实验3 空域滤波
实验要求
掌握利用模板对图像进行空域滤波操作，熟练掌握常用空域模板的使用。
- 利用均值模板平滑灰度图像。 具体内容：利用OpenCV对图像像素进行操作，分别利用3 * 3、5 * 5和9 * 9尺寸的均值模板平滑灰度图像
- 利用高斯模板平滑灰度图像。 具体内容：利用OpenCV对图像像素进行操作，分别利用3 * 3、5 * 5和9 * 9尺寸的高斯模板平滑灰度图像
- 利用Laplacian、Robert、Sobel模板锐化灰度图像。 具体内容：利用OpenCV对图像像素进行操作，分别利用Laplacian、Robert、Sobel模板锐化灰度图像
- 利用高提升滤波算法增强灰度图像。 具体内容：利用OpenCV对图像像素进行操作，设计高提升滤波算法增强图像
- 利用均值模板平滑彩色图像。 具体内容：利用OpenCV分别对图像像素的RGB三个通道进行操作，利用3 * 3、5 * 5和9 * 9尺寸的均值模板平滑彩色图像
- 利用高斯模板平滑彩色图像。 具体内容：利用OpenCV分别对图像像素的RGB三个通道进行操作，分别利用3 * 3、5 * 5和9 * 9尺寸的高斯模板平滑彩色图像
- 利用Laplacian、Robert、Sobel模板锐化灰度图像。 具体内容：利用OpenCV分别对图像像素的RGB三个通道进行操作，分别利用Laplacian、Robert、Sobel模板锐化彩色图像
实验完成情况
包括完成的实验内容及每个实验的完成程度
完成了7项内容：
-
利用均值模板平滑灰度图像 需要特别说明的是 int FilterProcessing(Mat src, Mat dst, Mat filter, double ProcessingMethod(Mat filterArea, Mat filter)) 和 double LinearFilterCalc(Mat filterArea, Mat linearFilter) 这两个函数。FilterProcessing是用滤波器对图像进行运算的函数，其中的double ProcessingMethod(Mat filterArea, Mat filter)是回调函数，表示运算方式。本实验中的滤波方法都是线性运算LinearFilterCalc，即利用滤波器对图像进行卷积。但中值滤波等非线性滤波器就要用其他的运算方式，到时只要替换回掉函数即可。把FilterProcessing函数写成这样也是为了提高函数的通用性，方便以后的实验。 均值滤波分别用3 * 3、5 * 5和9 * 9尺寸的均值模板，均值滤波器很容易直接得到，直接将滤波器代入上面提到的两个函数即可。
double LinearFilterCalc(Mat filterArea, Mat linearFilter) { double result = 0; for (int y = 0; y < filterArea.rows; y++) { for (int x = 0; x < filterArea.cols; x++) { result += (double(filterArea.at<uchar>(y, x)))*(linearFilter.at<double>(y, x)); } } return result; } /* padding and scan */ int FilterProcessing( Mat src, Mat dst, Mat filter, double ProcessingMethod(Mat filterArea, Mat filter)) { Mat src_padding=src.clone(); Mat filterArea; int padding = (filter.rows - 1) / 2; //padding the border copyMakeBorder(src, src_padding, padding, padding, padding, padding, BORDER_REPLICATE); if (dst.type() == CV_8U) { for (int y = padding; y < src_padding.rows - padding; y++) { for (int x = padding; x < src_padding.cols - padding; x++) { filterArea = src_padding(Range(y - padding, y + padding + 1), Range(x - padding, x + padding + 1)); dst.at<uchar>(y - padding, x - padding) = int(ProcessingMethod(filterArea, filter) + 0.5); } } } else if (dst.type() == CV_64F) { for (int y = padding; y < src_padding.rows - padding; y++) { for (int x = padding; x < src_padding.cols - padding; x++) { filterArea = src_padding(Range(y - padding, y + padding + 1), Range(x - padding, x + padding + 1)); dst.at<double>(y - padding, x - padding) = ProcessingMethod(filterArea, filter); } } } else { cout << "type error" << endl; } return 0; } int MeanFilterProcessing() { //Mat countFilter_3x3 = (Mat_<double>(3, 3) << // 1, 2, 3, // 4, 5, 6, // 7, 8, 9); Mat meanFilter_3x3 = ((double)1 / 9)*Mat::ones(3, 3, CV_64F);//CV_64F对应double，若CV_32F对饮double会报错 Mat meanFilter_5x5 = ((double)1 / 25)*Mat::ones(5, 5, CV_64F); Mat meanFilter_9x9 = ((double)1 / 81)*Mat::ones(9, 9, CV_64F); Mat meanImg_3x3 = Mat::zeros(gray.size(), gray.type()); Mat meanImg_5x5 = Mat::zeros(gray.size(), gray.type()); Mat meanImg_9x9 = Mat::zeros(gray.size(), gray.type()); FilterProcessing(gray, meanImg_3x3, meanFilter_3x3, LinearFilterCalc); FilterProcessing(gray, meanImg_5x5, meanFilter_5x5, LinearFilterCalc); FilterProcessing(gray, meanImg_9x9, meanFilter_9x9, LinearFilterCalc); namedWindow("Gray - 灰度图像", WINDOW_AUTOSIZE); namedWindow("meanFilter_3x3", WINDOW_AUTOSIZE); namedWindow("meanFilter_5x5", WINDOW_AUTOSIZE); namedWindow("meanFilter_5x5", WINDOW_AUTOSIZE); imshow("Gray - 灰度图像", gray); imshow("meanFilter_3x3", meanImg_3x3); imshow("meanFilter_5x5", meanImg_5x5); imshow("meanFilter_9x9", meanImg_9x9); waitKey(0); destroyAllWindows(); return 0; } - 利用高斯模板平滑灰度图像
值得一提的是int GaussianFilterGenerator(Mat gaussianFilter, double variance)函数，这是我写的专门用来生成任意大小的高斯滤波器的函数。除模板矩阵外，一开始写的函数有幅值和方差两个参数，如果幅值不够大，对9 * 9的模板来说边缘会出现很多零。于是后来改成生成的高斯滤波器中最小值为1，如此一来就不要幅值参数，只需要一个方差参数就够了。 用GaussianFilterGenerator分别生成3 * 3、5 * 5和9 * 9尺寸的均值滤波器后，再代入上面提到的滤波处理函数得到滤波结果。int GaussianFilterGenerator(Mat gaussianFilter, double variance) { int filterSize = gaussianFilter.rows; if (filterSize % 2 != 1) { cout << "Bad Gaussian filter size" << endl; return -1; } //vector<int> center(2); int center = filterSize / 2; int sum = 0; int centerValue; centerValue = int((1. / exp(-0.5*(double(pow((0 - center), 2) + pow((0 - center), 2))) / variance)) + 0.5); for (int y = 0; y < gaussianFilter.rows; y++) { for (int x = 0; x < gaussianFilter.cols; x++) { gaussianFilter.at<double>(y, x) = int((double(centerValue)*exp(-0.5*(double(pow((y - center), 2) + pow((x - center), 2))) / variance)) + 0.5); sum += (int)gaussianFilter.at<double>(y, x); } } cout << filterSize << "x" << filterSize << " Gaussian Filter is : " << endl; cout << "1/" << sum << " * " << endl << gaussianFilter << endl; gaussianFilter = (1 / (double(sum)))*gaussianFilter; return 0; } int GaussianFilterProcessing() { Mat gaussianFilter_3x3 = Mat::zeros(3, 3, CV_64F);//CV_64F对应double，若CV_32F对应double会报错 Mat gaussianFilter_5x5 = Mat::zeros(5, 5, CV_64F); Mat gaussianFilter_9x9 = Mat::zeros(9, 9, CV_64F); GaussianFilterGenerator(gaussianFilter_3x3, 0.72);//0.720 is square of 0.849 GaussianFilterGenerator(gaussianFilter_5x5, 1.92); GaussianFilterGenerator(gaussianFilter_9x9, 2.8854); Mat gaussianImg_3x3 = Mat::zeros(gray.size(), gray.type()); Mat gaussianImg_5x5 = Mat::zeros(gray.size(), gray.type()); Mat gaussianImg_9x9 = Mat::zeros(gray.size(), gray.type()); FilterProcessing(gray, gaussianImg_3x3, gaussianFilter_3x3, LinearFilterCalc); FilterProcessing(gray, gaussianImg_5x5, gaussianFilter_5x5, LinearFilterCalc); FilterProcessing(gray, gaussianImg_9x9, gaussianFilter_9x9, LinearFilterCalc); namedWindow("Gray - 灰度图像", WINDOW_AUTOSIZE); namedWindow("gaussianFilter_3x3", WINDOW_AUTOSIZE); namedWindow("gaussianFilter_5x5", WINDOW_AUTOSIZE); namedWindow("gaussianFilter_5x5", WINDOW_AUTOSIZE); imshow("Gray - 灰度图像", gray); imshow("gaussianFilter_3x3", gaussianImg_3x3); imshow("gaussianFilter_5x5", gaussianImg_5x5); imshow("gaussianFilter_9x9", gaussianImg_9x9); waitKey(0); destroyAllWindows(); return 0; } - 利用Laplacian、Robert、Sobel模板锐化灰度图像
