【圖像處理】透視變換 Perspective Transformationapp
透視變換(Perspective Transformation)是將圖片投影到一個新的視平面(Viewing Plane),也稱做投影映射(Projective Mapping)。通用的變換公式爲:less

u,v是原始圖片左邊,對應獲得變換後的圖片座標x,y,其中
。
變換矩陣
能夠拆成4部分,
表示線性變換,好比scaling,shearing和ratotion。
用於平移,
產生透視變換。因此能夠理解成仿射等是透視變換的特殊形式。通過透視變換以後的圖片一般不是平行四邊形(除非映射視平面和原來平面平行的狀況)。函數
重寫以前的變換公式能夠獲得:ui

因此,已知變換對應的幾個點就能夠求取變換公式。反之,特定的變換公式也能新的變換後的圖片。簡單的看一個正方形到四邊形的變換:
變換的4組對應點能夠表示成:
spa
根據變換公式獲得:.net

定義幾個輔助變量:orm

都爲0時變換平面與原來是平行的,能夠獲得:blog

不爲0時,獲得:圖片

求解出的變換矩陣就能夠將一個正方形變換到四邊形。反之,四邊形變換到正方形也是同樣的。因而,咱們經過兩次變換:四邊形變換到正方形+正方形變換到四邊形就能夠將任意一個四邊形變換到另外一個四邊形。ip

