added more transforms

Josh Klontz
1 parent 94f8e680
Showing 4 changed files with 75 additions and 17 deletions
sdk/plugins/algorithms.cpp
sdk/plugins/cvt.cpp
sdk/plugins/hist.cpp
sdk/plugins/integral.cpp
@@ -50,7 +50,7 @@ class AlgorithmsInitializer : public Initializer
         Globals->abbreviations.insert("SmallSIFT", "Open+LimitSize(512)+KeyPointDetector(SIFT)+KeyPointDescriptor(SIFT):KeyPointMatcher(BruteForce)");
         Globals->abbreviations.insert("SmallSURF", "Open+LimitSize(512)+KeyPointDetector(SURF)+KeyPointDescriptor(SURF):KeyPointMatcher(BruteForce)");
         Globals->abbreviations.insert("ColorHist", "Open+LimitSize(512)!EnsureChannels(3)+SplitChannels+Hist(256,0,8)+Cat+Normalize(L1):L2");
-        Globals->abbreviations.insert("IHH", "Open+SplitChannels/(Cvt(Gray)+Gradient+Bin(0,6.283,8,true))+Integral+Merge+IntegralSampler");
+        Globals->abbreviations.insert("IHH", "Open+(RG+MAdd(0.5))/(Cvt(Gray)+Gradient+Bin(0,360,8,true))+Merge+Integral+IntegralSampler+CvtFloat:L2");
         // Hash
         Globals->abbreviations.insert("FileName", "Name+Identity:Identical");
@@ -160,6 +160,63 @@ class EnsureChannelsTransform : public UntrainableTransform
 BR_REGISTER(Transform, EnsureChannelsTransform)
+/*!
+ * \ingroup transforms
+ * \brief Normalized RG color space.
+ * \author Josh Klontz \cite jklontz
+ */
+class RGTransform : public UntrainableTransform
+{
+    Q_OBJECT
+
+    void project(const Template &src, Template &dst) const
+    {
+        if (src.m().type() != CV_8UC3)
+            qFatal("Expected CV_8UC3 images.");
+
+        const Mat &m = src.m();
+        Mat R(m.size(), CV_8UC1); // R / (R+G+B)
+        Mat G(m.size(), CV_8UC1); // G / (R+G+B)
+
+        for (int i=0; i<m.rows; i++)
+            for (int j=0; j<m.cols; j++) {
+                Vec3b v = m.at<Vec3b>(i,j);
+                uchar& b = v[0];
+                uchar& g = v[1];
+                uchar& r = v[2];
+
+                R.at<uchar>(i, j) = saturate_cast<uchar>(255.0*r/(r+g+b));
+                G.at<uchar>(i, j) = saturate_cast<uchar>(255.0*g/(r+g+b));
+            }
+
+        dst.append(R);
+        dst.append(G);
+    }
+};
+
+BR_REGISTER(Transform, RGTransform)
+
+/*!
+ * \ingroup transforms
+ * \brief dst = a*src+b
+ * \author Josh Klontz \cite jklontz
+ */
+class MAddTransform : public UntrainableTransform
+{
+    Q_OBJECT
+    Q_PROPERTY(double a READ get_a WRITE set_a RESET reset_a STORED false)
+    Q_PROPERTY(double b READ get_b WRITE set_b RESET reset_b STORED false)
+    BR_PROPERTY(double, a, 1)
+    BR_PROPERTY(double, b, 0)
+
+    void project(const Template &src, Template &dst) const
+    {
+        src.m().convertTo(dst.m(), src.m().depth(), a, b);
+    }
+};
+
+BR_REGISTER(Transform, MAddTransform)
+
 } // namespace br
 #include "cvt.moc"
@@ -83,13 +83,13 @@ class BinTransform : public UntrainableTransform
     void project(const Template &src, Template &dst) const
     {
-        src.m().convertTo(dst, bins > 256 ? CV_16U : CV_8U, bins/(max-min));
+        src.m().convertTo(dst, bins > 256 ? CV_16U : CV_8U, bins/(max-min), -0.5 /* floor */);
         if (!split) return;
         Mat input = dst;
         QList<Mat> outputs; outputs.reserve(bins);
         for (int i=0; i<bins; i++)
-            outputs.append(input == i);
+            outputs.append(input == i); // Note: Matrix elements are 0 or 255
         dst.clear(); dst.append(outputs);
     }
 };
@@ -2,6 +2,8 @@
 #include <Eigen/Core>
 #include <openbr_plugin.h>
+#include "core/opencvutils.h"
+
 using namespace cv;
 namespace br
@@ -26,25 +28,25 @@ BR_REGISTER(Transform, IntegralTransform)
 /*!