这三个锐化模板与之前的平滑模板不同，会出现超过[0,255]范围的结果。对于结算得到的边缘图像，需要标定一下。对于锐化结果，直接用saturate_cast函数封闭两边边界。只要数据类型没问题，计算方面跟上面的过程没有太大区别。int LaplacianFilterProcessing(Mat src, Mat dst, Mat laplacianFilter, Mat laplacianFilterImg, double c)//其实没必要多拆出来这一个函数 { FilterProcessing(gray, laplacianFilterImg, laplacianFilter, LinearFilterCalc); //计算并标定锐化结果，直接saturate_cast<uchar> for (int y = 0; y < gray.rows; y++) { for (int x = 0; x < gray.cols; x++) { dst.at<uchar>(y, x) = saturate_cast<uchar>(src.at<uchar>(y, x) + int(c*int(laplacianFilterImg.at<double>(y, x)) + 0.5));//int(double + 0.5)用来四舍五入 } } return 0; } int LaplacianSharpen(Mat src, Mat dst, string title, double c, int filterNum)//dst是锐化后的结果图像 { Mat laplacianFilter_n4 = (Mat_<double>(3, 3) << 0, 1, 0, 1, -4, 1, 0, 1, 0); Mat laplacianFilter_n8 = (Mat_<double>(3, 3) << 1, 1, 1, 1, -8, 1, 1, 1, 1); Mat laplacianFilter; string filterTitle; switch (filterNum) { case 1: laplacianFilter = laplacianFilter_n4; filterTitle = "1 -4 1"; break; case 2: laplacianFilter = laplacianFilter_n8; filterTitle = "1 -8 1"; break; default: break; } Mat laplacianFilterImg = Mat::zeros(src.size(), CV_64F);//滤波后得到的边缘图像 LaplacianFilterProcessing(src, dst, laplacianFilter, laplacianFilterImg, c); /*标定Laplacian滤波得到的边缘结果*/ //LinearTransProcessing(laplacianFilterImg, dst, 0, 255);//自己写的函数，不如自带的成员方法convertTo()好用 double maxLap, minLap; minMaxLoc(laplacianFilterImg, &minLap, &maxLap, 0, 0); laplacianFilterImg.convertTo(laplacianFilterImg, CV_8U, 255. / (maxLap - minLap), (-255.*minLap) / (maxLap - minLap)); if (title.compare("") != 0)//标题不空则显示结果 { namedWindow(title + "Laplacian Filter Img" + filterTitle, WINDOW_AUTOSIZE); imshow(title + "Laplacian Filter Img" + filterTitle, laplacianFilterImg); namedWindow(title + "Laplacian Sharpen Img" + filterTitle, WINDOW_AUTOSIZE); imshow(title + "Laplacian Sharpen Img" + filterTitle, dst); namedWindow(title, WINDOW_AUTOSIZE); imshow(title, src); waitKey(0); destroyAllWindows(); } return 0; } int RobertSharpen(Mat src, Mat dst, string title, double c) { Mat robertFilterImg = Mat::zeros(src.size(), CV_64F);//滤波后得到的边缘图像 for (int y = 0; y < src.rows - 1; y++) { for (int x = 0; x < src.cols - 1; x++) { robertFilterImg.at<double>(y, x) = abs(src.at<uchar>(y + 1, x + 1) - src.at<uchar>(y, x)) + abs(src.at<uchar>(y + 1, x) - src.at<uchar>(y, x + 1)); dst.at <uchar>(y, x) = saturate_cast<uchar>(src.at<uchar>(y, x) + int(c*robertFilterImg.at<double>(y, x) + 0.5));//int(double + 0.5)用来四舍五入 } } /*标定Rober滤波得到的边缘结果*/ double maxRob, minRob; minMaxLoc(robertFilterImg, &minRob, &maxRob, 0, 0); robertFilterImg.convertTo(robertFilterImg, CV_8U, 255. / (maxRob - minRob), (-255.