看一段代碼:
- PerspectiveTransform::PerspectiveTransform(float inA11, float inA21,
- float inA31, float inA12,
- float inA22, float inA32,
- float inA13, float inA23,
- float inA33) :
- a11(inA11), a12(inA12), a13(inA13), a21(inA21), a22(inA22), a23(inA23),
- a31(inA31), a32(inA32), a33(inA33) {}
-
- PerspectiveTransform PerspectiveTransform::quadrilateralToQuadrilateral(float x0, float y0, float x1, float y1,
- float x2, float y2, float x3, float y3, float x0p, float y0p, float x1p, float y1p, float x2p, float y2p,
- float x3p, float y3p) {
- PerspectiveTransform qToS = PerspectiveTransform::quadrilateralToSquare(x0, y0, x1, y1, x2, y2, x3, y3);
- PerspectiveTransform sToQ =
- PerspectiveTransform::squareToQuadrilateral(x0p, y0p, x1p, y1p, x2p, y2p, x3p, y3p);
- return sToQ.times(qToS);
- }
-
- PerspectiveTransform PerspectiveTransform::squareToQuadrilateral(float x0, float y0, float x1, float y1, float x2,
- float y2, float x3, float y3) {
- float dx3 = x0 - x1 + x2 - x3;
- float dy3 = y0 - y1 + y2 - y3;
- if (dx3 == 0.0f && dy3 == 0.0f) {
- PerspectiveTransform result(PerspectiveTransform(x1 - x0, x2 - x1, x0, y1 - y0, y2 - y1, y0, 0.0f,
- 0.0f, 1.0f));
- return result;
- } else {
- float dx1 = x1 - x2;
- float dx2 = x3 - x2;
- float dy1 = y1 - y2;
- float dy2 = y3 - y2;
- float denominator = dx1 * dy2 - dx2 * dy1;
- float a13 = (dx3 * dy2 - dx2 * dy3) / denominator;
- float a23 = (dx1 * dy3 - dx3 * dy1) / denominator;
- PerspectiveTransform result(PerspectiveTransform(x1 - x0 + a13 * x1, x3 - x0 + a23 * x3, x0, y1 - y0
- + a13 * y1, y3 - y0 + a23 * y3, y0, a13, a23, 1.0f));
- return result;
- }
- }
-
- PerspectiveTransform PerspectiveTransform::quadrilateralToSquare(float x0, float y0, float x1, float y1, float x2,
- float y2, float x3, float y3) {
-
- return squareToQuadrilateral(x0, y0, x1, y1, x2, y2, x3, y3).buildAdjoint();
- }
-
- PerspectiveTransform PerspectiveTransform::buildAdjoint() {
-
- PerspectiveTransform result(PerspectiveTransform(a22 * a33 - a23 * a32, a23 * a31 - a21 * a33, a21 * a32
- - a22 * a31, a13 * a32 - a12 * a33, a11 * a33 - a13 * a31, a12 * a31 - a11 * a32, a12 * a23 - a13 * a22,
- a13 * a21 - a11 * a23, a11 * a22 - a12 * a21));
- return result;
- }
-
- PerspectiveTransform PerspectiveTransform::times(PerspectiveTransform other) {
- PerspectiveTransform result(PerspectiveTransform(a11 * other.a11 + a21 * other.a12 + a31 * other.a13,
- a11 * other.a21 + a21 * other.a22 + a31 * other.a23, a11 * other.a31 + a21 * other.a32 + a31
- * other.a33, a12 * other.a11 + a22 * other.a12 + a32 * other.a13, a12 * other.a21 + a22
- * other.a22 + a32 * other.a23, a12 * other.a31 + a22 * other.a32 + a32 * other.a33, a13
- * other.a11 + a23 * other.a12 + a33 * other.a13, a13 * other.a21 + a23 * other.a22 + a33
- * other.a23, a13 * other.a31 + a23 * other.a32 + a33 * other.a33));
- return result;
- }
-
- void PerspectiveTransform::transformPoints(vector<float> &points) {
- int max = points.size();
- for (int i = 0; i < max; i += 2) {
- float x = points[i];
- float y = points[i + 1];
- float denominator = a13 * x + a23 * y + a33;
- points[i] = (a11 * x + a21 * y + a31) / denominator;
- points[i + 1] = (a12 * x + a22 * y + a32) / denominator;
- }
- }
對一張透視圖片變換回正面圖的效果:
- int main(){
- Mat img=imread("boy.png");
- int img_height = img.rows;
- int img_width = img.cols;
- Mat img_trans = Mat::zeros(img_height,img_width,CV_8UC3);
- PerspectiveTransform tansform = PerspectiveTransform::quadrilateralToQuadrilateral(
- 0,0,
- img_width-1,0,
- 0,img_height-1,
- img_width-1,img_height-1,
- 150,250,
- 771,0,
- 0,1023,
- 650,1023
- );
- vector<float> ponits;
- for(int i=0;i<img_height;i++){
- for(int j=0;j<img_width;j++){
- ponits.push_back(j);
- ponits.push_back(i);
- }
- }
- tansform.transformPoints(ponits);
- for(int i=0;i<img_height;i++){
- uchar* t= img_trans.ptr<uchar>(i);
- for (int j=0;j<img_width;j++){
- int tmp = i*img_width+j;
- int x = ponits[tmp*2];
- int y = ponits[tmp*2+1];
- if(x<0||x>(img_width-1)||y<0||y>(img_height-1))
- continue;
- uchar* p = img.ptr<uchar>(y);
- t[j*3] = p[x*3];
- t[j*3+1] = p[x*3+1];
- t[j*3+2] = p[x*3+2];
- }
- }
- imwrite("trans.png",img_trans);
- return 0;
- }
透視變換的原理和矩陣求解請參見前一篇《透視變換 Perspective Transformation》。在OpenCV中也實現了透視變換的公式求解和變換函數。
求解變換公式的函數:
- Mat getPerspectiveTransform(const Point2f src[], const Point2f dst[])
輸入原始圖像和變換以後的圖像的對應4個點,即可以獲得變換矩陣。以後用求解獲得的矩陣輸入perspectiveTransform即可以對一組點進行變換:
- void perspectiveTransform(InputArray src, OutputArray dst, InputArray m)
注意這裏src和dst的輸入並非圖像,而是圖像對應的座標。應用前一篇的例子,作個相反的變換:
- int main( )
- {
- Mat img=imread("boy.png");
- int img_height = img.rows;
- int img_width = img.cols;
- vector<Point2f> corners(4);
- corners[0] = Point2f(0,0);
- corners[1] = Point2f(img_width-1,0);
- corners[2] = Point2f(0,img_height-1);
- corners[3] = Point2f(img_width-1,img_height-1);
- vector<Point2f> corners_trans(4);
- corners_trans[0] = Point2f(150,250);
- corners_trans[1] = Point2f(771,0);
- corners_trans[2] = Point2f(0,img_height-1);
- corners_trans[3] = Point2f(650,img_height-1);
-
- Mat transform = getPerspectiveTransform(corners,corners_trans);
- cout<<transform<<endl;
- vector<Point2f> ponits, points_trans;
- for(int i=0;i<img_height;i++){
- for(int j=0;j<img_width;j++){
- ponits.push_back(Point2f(j,i));
- }
- }
-
- perspectiveTransform( ponits, points_trans, transform);
- Mat img_trans = Mat::zeros(img_height,img_width,CV_8UC3);
- int count = 0;
- for(int i=0;i<img_height;i++){
- uchar* p = img.ptr<uchar>(i);
- for(int j=0;j<img_width;j++){
- int y = points_trans[count].y;
- int x = points_trans[count].x;
- uchar* t = img_trans.ptr<uchar>(y);
- t[x*3] = p[j*3];
- t[x*3+1] = p[j*3+1];
- t[x*3+2] = p[j*3+2];
- count++;
- }
- }
- imwrite("boy_trans.png",img_trans);
-
- return 0;
- }
獲得變換以後的圖片:

注意這種將原圖變換到對應圖像上的方式會有一些沒有被填充的點,也就是右圖中黑色的小點。解決這種問題一是用差值的方式,再一種比較簡單就是不用原圖的點變換後對應找新圖的座標,而是直接在新圖上找反向變換原圖的點。提及來有點繞口,具體見前一篇《透視變換 Perspective Transformation》的代碼應該就能懂啦。
除了getPerspectiveTransform()函數,OpenCV還提供了findHomography()的函數,不是用點來找,而是直接用透視平面來找變換公式。這個函數在特徵匹配的經典例子中有用到,也很是直觀:
- int main( int argc, char** argv )
- {
- Mat img_object = imread( argv[1], IMREAD_GRAYSCALE );
- Mat img_scene = imread( argv[2], IMREAD_GRAYSCALE );
- if( !img_object.data || !img_scene.data )
- { std::cout<< " --(!) Error reading images " << std::endl; return -1; }
-
-
- int minHessian = 400;
- SurfFeatureDetector detector( minHessian );
- std::vector<KeyPoint> keypoints_object, keypoints_scene;
- detector.detect( img_object, keypoints_object );
- detector.detect( img_scene, keypoints_scene );
-
-
- SurfDescriptorExtractor extractor;
- Mat descriptors_object, descriptors_scene;
- extractor.compute( img_object, keypoints_object, descriptors_object );
- extractor.compute( img_scene, keypoints_scene, descriptors_scene );
-
-
- FlannBasedMatcher matcher;
- std::vector< DMatch > matches;
- matcher.match( descriptors_object, descriptors_scene, matches );
- double max_dist = 0; double min_dist = 100;
-
-
- for( int i = 0; i < descriptors_object.rows; i++ )
- { double dist = matches[i].distance;
- if( dist < min_dist ) min_dist = dist;
- if( dist > max_dist ) max_dist = dist;
- }
-
- printf("-- Max dist : %f \n", max_dist );
- printf("-- Min dist : %f \n", min_dist );
-
-
- std::vector< DMatch > good_matches;
-
- for( int i = 0; i < descriptors_object.rows; i++ )
- { if( matches[i].distance < 3*min_dist )
- { good_matches.push_back( matches[i]); }
- }
-
- Mat img_matches;
- drawMatches( img_object, keypoints_object, img_scene, keypoints_scene,
- good_matches, img_matches, Scalar::all(-1), Scalar::all(-1),
- vector<char>(), DrawMatchesFlags::NOT_DRAW_SINGLE_POINTS );
-
-
- std::vector<Point2f> obj;
- std::vector<Point2f> scene;
-
- for( size_t i = 0; i < good_matches.size(); i++ )
- {
-
- obj.push_back( keypoints_object[ good_matches[i].queryIdx ].pt );
- scene.push_back( keypoints_scene[ good_matches[i].trainIdx ].pt );
- }
-
- Mat H = findHomography( obj, scene, RANSAC );
-
-
- std::vector<Point2f> obj_corners(4);
- obj_corners[0] = Point(0,0); obj_corners[1] = Point( img_object.cols, 0 );
- obj_corners[2] = Point( img_object.cols, img_object.rows ); obj_corners[3] = Point( 0, img_object.rows );
- std::vector<Point2f> scene_corners(4);
- perspectiveTransform( obj_corners, scene_corners, H);
-
- Point2f offset( (float)img_object.cols, 0);
- line( img_matches, scene_corners[0] + offset, scene_corners[1] + offset, Scalar(0, 255, 0), 4 );
- line( img_matches, scene_corners[1] + offset, scene_corners[2] + offset, Scalar( 0, 255, 0), 4 );
- line( img_matches, scene_corners[2] + offset, scene_corners[3] + offset, Scalar( 0, 255, 0), 4 );
- line( img_matches, scene_corners[3] + offset, scene_corners[0] + offset, Scalar( 0, 255, 0), 4 );
-
-
- imshow( "Good Matches & Object detection", img_matches );
- waitKey(0);
- return 0;
- }
代碼運行效果:

findHomography()函數直接經過兩個平面上相匹配的特徵點求出變換公式,以後代碼又對原圖的四個邊緣點進行變換,在右圖上畫出對應的矩形。這個圖也很好地解釋了所謂透視變換的「Viewing Plane」。