  * \ingroup transforms
- * \brief Sliding window object recognition from a multi-channel intergral image.
+ * \brief Sliding window feature extraction from a multi-channel intergral image.
  * \author Josh Klontz \cite jklontz
  */
-class IntegralSampler : public UntrainableTransform
+class IntegralSamplerTransform : public UntrainableTransform
 {
     Q_OBJECT
     Q_PROPERTY(int scales READ get_scales WRITE set_scales RESET reset_scales STORED false)
     Q_PROPERTY(float scaleFactor READ get_scaleFactor WRITE set_scaleFactor RESET reset_scaleFactor STORED false)
     Q_PROPERTY(float stepFactor READ get_stepFactor WRITE set_stepFactor RESET reset_stepFactor STORED false)
     Q_PROPERTY(int minSize READ get_minSize WRITE set_minSize RESET reset_minSize STORED false)
-    BR_PROPERTY(int, scales, 1)
-    BR_PROPERTY(float, scaleFactor, 1.25)
+    BR_PROPERTY(int, scales, 3)
+    BR_PROPERTY(float, scaleFactor, 2)
     BR_PROPERTY(float, stepFactor, 0.25)
     BR_PROPERTY(int, minSize, 8)
     void project(const Template &src, Template &dst) const
     {
         typedef Eigen::Map< const Eigen::Matrix<qint32,Eigen::Dynamic,1> > InputDescriptor;
-        typedef Eigen::Map< Eigen::Matrix<qint32,Eigen::Dynamic,1> > OutputDescriptor;
+        typedef Eigen::Map< Eigen::Matrix<float,Eigen::Dynamic,1> > OutputDescriptor;
         const Mat &m = src.m();
         if (m.depth() != CV_32S) qFatal("Expected CV_32S matrix depth.");
         const int channels = m.channels();
@@ -53,16 +55,16 @@ class IntegralSampler : public UntrainableTransform
         int descriptors = 0;
         int currentSize = min(m.rows, m.cols)-1;
         for (int scale=0; scale<scales; scale++) {
-            descriptors += ceil((m.rows-currentSize)*stepFactor/currentSize) *
-                    ceil((m.cols-currentSize)*stepFactor/currentSize);
+            descriptors += int(1+(m.rows-currentSize)/(currentSize*stepFactor)) *
+                           int(1+(m.cols-currentSize)/(currentSize*stepFactor));
             currentSize /= scaleFactor;
             if (currentSize < minSize)
                 break;
         }
-        Mat n(descriptors, channels, CV_32SC1);
+        Mat n(descriptors, channels, CV_32FC1);
         const qint32 *dataIn = (qint32*)m.data;
-        qint32 *dataOut = (qint32*)n.data;
+        float *dataOut = (float*)n.data;
         currentSize = min(m.rows, m.cols)-1;
         int index = 0;
         for (int scale=0; scale<scales; scale++) {
@@ -71,10 +73,10 @@ class IntegralSampler : public UntrainableTransform
                 for (int j=currentSize; j<m.cols; j+=currentStep) {
                     InputDescriptor a(dataIn+((i-currentSize)*rowStep+(j-currentSize)*channels), channels, 1);
                     InputDescriptor b(dataIn+((i-currentSize)*rowStep+ j             *channels), channels, 1);
-                    InputDescriptor c(dataIn+( i             *rowStep+(j-currentSize)*channels), channels, 1);
-                    InputDescriptor d(dataIn+( i             *rowStep+ j             *channels), channels, 1);
+                    InputDescriptor c(dataIn+(i              *rowStep+(j-currentSize)*channels), channels, 1);
+                    InputDescriptor d(dataIn+(i              *rowStep+ j             *channels), channels, 1);
                     OutputDescriptor y(dataOut+(index*channels), channels, 1);
-                    y = d-b-c+a;
+                    y = (d-b-c+a).cast<float>()/(currentSize*currentSize);
                     index++;
                 }
             }
@@ -82,12 +84,11 @@ class IntegralSampler : public UntrainableTransform
             if (currentSize < minSize)
                 break;
         }
-
         dst.m() = n;
     }
 };
-BR_REGISTER(Transform, IntegralSampler)
+BR_REGISTER(Transform, IntegralSamplerTransform)
 /*!
  * \ingroup transforms