*minRob) / (maxRob - minRob)); if (title != "")//标题不空则显示结果 { namedWindow(title + "Robert Filter Img", WINDOW_AUTOSIZE); imshow(title + "Robert Filter Img", robertFilterImg); namedWindow(title + "Robert Sharpen Img", WINDOW_AUTOSIZE); imshow(title + "Robert Sharpen Img", dst); namedWindow(title, WINDOW_AUTOSIZE); imshow(title, src); waitKey(0); destroyAllWindows(); } return 0; } int SobelSharpen(Mat src, Mat dst, string title, double c) { Mat sobelFilterImg = Mat::zeros(src.size(), CV_64F);//滤波后得到的边缘图像 Mat filterArea; Mat sobelFilter_x= (Mat_<double>(3, 3) << -1, 0, 1, -2, 0, 2, -1, 0, 1); Mat sobelFilter_y = (Mat_<double>(3, 3) << -1, -2, -1, 0, 0, 0, 1, 2, 1); for (int y = 1; y < src.rows - 1; y++) { for (int x = 1; x < src.cols - 1; x++) { filterArea = src(Range(y - 1, y + 1 + 1), Range(x - 1, x + 1 + 1)); sobelFilterImg.at<double>(y, x) = abs(LinearFilterCalc(filterArea, sobelFilter_x)) + abs(LinearFilterCalc(filterArea, sobelFilter_y)); dst.at <uchar>(y, x) = saturate_cast<uchar>(src.at<uchar>(y, x) + int(c*sobelFilterImg.at<double>(y, x) + 0.5));//int(double + 0.5)用来四舍五入 } } /*标定Sobel滤波得到的边缘结果*/ double maxSob, minSob; minMaxLoc(sobelFilterImg, &minSob, &maxSob, 0, 0); sobelFilterImg.convertTo(sobelFilterImg, CV_8U, 255. / (maxSob - minSob), (-255.*minSob) / (maxSob - minSob)); if (title != "")//标题不空则显示结果 { namedWindow(title + "Sobel Filter Img", WINDOW_AUTOSIZE); imshow(title + "Sobel Filter Img", sobelFilterImg); namedWindow(title + "Sobel Sharpen Img", WINDOW_AUTOSIZE); imshow(title + "Sobel Sharpen Img", dst); namedWindow(title, WINDOW_AUTOSIZE); imshow(title, src); waitKey(0); destroyAllWindows(); } return 0; } int SharpenFilterProcessing() { /*Laplacian*/ int lablacianFilterNum = 1;//1为中心为-4的laplacian模板，2为中心为-8的 double cLap = -1; Mat laplacianSharpenImg = Mat::zeros(gray.size(), gray.type()); /*Robert*/ double cRob = 1; Mat robertSharpenImg = Mat::zeros(gray.size(), gray.type()); /*Sobel*/ double cSob = 0.5; Mat sobelSharpenImg = Mat::zeros(gray.size(), gray.type()); char choice; while (1) { cout << "清选择锐化模板\n" << "1 Laplacian\n" << "2 Robert\n" << "3 Sobel\n" << "按w返回上一级，按q退出："; cin >> choice; if (choice == 'q') { exit(0); } else { switch (choice) { case '1': LaplacianSharpen(gray, laplacianSharpenImg, "灰度图像 ", cLap, lablacianFilterNum); break; case '2': RobertSharpen(gray, robertSharpenImg, "灰度图像 ", cRob);//系数取正会加强较亮的部分，系数取负会加强较暗的部分 break; case '3': SobelSharpen(gray, sobelSharpenImg, "灰度图像 ", cSob);//系数取正会加强较亮的部分，系数取负会加强较暗的部分 break; case 'w': return 0; default: cout << "无效的输入" << endl; break; } } } return 0; } -
利用高提升滤波算法增强灰度图像
用原图像减去平滑后的图像得到边缘，然后再用得到的边缘增强原图像。平滑方法提供了上面的均值和高斯两种方法共六种模板来选择。int HighboostFilterProcessing() { double k = 1.5;//k>1时为高提升滤波 Mat meanFilter_3x3 = ((double)1 / 9)*Mat::ones(3, 3, CV_64F);//CV_64F对应double，若CV_32F对饮double会报错 Mat meanFilter_5x5 = ((double)1 / 25)*Mat::ones(5, 5, CV_64F); Mat meanFilter_9x9 = ((double)1 / 81)*Mat::ones(9, 9, CV_64F); Mat gaussianFilter_3x3 = Mat::zeros(3, 3, CV_64F);//CV_64F对应double，若CV_32F对应double会报错 Mat gaussianFilter_5x5 = Mat::zeros(5, 5, CV_64F); Mat gaussianFilter_9x9 = Mat::zeros(9, 9, CV_64F); GaussianFilterGenerator(gaussianFilter_3x3, 0.72); GaussianFilterGenerator(gaussianFilter_5x5, 1.92); GaussianFilterGenerator(gaussianFilter_9x9, 2.8854); Mat filter; char choice; while (1) { cout << "清选择计算 unsharp mask 时所用的平滑滤波器\n" << "1 3x3 均值\n" << "2 5x5 均值\n" << "3 9x9 均值\n" << "4 3x3 高斯\n" << "5 