Peter M. Groen / openbr

Commit 66160b306237c41a1d5a6dd8673254fbbe38df32

Authored by Scott Klum
1 parent a8dad531

Procrustes transform: R matrix calculated

Inline Side-by-side
Showing 2 changed files with 112 additions and 2 deletions
openbr/plugins/regions.cpp
  View file @66160b3
... ... @@ -201,9 +201,11 @@ class RectFromPointsTransform : public UntrainableTransform
201 201 Q_PROPERTY(QList<int> indices READ get_indices WRITE set_indices RESET reset_indices STORED false)
202 202 Q_PROPERTY(double padding READ get_padding WRITE set_padding RESET reset_padding STORED false)
203 203 Q_PROPERTY(double aspectRatio READ get_aspectRatio WRITE set_aspectRatio RESET reset_aspectRatio STORED false)
  204 + Q_PROPERTY(bool crop READ get_crop WRITE set_crop RESET reset_crop STORED false);
204 205 BR_PROPERTY(QList<int>, indices, QList<int>())
205 206 BR_PROPERTY(double, padding, 0)
206 207 BR_PROPERTY(double, aspectRatio, 1.0)
  208 + BR_PROPERTY(bool, crop, true)
207 209  
208 210 void project(const Template &src, Template &dst) const
209 211 {
... ... @@ -236,7 +238,8 @@ class RectFromPointsTransform : public UntrainableTransform
236 238 double deltaHeight = width/aspectRatio - height;
237 239 height += deltaHeight;
238 240  
239   - dst.m() = src.m()(Rect(std::max(0.0, minX - deltaWidth/2.0), std::max(0.0, minY - deltaHeight/2.0), std::min((double)src.m().cols, width), std::min((double)src.m().rows, height)));
  241 + if (crop) dst.m() = src.m()(Rect(std::max(0.0, minX - deltaWidth/2.0), std::max(0.0, minY - deltaHeight/2.0), std::min((double)src.m().cols, width), std::min((double)src.m().rows, height)));
  242 + else dst.m() = src.m();
240 243 }
241 244 };
242 245  
... ...
openbr/plugins/stasm4.cpp
  View file @66160b3
1 1 #include <stasm_lib.h>
2 2 #include <stasmcascadeclassifier.h>
3   -#include <opencv2/objdetect/objdetect.hpp>
  3 +#include <opencv2/opencv.hpp>
4 4 #include "openbr_internal.h"
  5 +#include "openbr/core/qtutils.h"
  6 +#include "openbr/core/opencvutils.h"
  7 +#include <QString>
  8 +#include <Eigen/SVD>
5 9  
6 10 using namespace cv;
7 11  
... ... @@ -79,6 +83,109 @@ class StasmTransform : public UntrainableTransform
79 83  
80 84 BR_REGISTER(Transform, StasmTransform)
81 85  
  86 +#include <iostream>
  87 +
  88 +/*!
  89 + * \ingroup transforms
  90 + * \brief Wraps STASM key point detector
  91 + * \author Scott Klum \cite sklum
  92 + */
  93 +class ProcrustesTransform : public Transform
  94 +{
  95 + Q_OBJECT
  96 +
  97 + Q_PROPERTY(QString principalShapePath READ get_principalShapePath WRITE set_principalShapePath RESET reset_principalShapePath STORED false)
  98 + BR_PROPERTY(QString, principalShapePath, QString())
  99 +
  100 + Eigen::MatrixXf meanShape;
  101 +
  102 + void train(const TemplateList &data)
  103 + {
  104 + QList< QList<cv::Point2f> > normalizedPoints;
  105 +
  106 + // Normalize all sets of points
  107 + foreach (br::Template datum, data) {
  108 + QList<cv::Point2f> points = OpenCVUtils::toPoints(datum.file.points());
  109 +
  110 + if (points.empty()) {
  111 + continue;
  112 + }
  113 +
  114 + cv::Scalar mean = cv::mean(points.toVector().toStdVector());
  115 + for (int i = 0; i < points.size(); i++) {
  116 + points[i].x -= mean[0];
  117 + points[i].y -= mean[1];
  118 + }
  119 +
  120 + float norm = cv::norm(points.toVector().toStdVector());
  121 + for (int i = 0; i < points.size(); i++) {
  122 + points[i].x /= norm;
  123 + points[i].y /= norm;
  124 + }
  125 +
  126 + normalizedPoints.append(points);
  127 + }
  128 +
  129 + // Determine mean shape
  130 + Eigen::MatrixXf shapeTest(normalizedPoints[0].size(), 2);
  131 +
  132 + cv::Mat shapeBuffer(normalizedPoints[0].size(), 2, CV_32F);
  133 +
  134 + for (int i = 0; i < normalizedPoints[0].size(); i++) {
  135 +
  136 + double x = 0;
  137 + double y = 0;
  138 +
  139 + for (int j = 0; j < normalizedPoints.size(); j++) {
  140 + x += normalizedPoints[j][i].x;
  141 + y += normalizedPoints[j][i].y;
  142 + }
  143 +
  144 + x /= (double)normalizedPoints.size();
  145 + y /= (double)normalizedPoints.size();
  146 +
  147 + shapeBuffer.at<float>(i,0) = x;
  148 + shapeBuffer.at<float>(i,1) = y;
  149 +
  150 + shapeTest(i,0) = x;
  151 + shapeTest(i,1) = y;
  152 + }
  153 +
  154 + meanShape = shapeTest;
  155 + }
  156 +
  157 + void project(const Template &src, Template &dst) const
  158 + {
  159 + QList<QPointF> points = src.file.points();
  160 +
  161 + cv::Scalar mean = cv::mean(OpenCVUtils::toPoints(points).toVector().toStdVector());
  162 +
  163 + for (int i = 0; i < points.size(); i++) points[i] -= QPointF(mean[0],mean[1]);
  164 +
  165 + float norm = cv::norm(OpenCVUtils::toPoints(points).toVector().toStdVector());
  166 +
  167 + Eigen::MatrixXf srcPoints(points.size(), 2);
  168 + for (int i = 0; i < points.size(); i++) {
  169 + srcPoints(i,0) = points[i].x()/norm;
  170 + srcPoints(i,1) = points[i].y()/norm;
  171 + }
  172 +
  173 + Eigen::JacobiSVD<Eigen::MatrixXf> svd(srcPoints.transpose()*meanShape, Eigen::ComputeThinU | Eigen::ComputeThinV);
  174 +
  175 + Eigen::MatrixXf R = svd.matrixU()*svd.matrixV().transpose();
  176 +
  177 + std::cout << R(1,0) << std::endl;
  178 + // Determine transformation matrix
  179 +
  180 + // Apply transformation matrix
  181 + //dst.file.setPoints(meanShape);*/
  182 + dst.m() = src.m();
  183 + }
  184 +
  185 +};
  186 +
  187 +BR_REGISTER(Transform, ProcrustesTransform)
  188 +
82 189 } // namespace br
83 190  
84 191 #include "stasm4.moc"
... ...