PARP Research Group

Universidad de Murcia

---

examples/completeExample/completeExample.cpp Go to the documentation of this file. /* * Copyright (C) 2007, 2008, 2009, 2010, 2011, 2012. PARP Research Group. * <http://perception.inf.um.es> * University of Murcia, Spain. * * This file is part of the QVision library. * * QVision is free software: you can redistribute it and/or modify * it under the terms of the GNU Lesser General Public License as * published by the Free Software Foundation, version 3 of the License. * * QVision is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU Lesser General Public License for more details. * * You should have received a copy of the GNU Lesser General Public * License along with QVision. If not, see <http://www.gnu.org/licenses/>. */ #ifndef DOXYGEN_IGNORE_THIS // Do not generate any doxygen documentation from here on. /************************************** INCLUDE SECTION **************************************/ // Main QVApplication class (always needed): #include <QVApplication> // Video source reader block (virtually always needed): #include <QVVideoReaderBlock> // Main GUI, image and function plot windows: #include <QVDefaultGUI> #include <QVImageCanvas> #include <QVFunctionPlot> // Image processing and matrix functionality: #include <qvipp.h> #include <qvip.h> #include <qvmatrixalgebra.h> /************************************** CUSTOMIZED WINDOWS SECTION **************************************/ // First, you need to declare new types that are going to be used as properties: typedef QMap<sFloat,QPointF> TMapFloatPoints; Q_DECLARE_METATYPE(TMapFloatPoints); typedef QList<QLineF> TListLines; Q_DECLARE_METATYPE(TListLines); // Two new customized QVImageCanvas subclasses: one to draw float corner response image, with corresponding maximum // response points, and another to draw Canny image, with regression lines in selected squares: class ResponseImageCanvas : public QVImageCanvas { public: ResponseImageCanvas(QString name) : QVImageCanvas(name) { // Add extra properties to draw in canvas: addProperty<QMap<sFloat,QPointF> >("Max points", inputFlag); } ; void custom_viewer() { // Read properties to draw in canvas: QMap<sFloat,QPointF> maxPoints = getPropertyValue< QMap<sFloat,QPointF> >("Max points"); // Use getQVPainter() and standard QT QPainter class methods to draw in canvas: getQVPainter()->setPen(Qt::yellow); QMap<sFloat, QPointF>::iterator i = maxPoints.begin(); int order = 1; while (i != maxPoints.end()) { QPointF point = i.value() + QPointF(0.5,0.5); sFloat response = i.key(); getQVPainter()->drawLine(point+QPointF(-2,0),point+QPointF(+2,0)); getQVPainter()->drawLine(point+QPointF(0,-2),point+QPointF(0,+2)); getQVPainter()->drawTextUnscaled(point, QString("%1(%2)").arg(order).arg(-response)); i++; order++; } }; }; class CannyImageCanvas : public QVImageCanvas { public: CannyImageCanvas(QString name) : QVImageCanvas(name) { addProperty<int>("lr", inputFlag); addProperty< QList<QLineF> >("Regression lines", inputFlag); } ; void custom_viewer(){ QVPolyline poly = getPropertyValue<QVPolyline>("poly select"); QList<QLineF> regLines = getPropertyValue< QList<QLineF> >("Regression lines"); const int lr = getPropertyValue<int>("lr"); getQVPainter()->setPen(Qt::green); foreach(QPoint point,poly) getQVPainter()->drawRect(point.x()-lr,point.y()-lr,2*lr+1,2*lr+1); getQVPainter()->setPen(Qt::red); foreach(QLineF line,regLines) getQVPainter()->drawLine(line.translated(0.5,0.5)); }; }; /************************************** AUXILIARY FUNCTIONS **************************************/ // The following functions illustrate use of main data classes (QVImage, QVMatrix, QVVector, ...) and functions // (Intel IPP image processing functions, matrix algebra, ...). Native QT classes (like QMap, QList, QPoint, ...) are // also extensively used: QMap<sFloat,QPointF> findMaximalPoints(const QVImage<sFloat> &image, const double threshold, const int windowRadius) { QVImage<sFloat> maxImage; FilterMax(image, maxImage, QSize(2*windowRadius+1,2*windowRadius+1), // Mask