5x5 高斯\n" << "6 9x9 高斯\n" << "按w返回上一级，按q退出："; cin >> choice; if (choice == 'q') { exit(0); } else { switch (choice) { case '1': filter = meanFilter_3x3; break; case '2': filter = meanFilter_5x5; break; case '3': filter = meanFilter_9x9; break; case '4': filter = gaussianFilter_3x3; break; case '5': filter = gaussianFilter_5x5; break; case '6': filter = gaussianFilter_9x9; break; case 'w': return 0; default: cout << "无效的输入" << endl; continue; //break; } } Mat highboostFilterImg = Mat::zeros(gray.size(), gray.type()); Mat unsharpMask= Mat::zeros(gray.size(), CV_64F);//非锐化掩蔽 Mat blurImg = Mat::zeros(gray.size(), gray.type()); FilterProcessing(gray, blurImg, filter, LinearFilterCalc); gray.convertTo(gray, CV_64F); blurImg.convertTo(blurImg, CV_64F); unsharpMask = gray - blurImg; gray.convertTo(gray, CV_8U); for (int y = 0; y < highboostFilterImg.rows; y++) { for (int x = 0; x < highboostFilterImg.cols; x++) { highboostFilterImg.at <uchar>(y, x) = saturate_cast<uchar>(gray.at<uchar>(y, x) + k*unsharpMask.at<double>(y, x)); } } //标定 scaling unsharp mask double maxMast, minMast; minMaxLoc(unsharpMask, &minMast, &maxMast, 0, 0); unsharpMask.convertTo(unsharpMask, CV_8U, 255. / (maxMast - minMast), (-255.*minMast) / (maxMast - minMast)); namedWindow("Gray - 灰度图像", WINDOW_AUTOSIZE); imshow("Gray - 灰度图像", gray); namedWindow("Unsharp Mask with scaling", WINDOW_AUTOSIZE); imshow("Unsharp Mask with scaling", unsharpMask); namedWindow("Highboost Filter Img", WINDOW_AUTOSIZE); imshow("Highboost Filter Img", highboostFilterImg); waitKey(0); destroyAllWindows(); } return 0; } - 利用均值模板平滑彩色图像BGR通道
三个通道分别平滑再合并。BGRFilterProcessing和ShowBGRChannels两个函数是专门为彩色图像处理写的函数。int BGRFilterProcessing( Mat src, Mat dst, Mat filter, Mat BGR_bChannel, Mat BGR_gChannel, Mat BGR_rChannel, double ProcessingMethod(Mat filterArea, Mat filter)) { vector<Mat> BGRchannels; split(image, BGRchannels); Mat bChannel, gChannel, rChannel; bChannel = BGRchannels.at(0); gChannel = BGRchannels.at(1); rChannel = BGRchannels.at(2); if (bChannel.size() != gray.size()) { cout << "size error !" << endl; return 0; } FilterProcessing(bChannel, BGR_bChannel, filter, ProcessingMethod); FilterProcessing(gChannel, BGR_gChannel, filter, ProcessingMethod); FilterProcessing(rChannel, BGR_rChannel, filter, ProcessingMethod); vector<Mat> BGRchannels_merge(3); BGRchannels_merge.at(0) = BGR_bChannel; BGRchannels_merge.at(1) = BGR_gChannel; BGRchannels_merge.at(2) = BGR_rChannel; merge(BGRchannels_merge, dst); return 0; } int ShowBGRChannels( Mat BGR_bChannel, Mat BGR_gChannel, Mat BGR_rChannel, string title) { vector<Mat> BGRchannels; split(image, BGRchannels); Mat bChannel, gChannel, rChannel; bChannel = BGRchannels.at(0); gChannel = BGRchannels.at(1); rChannel = BGRchannels.at(2); string bChannelTitle = "B channel:" + title; string gChannelTitle = "G channel:" + title; string rChannelTitle = "R channel:" + title; namedWindow("B channel", WINDOW_AUTOSIZE); imshow("B channel", bChannel); namedWindow(bChannelTitle, WINDOW_AUTOSIZE); imshow(bChannelTitle, BGR_bChannel); namedWindow("G