Extraction

Fingerprint extraction module for DBOX

This library is a continuation of Preprocessing module, it fully supports the outputs from that library.

Dependencies:

The mentioned or newer versions are recommended

Getting Started:

You need to provide valid paths to these libraries and their header files in .pro file.
Build and run the project to generate .so (.dll / .lib) files
Include the library and header files to your own application
Copy the 'core' folder to your root project directory

This library supports QThread

API

Required

int loadInput(cv::Mat imgOriginal, cv::Mat imgSkeleton, cv::Mat orientationMap, int fpQuality = 100, cv::Mat qualityMap = cv::Mat(0,0,CV_8UC1), cv::Mat imgSkeletonInverted = cv::Mat(0,0,CV_8UC1));
int loadInput(PREPROCESSING_RESULTS preprocessingResults);  
int loadInput(QMap<QString, PREPROCESSING_RESULTS> preprocessingResults);  
  
void start();

Usage:

Load the input parameters, they can be added straight from the Preprocessing library

If you want to add the input parameters manually:
- If you want to use the Orientation Fixer you have to load the inverted skeleton map
- If you want to have quality values for each minutia you have to load a quality map

Just call start()

Optional:

int setExtractionParams(CAFFE_FILES extractionFiles, int extractionBlockSize);  
  
int setFeatures(bool useISOConverter, bool useOrientationFixer = true, bool useVariableBlockSize = false);  
  
int setCPUOnly(bool enabled);

Usage:

With setExtractionParams(...) you can set the Caffe model files and parameteres required for classification with neural network and the block size used during the classification
With setFeatures(...) you can set some optional features:
- you can convert the founded minutia to ISO/IEC 19794-2:2005 format (important notice: after use you need to free up the memory of the ISO templates, because they are dynamic arrays)
- you can use Orientation Fixer to get minutiae angles from 0 to 360 degrees instead of 0 to 180 degrees
- you can use Variable Block Size during classification, in some cases it can be effective
With setCPUOnly(...) you can force the library to use the CPU during classification (notice: it can be slower than the GPU)

SIGNALS:

void extractionDoneSignal(EXTRACTION_RESULTS results);  
  
void extractionDoneSignal(QVector<MINUTIA> minutiae);  
  
void extractionDoneSignal(unsigned char* minutiaeISO);  
  
void extractionSequenceDoneSignal(QMap<QString, EXTRACTION_RESULTS> results);

void extractionSequenceDoneSignal(QMap<QString, QVector<MINUTIA>> minutiaeMap);  

void extractionSequenceDoneSignal(QMap<QString, unsigned char *> minutiaeISO);  
  
void extractionDurationsSignal(EXTRACTION_DURATIONS durations);  
  
void extractionErrorSignal(int errorCode);

Important notice:

The first three signals are emitted only if you loaded one input
The next three signals are emitted only if you loaded a sequence of input
Signals with ISO templates are emmitted only if you activated that in setFeatures(...) (remind: do not forget to free up the memory)
You get extractionDurationSignal with duration values in ms for each phase during extraction if it's finished successfully
You get extractionErrorSignal with the error code if an error occured during preprocessing

A simple example how to use signals in your application
yourclass.h:

#include "extraction.h"

class YourClass: public QObject
{
    Q_OBJECT  
  
private:  
    Extraction e;  
    
private slots:
    void extractionResultsSlot(EXTRACTION_RESULTS result);
}

yourclass.cpp:

#include "yourclass.h"
YourClass::YourClass()
{
    qRegisterMetaType<EXTRACTION_RESULTS >("EXTRACTION_RESULTS");
    connect(&e, SIGNAL(extractionDoneSignal(EXTRACTION_RESULTS)), this, SLOT(extractionResultsSlot(EXTRACTION_RESULTS)));
}

void YourClass::extractionResultsSlot(EXTRACTION_RESULTS result)
{
    ...
}

For more please visit Qt Signals & Slots.

stupel/Extraction

Extraction

API