size QPoint(0,0), // Anchor image.getROI().topLeft() + QPoint(windowRadius, windowRadius) // Destination ); const QRect ROI = maxImage.getROI(); const int maxStep = maxImage.getStep() / sizeof(sFloat), imageStep = image.getStep() / sizeof(sFloat); QMap<sFloat, QPointF> sortedPoints; sFloat *actualPtr = (sFloat *) image.getReadData() + (imageStep + 1) * windowRadius; sFloat *maxPtr = (sFloat *) maxImage.getReadData() + (maxStep + 1) * windowRadius; for(int j = 0; j < ROI.height(); j++, actualPtr += imageStep, maxPtr += maxStep) for(int i = 0; i < ROI.width(); i++) if ( (actualPtr[i] >= threshold) and (maxPtr[i] == actualPtr[i]) ) // We store negative response (QMap sorted in ascending order): sortedPoints.insertMulti(-actualPtr[i], QPointF(i+ROI.x(), j+ROI.y())); return sortedPoints; } QVImage<uChar> getCannyImage(const QVImage<uChar> imageGray, QRect CannyROI, const double thrLow, const double thrHigh) { uInt cols = imageGray.getCols(), rows = imageGray.getRows(); QVImage<sFloat> imageFloat(imageGray),dX(cols,rows), dY(cols,rows), dXNeg(cols,rows); QVImage<uChar> imageCanny(cols,rows), buffer; QVImage<uChar> imageDest(imageGray); imageFloat.setROI(CannyROI); imageCanny.setROI(CannyROI); FilterSobelHorizMask(imageFloat, dY, ippMskSize3x3); FilterSobelVertMask(imageFloat, dX, ippMskSize3x3); MulC(dX, -1, dXNeg); CannyGetSize(imageCanny, buffer); Canny(dXNeg, dY, imageCanny, qMin(thrLow,thrHigh) /* IPP Canny fails if thrLow > thrHigh */, thrHigh, buffer); Copy(imageCanny, imageDest, QPoint(CannyROI.x(),CannyROI.y())); imageDest.resetROI(); return imageDest; } QList<QLineF> getRegressionLines(const QVImage<uChar> &imageCanny, const QRect &CannyROI, const QVPolyline &CannySelPoints, int lr) { QList<QLineF> regLines; for(int k=0;k<CannySelPoints.size();k++) { QPointF mean(0,0); QList<QPointF> listPoints; QPointF point = CannySelPoints[k]; for(int j=point.y()-lr;j<=point.y()+lr;j++) for(int i=point.x()-lr;i<=point.x()+lr;i++) if(CannyROI.contains(i,j)) if(imageCanny(i,j) == 255) { listPoints.append(QPointF(i,j)); mean += QPointF(i,j); } const int n = listPoints.size(); if(n >= 2) { // Two points minimum to estimate regression line. mean /= n; QVMatrix M(n,2); // n x 2 matrix; for(int k=0;k<n;k++) { M(k,0) = listPoints[k].x() - mean.x(); M(k,1) = listPoints[k].y() - mean.y(); } QVMatrix covMat = M.transpose()*M / n, Q; QVVector lambda; eigenDecomposition(covMat,lambda,Q); QVVector mainEigenvector = Q.getCol(0); QPointF pointEV(mainEigenvector(0),mainEigenvector(1)); regLines.append(QLineF(mean+lr*pointEV,mean-lr*pointEV)); } } return regLines; } /************************************** PROCESSING BLOCK(S) **************************************/ // Our customized processing blocks inherit form QVProcessingBlock class. If we are not interested in multithreading, // we will normally use just one processing block: class MyProcessingBlock: public QVProcessingBlock { private: QVIndexedStringList filterList; // To select from list of strings with automatically generated combobox in GUI. public: MyProcessingBlock(QString name = QString()): QVProcessingBlock(name) { // Constructor // Add input parameters (bools, ints, doubles, ...). For all of them, corresponding GUI widgets and // read from command lines are automatically included: addProperty<bool>("box", inputFlag, false, "Apply box filter"); addProperty<int>("boxrows", inputFlag, 3, "Row box size", 1, 5); addProperty<int>("boxcols", inputFlag, 3, "Col box size", 1, 5); addProperty<double>("thhi", inputFlag, 150, "High threshold for Canny operator", 50, 1000); addProperty<double>("thlo", inputFlag, 50, "Low threshold for Canny operator", 10, 500); addProperty<double>("thMax", inputFlag, 50, "Max threshold for Hessian/Harris", 1.0, 255.0); addProperty<int>("lr", inputFlag|outputFlag, 5, "Radius for Canny regression lines", 2, 20); // Also comboboxes automatically generated in GUI from lists of strings (QVIndexedStringList