channel", WINDOW_AUTOSIZE); imshow("G channel", gChannel); namedWindow(gChannelTitle, WINDOW_AUTOSIZE); imshow(gChannelTitle, BGR_gChannel); namedWindow("R channel", WINDOW_AUTOSIZE); imshow("R channel", rChannel); namedWindow(rChannelTitle, WINDOW_AUTOSIZE); imshow(rChannelTitle, BGR_rChannel); waitKey(0); destroyAllWindows(); return 0; } int BGRMeanFilterProcessing() { Mat meanFilter_3x3 = ((double)1 / 9)*Mat::ones(3, 3, CV_64F);//CV_64F对应double，若CV_32F对饮double会报错 Mat meanFilter_5x5 = ((double)1 / 25)*Mat::ones(5, 5, CV_64F); Mat meanFilter_9x9 = ((double)1 / 81)*Mat::ones(9, 9, CV_64F); Mat meanFilterImg_3x3 = Mat::zeros(image.size(), image.type());//这里一定要给三个结果对象分配内存空间，不然调用函数得到的结果无法保留 Mat meanFilterImg_5x5 = Mat::zeros(image.size(), image.type()); Mat meanFilterImg_9x9 = Mat::zeros(image.size(), image.type()); //Mat mean_bChannel_3x3 = Mat::zeros(gray.size(), CV_64F); Mat mean_bChannel_3x3 = Mat::zeros(gray.size(), gray.type()); Mat mean_gChannel_3x3 = Mat::zeros(gray.size(), gray.type()); Mat mean_rChannel_3x3 = Mat::zeros(gray.size(), gray.type()); Mat mean_bChannel_5x5 = Mat::zeros(gray.size(), gray.type()); Mat mean_gChannel_5x5 = Mat::zeros(gray.size(), gray.type()); Mat mean_rChannel_5x5 = Mat::zeros(gray.size(), gray.type()); Mat mean_bChannel_9x9 = Mat::zeros(gray.size(), gray.type()); Mat mean_gChannel_9x9 = Mat::zeros(gray.size(), gray.type()); Mat mean_rChannel_9x9 = Mat::zeros(gray.size(), gray.type()); BGRFilterProcessing( image, meanFilterImg_3x3, meanFilter_3x3, mean_bChannel_3x3, mean_gChannel_3x3, mean_rChannel_3x3, LinearFilterCalc); BGRFilterProcessing( image, meanFilterImg_5x5, meanFilter_5x5, mean_bChannel_5x5, mean_gChannel_5x5, mean_rChannel_5x5, LinearFilterCalc); BGRFilterProcessing( image, meanFilterImg_9x9, meanFilter_9x9, mean_bChannel_9x9, mean_gChannel_9x9, mean_rChannel_9x9, LinearFilterCalc); ShowBGRChannels(mean_bChannel_3x3, mean_gChannel_3x3, mean_rChannel_3x3, "Mean Filter 3x3"); ShowBGRChannels(mean_bChannel_5x5, mean_gChannel_5x5, mean_rChannel_5x5, "Mean Filter 5x5"); ShowBGRChannels(mean_bChannel_9x9, mean_gChannel_9x9, mean_rChannel_9x9, "Mean Filter 9x9"); namedWindow("原始图像", WINDOW_AUTOSIZE); imshow("原始图像", image); namedWindow("Mean Filter 3x3", WINDOW_AUTOSIZE); namedWindow("Mean Filter 5x5", WINDOW_AUTOSIZE); namedWindow("Mean Filter 9x9", WINDOW_AUTOSIZE); imshow("Mean Filter 3x3", meanFilterImg_3x3); imshow("Mean Filter 5x5", meanFilterImg_5x5); imshow("Mean Filter 9x9", meanFilterImg_9x9); waitKey(0); destroyAllWindows(); return 0; } - 利用高斯模板平滑彩色图像BGR通道
