最近有个需求:拍摄证件或纸质文件上传时,需要自动将拍摄背景去除,只保留证件或文件那部分的图像。
先来一张效果图
首先使用opencv提供的CvVideoCamera类来加载视频流
实现CvVideoCameraDelegate的方法:
- (void)processImage:(Mat &)mat;
这个代理方法能实时获取摄像头输入的每一帧图像
- (void)processImage:(Mat &)mat
{
Mat src_gray, filtered, edges, dilated_edges;
//获取灰度图像
cvtColor(mat, src_gray, COLOR_BGR2GRAY);
//滤波,模糊处理,消除某些背景干扰信息
blur(src_gray, filtered, cv::Size(3, 3));
//腐蚀操作,消除某些背景干扰信息
erode(filtered, filtered, Mat(),cv::Point(-1, -1), 3, 1, 1);
int thresh = 35;
//边缘检测
Canny(filtered, edges, thresh, thresh*3, 3);
//膨胀操作,尽量使边缘闭合
dilate(edges, dilated_edges, Mat(), cv::Point(-1, -1), 3, 1, 1);
vector<vector<cv::Point> > contours, squares, hulls;
//寻找边框
findContours(dilated_edges, contours, RETR_LIST, CHAIN_APPROX_SIMPLE);
vector<cv::Point> hull, approx;
for (size_t i = 0; i < contours.size(); i++)
{
//边框的凸包
convexHull(contours[i], hull);
//多边形拟合凸包边框(此时的拟合的精度较低)
approxPolyDP(Mat(hull), approx, arcLength(Mat(approx), true)*0.02, true);
//筛选出面积大于某一阈值的,且四边形的各个角度都接近直角的凸四边形
if (approx.size() == 4 && fabs(contourArea(Mat(approx))) > 40000 &&
isContourConvex(Mat(approx)))
{
double maxCosine = 0;
for (int j = 2; j < 5; j++)
{
double cosine = fabs(getAngle(approx[j%4], approx[j-2], approx[j-1]));
maxCosine = MAX(maxCosine, cosine);
}
//角度大概72度
if (maxCosine < 0.3)
{
squares.push_back(approx);
hulls.push_back(hull);
}
}
}
vector<cv::Point> largest_square;
//找出外接矩形最大的四边形
int idex = findLargestSquare(squares, largest_square);
if (largest_square.size() == 0 || idex == -1) return;
//找到这个最大的四边形对应的凸边框,再次进行多边形拟合,此次精度较高,拟合的结果可能是大于4条边的多边形
//接下来的操作,主要是为了解决,证件有圆角时检测到的四个顶点的连线会有切边的问题
hull = hulls[idex];
approxPolyDP(Mat(hull), approx, 3, true);
vector<cv::Point> newApprox;
double maxL = arcLength(Mat(approx), true)*0.02;
//找到高精度拟合时得到的顶点中 距离小于 低精度拟合得到的四个顶点 maxL的顶点,排除部分顶点的干扰
for (cv::Point p : approx)
{
if (!(getSpacePointToPoint(p, largest_square[0]) > maxL &&
getSpacePointToPoint(p, largest_square[1]) > maxL &&
getSpacePointToPoint(p, largest_square[2]) > maxL &&
getSpacePointToPoint(p, largest_square[3]) > maxL))
{
newApprox.push_back(p);
}
}
//找到剩余顶点连线中,边长大于 2 * maxL的四条边作为四边形物体的四条边
vector<Vec4i> lines;
for (int i = 0; i < newApprox.size(); i++)
{
cv::Point p1 = newApprox[i];
cv::Point p2 = newApprox[(i+1)%newApprox.size()];
if (getSpacePointToPoint(p1, p2) > 2 * maxL)
{
lines.push_back(Vec4i(p1.x, p1.y, p2.x,p2.y));
}
}
//计算出这四条边中 相邻两条边的交点,即物体的四个顶点
vector<cv::Point> cornors1;
for (int i = 0; i < lines.size(); i++)
{
cv::Point cornor = computeIntersect(lines[i],lines[(i+1)%lines.size()]);
cornors1.push_back(cornor);
}
//绘制出四条边
for (int i = 0; i < cornors1.size(); i++)
{
line(mat, cornors1[i], cornors1[(i+1)%cornors1.size()], Scalar(0,0,255), 5);
}
}
相关自定义函数:
#pragma mark =========== 寻找最大边框 ===========
int findLargestSquare(const vector<vector<cv::Point> >& squares, vector<cv::Point>& biggest_square)
{
if (!squares.size()) return -1;
int max_width = 0;
int max_height = 0;
int max_square_idx = 0;
for (int i = 0; i < squares.size(); i++)
{
cv::Rect rectangle = boundingRect(Mat(squares[i]));
