人工智能项目资料-基于OpenCV已训练好的分类器进行人脸检测，并替换为目标人脸.zip

#include <iostream> #include <vector> #include <opencv2\imgproc\types_c.h> #include <opencv2\objdetect.hpp> #include <opencv2\features2d.hpp> #include <opencv2\calib3d.hpp> #include <opencv2\photo.hpp> #include "Histogram1D.hpp" using namespace cv; using namespace std; vector<Rect> DetectFace(Mat target); Mat Change2Gray(Mat ori); void HistSpecify(const Mat &src, const Mat &ref, Mat &result); void EqualizeImage(const Mat &src, Mat &dst); int main() { Mat img, img_gray; Mat target, target_gray; target = imread("target.jpg"); img = imread("ori.jpg"); //resize(target, target, Size(img.cols, img.rows)); //直方图规定化 HistSpecify(target, img, target); //获取灰度图 img_gray = Change2Gray(img); target_gray = Change2Gray(target); //检测两张图的脸部 vector<Rect> faces_1 = DetectFace(target_gray); vector<Rect> faces_2 = DetectFace(img_gray); Mat target_roi; if (faces_1.size() == 0) { cout << "目标图没有人脸"; getchar(); return -1; } //Debug:脸部画矩形标记 //rectangle(target, Point(faces_1[0].x, faces_1[0].y), Point(faces_1[0].x + faces_1[0].width, faces_1[0].y + faces_1[0].height), Scalar(0, 255, 0), 1, 8); Rect rect(faces_1[0].x, faces_1[0].y, faces_1[0].width, faces_1[0].height); if (faces_2.size() == 0) { cout << "原图没有人脸"; getchar(); return -1; } //Debug:脸部画矩形标记 //rectangle(img, Point(faces_2[0].x, faces_2[0].y), Point(faces_2[0].x + faces_2[0].width, faces_2[0].y + faces_2[0].height), Scalar(0, 255, 0), 1, 8); //截取目标脸部 target_roi = target(rect); //将目标脸部大小放缩到原图脸部大小 resize(target_roi, target_roi, Size(faces_2[0].width, faces_2[0].height)); //imshow("cut", target_roi); //创建遮罩 Mat mask = Mat::zeros(img.rows, img.cols, CV_8UC3); Mat src = Mat::zeros(img.rows, img.cols, CV_8UC3); Mat white = Mat::zeros(faces_2[0].width, faces_2[0].height, CV_8UC3); white.setTo(Scalar(256, 256, 256)); Rect roi_rect(faces_2[0].x, faces_2[0].y, faces_2[0].width, faces_2[0].height); white.copyTo(mask(roi_rect)); target_roi.copyTo(src(roi_rect)); //边界融合 Mat result = Mat::zeros(img.rows, img.cols, CV_8UC3); Point center(faces_2[0].x + (int)(faces_2[0].width / 2), faces_2[0].y + (int)(faces_2[0].height / 2)); seamlessClone(src, img, mask, center, result, NORMAL_CLONE); //高斯模糊 GaussianBlur(img, img, Size(5, 5), 0); //抠出区域 /*bitwise_not(mask, mask_turn); bitwise_and(img, mask_turn, img); target_roi.copyTo(mask(roi_rect)); add(mask, img, img); GaussianBlur(img, img, Size(9, 9), 1);*/ //imshow("mask", mask); int c = 0; imshow("target", target); imshow("ori", img); imshow("res", result); while (c != 27) { c = waitKey(1); if (c == 102) { flip(target_roi, target_roi, 1); target_roi.copyTo(src(roi_rect)); seamlessClone(src, img, mask, center, result, NORMAL_CLONE); imshow("target", target); imshow("ori", img); imshow("res", result); } } imwrite("output.jpg", result); return 0; } //检测人脸位置 vector<Rect> DetectFace(Mat target) { CascadeClassifier faceCascade; faceCascade.load("haarcascade_frontalface_alt.xml"); vector<Rect> faces; faceCascade.detectMultiScale(target, faces, 1.2, 6, 0, Size(0, 0)); return faces; } //获取灰度图像 Mat Change2Gray(Mat ori) { Mat gray; if (ori.channels() == 3) { cvtColor(ori, gray, CV_BGR2GRAY); } else { gray = ori; } return gray; } /** * @brief EqualizeImage 对灰度图像进行直方图均衡化 * @param src 输入图像 * @param dst 均衡化后的图像 */ void EqualizeImage(const Mat &src, Mat &dst) { Histogram1D hist1D; Mat hist = hist1D.getHistogram(src); hist /= (src.rows * src.cols); // 对得到的灰度直方图进行归一化,得到密度（0～1） float cdf[256] = { 0 }; // 灰度的累积概率 Mat lut(1, 256, CV_8U); // 创建用于灰度变换的查找表 for (int i = 0; i < 256; i++) { // 计算灰度级的累积概率 if (i == 0) { cdf[i] = hist.at<float>(i); } else { cdf[i] = cdf[i - 1] + hist.at<float>(i); } lut.at<uchar>(i) = static_cast<uchar>(255 * cdf[i]); // 创建灰度的查找表 } LUT(src, lut, dst); // 应用查找表，进行灰度变化，得到均衡化后的图像 } /** * @brief HistSpecify 对灰度图像进行直方图规定化 * @param src 输入图像 * @param ref 参考图像，解析参考图像的直方图并用于规定化 * @param result 直方图规定化后的图像 * @note 手动设置一个直方图并用于规定化比较麻烦，这里使用一个参考图像来进行 */ void HistSpecify(const Mat &src, const Mat &ref, Mat &result) { Histogram1D hist1D; Mat src_hist = hist1D.getHistogram(src); Mat dst_hist = hist1D.getHistogram(ref); float src_cdf[256] = { 0 }; float dst_cdf[256] = { 0 }; // 直方图进行归一化处理 src_hist /= (src.rows * src.cols); dst_hist /= (ref.rows * ref.cols); // 计算原始直方图和规定直方图的累积概率 for (int i = 0; i < 256; i++) { if (i == 0) { src_cdf[i] = src_hist.at<float>(i); dst_cdf[i] = dst_hist.at<float>(i); } else { src_cdf[i] = src_cdf[i - 1] + src_hist.at<float>(i); dst_cdf[i] = dst_cdf[i - 1] + dst_hist.at<float>(i); } } // 累积概率的差值 float diff_cdf[256][256]; for (int i = 0; i < 256; i++) { for (int j = 0; j < 256; j++) { diff_cdf[i][j] = fabs(src_cdf[i] - dst_cdf[j]); } } // 构建灰度级映射表 Mat lut(1, 256, CV_8U); for (int i = 0; i < 256; i++) { // 查找源灰度级为ｉ的映射灰度 //　和ｉ的累积概率差值最小的规定化灰度 float min = diff_cdf[i][0]; int index = 0; for (int j = 1; j < 256; j++) { if (min > diff_cdf[i][j]) { min = diff_cdf[i][j]; index = j; } } lut.at<uchar>(i) = static_cast<uchar>(index); } // 应用查找表，做直方图规定化 LUT(src, lut, result); }