OpenANN
1.1.0
An open source library for artificial neural networks.
|
Creates distorted images. More...
#include <Distorter.h>
Public Member Functions | |
Distorter (double sigma=5.0, double alpha=36.0/255.0, double beta=15.0, double gammaX=15.0, double gammaY=15.0) | |
void | createDistortionMap (int rows, int cols) |
void | applyDistortions (Eigen::MatrixXd &instances, int rows, int cols) |
void | applyDistortion (Eigen::VectorXd &instance, int rows, int cols) |
Public Attributes | |
double | sigma |
elastic distortion More... | |
double | alpha |
elastic scaling More... | |
double | beta |
maximal absolute rotation More... | |
double | gammaX |
maximal horizontal scaling (percent) More... | |
double | gammaY |
maximal vertical scaling (percent) More... | |
int | gaussianKernelSize |
has to be odd More... | |
Eigen::MatrixXd | gaussianKernel |
Eigen::MatrixXd | distortionH |
Eigen::MatrixXd | distortionV |
Creates distorted images.
Available distortions are:
To apply elastic distortions, the original image is distorted with random displacements and then convolved with a Gaussian kernel of width . Then the values are rescaled by the factor . Afterwards the image will be rotated by degrees and scaled horizontally by the factor and vertically by .
Source: http://www.codeproject.com/Articles/16650/Neural-Network-for-Recognition-of-Handwritten-Digi
|
inline |
|
inline |
|
inline |
|
inline |
double Distorter::alpha |
elastic scaling
double Distorter::beta |
maximal absolute rotation
Eigen::MatrixXd Distorter::distortionH |
Eigen::MatrixXd Distorter::distortionV |
double Distorter::gammaX |
maximal horizontal scaling (percent)
double Distorter::gammaY |
maximal vertical scaling (percent)
Eigen::MatrixXd Distorter::gaussianKernel |
int Distorter::gaussianKernelSize |
has to be odd
double Distorter::sigma |
elastic distortion