if ((rectangle.width >= max_width) && (rectangle.height >= max_height))
{
max_width = rectangle.width;
max_height = rectangle.height;
max_square_idx = i;
}
}
biggest_square = squares[max_square_idx];
return max_square_idx;
}
/**
根据三个点计算中间那个点的夹角 pt1 pt0 pt2
*/
double getAngle(cv::Point pt1, cv::Point pt2, cv::Point pt0)
{
double dx1 = pt1.x - pt0.x;
double dy1 = pt1.y - pt0.y;
double dx2 = pt2.x - pt0.x;
double dy2 = pt2.y - pt0.y;
return (dx1*dx2 + dy1*dy2)/sqrt((dx1*dx1 + dy1*dy1)*(dx2*dx2 + dy2*dy2) + 1e-10);
}
/**
点到点的距离
@param p1 点1
@param p2 点2
@return 距离
*/
double getSpacePointToPoint(cv::Point p1, cv::Point p2)
{
int a = p1.x-p2.x;
int b = p1.y-p2.y;
return sqrt(a * a + b * b);
}
/**
两直线的交点
@param a 线段1
@param b 线段2
@return 交点
*/
cv::Point2f computeIntersect(cv::Vec4i a, cv::Vec4i b)
{
int x1 = a[0], y1 = a[1], x2 = a[2], y2 = a[3], x3 = b[0], y3 = b[1], x4 = b[2], y4 = b[3];
if (float d = ((float)(x1 - x2) * (y3 - y4)) - ((y1 - y2) * (x3 - x4)))
{
cv::Point2f pt;
pt.x = ((x1 * y2 - y1 * x2) * (x3 - x4) - (x1 - x2) * (x3 * y4 - y3 * x4)) / d;
pt.y = ((x1 * y2 - y1 * x2) * (y3 - y4) - (y1 - y2) * (x3 * y4 - y3 * x4)) / d;
return pt;
}
else
return cv::Point2f(-1, -1);
}
/**
对多个点按顺时针排序
@param corners 点的集合
*/
void sortCorners(std::vector<cv::Point2f>& corners)
{
if (corners.size() == 0) return;
//先延 X轴排列
cv::Point pl = corners[0];
int index = 0;
for (int i = 1; i < corners.size(); i++)
{
cv::Point point = corners[i];
if (pl.x > point.x)
{
pl = point;
index = i;
}
}
corners[index] = corners[0];
corners[0] = pl;
cv::Point lp = corners[0];
for (int i = 1; i < corners.size(); i++)
{
for (int j = i+1; j<corners.size(); j++)
{
cv::Point point1 = corners[i];
cv::Point point2 = corners[j];
if ((point1.y-lp.y*1.0)/(point1.x-lp.x)>(point2.y-lp.y*1.0)/(point2.x-lp.x))
{
cv::Point temp = point1;
corners[i] = corners[j];
corners[j] = temp;
}
}
}
}
根据四边形的四个顶点,提取目标图像
//对顶点顺时针排序
sortCorners(_corners);
//计算目标图像的尺寸
cv::Point2f p0 = _corners[0];
cv::Point2f p1 = _corners[1];
cv::Point2f p2 = _corners[2];
cv::Point2f p3 = _corners[3];
float space0 = getSpacePointToPoint(p0, p1);
float space1 = getSpacePointToPoint(p1, p2);
float space2 = getSpacePointToPoint(p2, p3);
float space3 = getSpacePointToPoint(p3, p0);
float width = space1 > space3 ? space1 : space3;
float height = space0 > space2 ? space0 : space2;
cv::Mat quad = cv::Mat::zeros(height * 3, width * 3, CV_8UC3);
std::vector<cv::Point2f> quad_pts;
quad_pts.push_back(cv::Point2f(0, quad.rows));
quad_pts.push_back(cv::Point2f(0, 0));
quad_pts.push_back(cv::Point2f(quad.cols, 0));
quad_pts.push_back(cv::Point2f(quad.cols, quad.rows));
//提取图像
cv::Mat transmtx = cv::getPerspectiveTransform(_corners , quad_pts);
cv::warpPerspective(mat, quad, transmtx, quad.size());
如果调用getPerspectiveTransform方法崩溃,请参照我的另一篇文章 opencv 调用 getPerspectiveTransform 函数报错
最后可以利用 拉普拉斯算子可以增强局部的图像对比度,是图像更清晰
Mat imageMat;
Mat kernel = (Mat_<float>(3,3) << 0, -1, 0, -1, 5, -1, 0, -1, 0);
filter2D(quad, imageMat, quad.depth(), kernel);
//Mat --> UIImage
self.imageView.image = MatToUIImage(imageMat);