std::pow(scaleFactor, (double)(level - firstLevel));
}

先构建层度空间，每一层的大小为：

 float scale = 1/getScale(level, firstLevel, scaleFactor);
        Size sz(cvRound(image.cols*scale), cvRound(image.rows*scale));
        Size wholeSize(sz.width + border*2, sz.height + border*2);
        Mat temp(wholeSize, image.type()), masktemp;
        imagePyramid[level] = temp(Rect(border, border, sz.width, sz.height));

对每一层检测的点进行处理，从而对不同层检测的点还原到原图像上！

        if (level != firstLevel)
        {
            float scale = getScale(level, firstLevel, scaleFactor);
            for (vector<KeyPoint>::iterator keypoint = keypoints.begin(),
                 keypointEnd = keypoints.end(); keypoint != keypointEnd; ++keypoint)
                keypoint->pt *= scale;
        }
        // And add the keypoints to the output
        _keypoints.insert(_keypoints.end(), keypoints.begin(), keypoints.end());