三个通道分别平滑再合并。int BGRGaussianFilterProcessing() { Mat gaussianFilter_3x3 = Mat::zeros(3, 3, CV_64F);//CV_64F对应double，若CV_32F对应double会报错 Mat gaussianFilter_5x5 = Mat::zeros(5, 5, CV_64F); Mat gaussianFilter_9x9 = Mat::zeros(9, 9, CV_64F); GaussianFilterGenerator(gaussianFilter_3x3, 0.72); GaussianFilterGenerator(gaussianFilter_5x5, 1.92); GaussianFilterGenerator(gaussianFilter_9x9, 2.8854); Mat gaussianFilterImg_3x3 = Mat::zeros(image.size(), image.type());//这里一定要给三个结果对象分配内存空间，不然调用函数得到的结果无法保留 Mat gaussianFilterImg_5x5 = Mat::zeros(image.size(), image.type()); Mat gaussianFilterImg_9x9 = Mat::zeros(image.size(), image.type()); Mat gaussian_bChannel_3x3 = Mat::zeros(gray.size(), gray.type()); Mat gaussian_gChannel_3x3 = Mat::zeros(gray.size(), gray.type()); Mat gaussian_rChannel_3x3 = Mat::zeros(gray.size(), gray.type()); Mat gaussian_bChannel_5x5 = Mat::zeros(gray.size(), gray.type()); Mat gaussian_gChannel_5x5 = Mat::zeros(gray.size(), gray.type()); Mat gaussian_rChannel_5x5 = Mat::zeros(gray.size(), gray.type()); Mat gaussian_bChannel_9x9 = Mat::zeros(gray.size(), gray.type()); Mat gaussian_gChannel_9x9 = Mat::zeros(gray.size(), gray.type()); Mat gaussian_rChannel_9x9 = Mat::zeros(gray.size(), gray.type()); BGRFilterProcessing( image, gaussianFilterImg_3x3, gaussianFilter_3x3, gaussian_bChannel_3x3, gaussian_gChannel_3x3, gaussian_rChannel_3x3, LinearFilterCalc); BGRFilterProcessing( image, gaussianFilterImg_5x5, gaussianFilter_5x5, gaussian_bChannel_5x5, gaussian_gChannel_5x5, gaussian_rChannel_5x5, LinearFilterCalc); BGRFilterProcessing( image, gaussianFilterImg_9x9, gaussianFilter_9x9, gaussian_bChannel_9x9, gaussian_gChannel_9x9, gaussian_rChannel_9x9, LinearFilterCalc); ShowBGRChannels(gaussian_bChannel_3x3, gaussian_gChannel_3x3, gaussian_rChannel_3x3, "Gaussian Filter 3x3"); ShowBGRChannels(gaussian_bChannel_5x5, gaussian_gChannel_5x5, gaussian_rChannel_5x5, "Gaussian Filter 5x5"); ShowBGRChannels(gaussian_bChannel_9x9, gaussian_gChannel_9x9, gaussian_rChannel_9x9, "Gaussian Filter 9x9"); namedWindow("原始图像", WINDOW_AUTOSIZE); namedWindow("Gaussian Filter 3x3", WINDOW_AUTOSIZE); namedWindow("Gaussian Filter 5x5", WINDOW_AUTOSIZE); namedWindow("Gaussian Filter 5x5", WINDOW_AUTOSIZE); imshow("原始图像", image); imshow("Gaussian Filter 3x3", gaussianFilterImg_3x3); imshow("Gaussian Filter 5x5", gaussianFilterImg_5x5); imshow("Gaussian Filter 9x9", gaussianFilterImg_9x9); waitKey(0); destroyAllWindows(); return 0; } - 利用Laplacian、Robert、Sobel模板锐化彩色图像BGR通道
三个通道分别锐化再合并。int BGRSharpenFilterProcessing() { vector<Mat> BGRchannels; split(image, BGRchannels); Mat bChannel, gChannel, rChannel; bChannel = BGRchannels.at(0); gChannel = BGRchannels.at(1); rChannel = BGRchannels.at(2); vector<Mat> BGRchannels_merge(3); /*Laplacian*/ int lablacianFilterNum = 1;//1为中心为-4的laplacian模板，2为中心为-8的 double cLap = -1; Mat laplacianSharpen_bChannel = Mat::zeros(gray.size(), gray.type()); Mat laplacianSharpen_gChannel = Mat::zeros(gray.size(), gray.type()); Mat laplacianSharpen_rChannel = Mat::zeros(gray.size(), gray.type()); Mat laplacianSharpenImg = Mat::zeros(image.size(), image.type()); /*Robert*/ double cRob = 1; Mat robertSharpen_bChannel = Mat::zeros(gray.size(), gray.type()); Mat robertSharpen_gChannel = Mat::zeros(gray.size(), gray.type()); Mat robertSharpen_rChannel = Mat::zeros(gray.size(), gray.type()); Mat robertSharpenImg = Mat::zeros(image.size(), image.type()); /*Sobel*/ double cSob = 0.5; Mat sobelSharpen_bChannel = Mat::zeros(gray.size(), gray.type()); Mat sobelSharpen_gChannel = Mat::zeros(gray.size(), gray.type()); Mat sobelSharpen_rChannel = Mat::zeros(gray.size(), gray.type()); Mat sobelSharpenImg = Mat::zeros(image.size(), image.type()); char choice; while (1) { cout << "清选择锐化模板\n" << "1 Laplacian\n" << "2 Robert\n" << "3 Sobel\n" << "按w返回上一级，按q退出："; cin >> choice; if (choice == 'q') { exit(0); } else { switch (choice) { case '1': LaplacianSharpen(bChannel, laplacianSharpen_bChannel, "B channel ", cLap, lablacianFilterNum); LaplacianSharpen(gChannel, laplacianSharpen_gChannel, "G channel ", cLap, lablacianFilterNum); LaplacianSharpen(rChannel, laplacianSharpen_rChannel, "R channel ", cLap, lablacianFilterNum); BGRchannels_merge.at(0) = laplacianSharpen_bChannel; BGRchannels_merge.at(1) = laplacianSharpen_gChannel; BGRchannels_merge.at(2) = laplacianSharpen_rChannel; merge(BGRchannels_merge, laplacianSharpenImg); namedWindow("Original - 原始图像", WINDOW_AUTOSIZE); imshow("Original - 原始图像", image); namedWindow("Laplacian Sharpen Img", WINDOW_AUTOSIZE); imshow("Laplacian Sharpen Img", laplacianSharpenImg); waitKey(0); destroyAllWindows(); break; case '2': RobertSharpen(bChannel, laplacianSharpen_bChannel, "B channel ", cRob); RobertSharpen(gChannel, laplacianSharpen_gChannel, "G channel ", cRob); RobertSharpen(rChannel, laplacianSharpen_rChannel, "R channel ", cRob); BGRchannels_merge.at(0) = laplacianSharpen_bChannel; BGRchannels_merge.at(1) = laplacianSharpen_gChannel; BGRchannels_merge.at(2) = laplacianSharpen_rChannel; merge(BGRchannels_merge, laplacianSharpenImg); namedWindow("Original - 原始图像", WINDOW_AUTOSIZE); imshow("Original - 原始图像", image); namedWindow("Robert Sharpen Img", WINDOW_AUTOSIZE); imshow("Robert Sharpen Img", laplacianSharpenImg); waitKey(0); destroyAllWindows(); break; case '3': SobelSharpen(bChannel, sobelSharpen_bChannel, "B channel ", cSob); SobelSharpen(gChannel, sobelSharpen_gChannel, "G channel ", cSob); SobelSharpen(rChannel, sobelSharpen_rChannel, "R channel ", cSob); BGRchannels_merge.at(0) = sobelSharpen_bChannel; BGRchannels_merge.at(1) = sobelSharpen_gChannel; BGRchannels_merge.at(2) = sobelSharpen_rChannel; merge(BGRchannels_merge, sobelSharpenImg); namedWindow("Original - 原始图像", WINDOW_AUTOSIZE); imshow("Original - 原始图像", image); namedWindow("Sobel Sharpen Img", WINDOW_AUTOSIZE); imshow("Sobel Sharpen Img", sobelSharpenImg); waitKey(0); destroyAllWindows(); break; case 'w': return 0; default: cout << "无效的输入" << endl; break; } } } return 0; }
实验中的问题
包括在实验中遇到的问题，以及解决问题的方法
-
仍然是关于数据类型的问题，凡是涉及计算过程和结果的量我都使用了double型，因此允许了小数和负值的相关计算，提高了函数的通用性，也稍微增加了一些运算代价。
-
FilterProcessing本来只能得到CV_8U的结果，后来做到锐化算子时发现不能适用了，于是函数内增加了数据类型判断和分别处理。
-
一开始写的时候处理彩色图像用了大段重复代码，看起来繁琐而且不容易编辑修改，于是整合了一些代码写了BGRFilterProcessing和ShowBGRChannels两个函数对彩色图像进行处理。不过现在看来BGRFilterProcessing函数的接口有一点过度设计，还不够好用。
-
大量的时间已不是花在数据结构和函数用法等初级问题上，而是对函数接口设计、算法的精确实现、性能改进等方面有了更多思考，努力追求结果的精确，为了一点小小的改进往往要花费大量时间以及思考和设计，因此本次实验总共写了八百多行代码。虽仍有不足之处，但经过反复确认不会出现错误。