properties): filterList.append("None"); filterList.append("Hessian"); filterList.append("Harris"); // firstFilter.append("..."); addProperty<QVIndexedStringList>("Filter", inputFlag, filterList, "Apply corner response filter"); // ROI rectangle and list of points to be linked as input from corresponding image window: addProperty<QRect>("CannyROI", inputFlag); addProperty<QVPolyline>("CannySelPoints", inputFlag); // One Input RGB image (3 channels), and two (one channel) output images (one of bytes, other of floats): addProperty< QVImage<uChar,3> >("Input RGB image", inputFlag|outputFlag); addProperty< QVImage<uChar,1> >("Output image byte", outputFlag); addProperty< QVImage<sFloat,1> >("Output image float", outputFlag); // One output property for list of maximum response points, regression lines and R, G, B histograms: addProperty< QMap<sFloat,QPointF> >("Max points", outputFlag); addProperty< QList<QLineF> >("Regression lines", outputFlag); addProperty< QList<double> >("R histogram", outputFlag); addProperty< QList<double> >("G histogram", outputFlag); addProperty< QList<double> >("B histogram", outputFlag); // An example trigger button to save a snapshot of current RGB image: addTrigger("Snapshot"); } // Generic callback method for trigger buttons: void processTrigger(const QString triggerName) { if(triggerName == "Snapshot") { QVImage<uchar,3> imageRGB = getPropertyValue< QVImage<uChar,3> >("Input RGB image"); QImage qimage = imageRGB; qimage.save("snapshot.jpg"); } // else if (triggerName == "...") } // Main iteration method for main processing block: void iterate() { // Get input RGB image: QVImage<uchar,3> imageRGB = getPropertyValue< QVImage<uChar,3> >("Input RGB image"); // Convert RGB image to gray: QVImage<uchar,1> imageGray = imageRGB; // Get current values of GUI parameters: const QVIndexedStringList selectedFilter = getPropertyValue<QVIndexedStringList>("Filter"); const bool applyBox = getPropertyValue<bool>("box"); const int boxRows = getPropertyValue<int>("boxrows"); const int boxCols = getPropertyValue<int>("boxcols"); const double thrLow = getPropertyValue<double>("thhi"); const double thrHigh = getPropertyValue<double>("thlo"); const double thMax = getPropertyValue<double>("thMax"); const int lr = getPropertyValue<int>("lr"); // Read GUI input linked from image windows: QRect CannyROI = getPropertyValue<QRect>("CannyROI"); QVPolyline CannySelPoints = getPropertyValue<QVPolyline>("CannySelPoints"); // Time statistics (put one after each stage of computation): timeFlag("Read input parameters"); // First, we compute the Canny operator on the byte image (ouput will be binary, 0 or 255 values): uInt cols = imageGray.getCols(), rows = imageGray.getRows(); // Define ROI for Canny: bool computeCanny = true; if (CannyROI == QRect()) { // User did not select rectangle: work in whole image: CannyROI = imageGray.getROI(); } else if(CannyROI.height()>=5 and CannyROI.width()>=5) { // If rectangle by user is big enough, define ROI accordingly: if(CannyROI.bottom() == static_cast<int>(rows)-1) CannyROI.setBottom(CannyROI.bottom()-1); // Solves IPP boundary problem in Canny function. } else { // Problematic ROI (too small ROI) for IPP Canny function: do not compute Canny image: computeCanny = false; } // Obtain Canny image (only in selected region of interest (ROI): QVImage<uChar> imageCanny; if(computeCanny) imageCanny = getCannyImage(imageGray,CannyROI,thrLow,thrHigh); else imageCanny = imageGray; timeFlag("Canny operator"); // Get regression lines on every input square marked by user: QList<QLineF> regLines = getRegressionLines(imageCanny,CannyROI,CannySelPoints,lr); timeFlag("Regression lines"); // If corresponding checkbox selected, previously apply box filter with requested box size values: if(applyBox) { // We need an auxiliary image, because IPP FilterBox is not an inplace operation: QVImage<uChar> imageGray2(cols,rows); FilterBox(imageGray, imageGray2, QSize(boxCols,boxRows), /*anchor*/ QPoint(0,0),/* destROIOffset */ QPoint(boxCols/2,boxRows/2)); imageGray = imageGray2; // Recover original ROI (which was reduced by FilterBox): imageGray.resetROI(); } timeFlag("Box filter"); // Apply selected response filter: QVImage<sFloat> imageFloat(cols,rows); int choice = selectedFilter.getIndex(); switch (choice) { case 0: { // None imageFloat = imageGray; break; } case 1: { // Hessian FilterHessianCornerResponseImage(imageGray, imageFloat, /* destROIOffset */ QPoint(2,2)); break; } case 2: { // Harris FilterHarrisCornerResponseImage(imageGray, imageFloat); // Underlying direct IPP call, no need for destROIoffset. break; } } timeFlag("Corner response filter"); // Recover original ROI (which was reduced by previous filters Hessian/Harris): imageFloat.resetROI(); // Normalize float image to get values in [0..255]: FilterNormalize(imageFloat, imageFloat); timeFlag("Normalization"); QMap<sFloat,QPointF> maxPoints; if(choice != 0) { const int radius = 3; maxPoints = findMaximalPoints(imageFloat, thMax, radius); } timeFlag("Get maximum response points"); // Histograms of R, G, B channels: QVImage<uChar> selected; Copy(imageRGB,0,selected); const QVector<int> RHistogramInt = HistogramRange(selected); Copy(imageRGB,1,selected); const QVector<int> GHistogramInt = HistogramRange(selected); Copy(imageRGB,2,selected); const QVector<int> BHistogramInt = HistogramRange(selected); QList<double> RHistogram,GHistogram,BHistogram; foreach(int val,RHistogramInt) RHistogram.append(val); foreach(int val,GHistogramInt) GHistogram.append(val); foreach(int val,BHistogramInt) BHistogram.append(val); timeFlag("Obtain RGB histograms"); // Finally, we set the output properties to their corresponding values: setPropertyValue< QVImage<uchar,1> >("Output image byte", imageCanny); setPropertyValue< QVImage<sFloat,1> >("Output image float", imageFloat); setPropertyValue< QMap<sFloat,QPointF> >("Max points", maxPoints); setPropertyValue< QList<QLineF> >("Regression lines", regLines); setPropertyValue< QList<double> >("R histogram", RHistogram); setPropertyValue< QList<double> >("G histogram", GHistogram); setPropertyValue< QList<double> >("B histogram", BHistogram); timeFlag("Write output parameters"); } }; // Main function always has the same structure: // 1. Create a QVApplication object. // 2. Create image source block(s) // 3. Create main processing block(s) (in our case, just one MyProcessingBlock). // 4. Create main GUI (QVDefaultGUI) and extra GUI elements (QVImageCanvas and subclasses,QVFunctionPlot, ...), // 5. Adequately link properties among objects of type 2-4. // 6. Call exec method on QVApplication. int main(int argc, char *argv[]) { // Application: QVApplication app(argc, argv, "Example program for QVision library."); // Image sources: QVVideoReaderBlock camera("Video"); // Processing blocks: MyProcessingBlock filterBlock("Image filter block"); // GUI elements: QVDefaultGUI interface; QVImageCanvas inputImage("Input RGB image"); CannyImageCanvas outputImageByte("Output image byte"); ResponseImageCanvas outputImageFloat("Output image float"); QVFunctionPlot RGBHistogramPlot("RGB histogram"); // Link properties: camera.linkProperty(&filterBlock,"Input RGB image"); filterBlock.linkProperty("Input RGB image", inputImage); filterBlock.linkProperty("Output image byte", outputImageByte); filterBlock.linkProperty("Regression lines", outputImageByte, "Regression lines"); filterBlock.linkProperty("Output image float", outputImageFloat); filterBlock.linkProperty("Max points", outputImageFloat,"Max points"); filterBlock.linkProperty("lr",outputImageByte,"lr"); filterBlock.linkProperty("R histogram",RGBHistogramPlot); filterBlock.linkProperty("G histogram",RGBHistogramPlot); filterBlock.linkProperty("B histogram",RGBHistogramPlot); outputImageByte.linkSelectedRectangle(filterBlock,"CannyROI"); outputImageByte.linkSelectedPolyline(filterBlock,"CannySelPoints"); // Execute application: return app.exec(); } #endif

---

QVision framework. PARP research group. Copyright © 2007, 2008, 2009, 2010, 2011.