view src/DistanceMetric.cpp @ 157:d6c1556fadd0 refactors

Default is actually Manhattan, not Euclidean (it just looks like squared-Euclidean for energy vectors). Add Euclidean as another alternative.

/* -*- c-basic-offset: 4 indent-tabs-mode: nil -*-  vi:set ts=8 sts=4 sw=4: */

/*
    Vamp feature extraction plugin using the MATCH audio alignment
    algorithm.

    Centre for Digital Music, Queen Mary, University of London.
    This file copyright 2007 Simon Dixon, Chris Cannam and QMUL.
    
    This program is free software; you can redistribute it and/or
    modify it under the terms of the GNU General Public License as
    published by the Free Software Foundation; either version 2 of the
    License, or (at your option) any later version.  See the file
    COPYING included with this distribution for more information.
*/

#include "DistanceMetric.h"

#include <cassert>
#include <cmath>
#include <iostream>

using namespace std;

//#define DEBUG_DISTANCE_METRIC 1

DistanceMetric::DistanceMetric(Parameters params) :
    m_params(params)
{
#ifdef DEBUG_DISTANCE_METRIC
    cerr << "*** DistanceMetric: norm = " << m_params.norm
         << endl;
#endif
}

double
DistanceMetric::calcDistance(const vector<double> &f1,
			     const vector<double> &f2)
{
    double d = 0;
    double sum = 0;
    double eps = 1e-16;

    int featureSize = f1.size();
    assert(int(f2.size()) == featureSize);

    if (m_params.metric == Cosine) {

        double num = 0, denom1 = 0, denom2 = 0;
        
        for (int i = 0; i < featureSize; ++i) {
            num += f1[i] * f2[i];
            denom1 += f1[i] * f1[i];
            denom2 += f2[i] * f2[i];
        }

        d = 1.0 - (num / (eps + sqrt(denom1 * denom2)));

        if (m_params.noise == AddNoise) {
            d += 1e-2;
        }
        if (d > 1.0) d = 1.0;
        
        return d; // normalisation param ignored

    }

    if (m_params.metric == Manhattan) {
        for (int i = 0; i < featureSize; i++) {
            d += fabs(f1[i] - f2[i]);
            sum += fabs(f1[i]) + fabs(f2[i]);
        }
    } else {
        // Euclidean
        for (int i = 0; i < featureSize; i++) {
            d += (f1[i] - f2[i]) * (f1[i] - f2[i]);
            sum += fabs(f1[i]) + fabs(f2[i]);
        }
        d = sqrt(d);
    }

    double noise = 1e-3 * featureSize;
    if (m_params.noise == AddNoise) {
        d += noise;
        sum += noise;
    }
    
    if (sum == 0) {
        return 0;
    }

    double distance = 0;

    if (m_params.norm == NormaliseDistanceToSum) {

        distance = d / sum; // 0 <= d/sum <= 2

    } else if (m_params.norm == NormaliseDistanceToLogSum) {

        // note if this were to be restored, it would have to use
        // totalEnergies vector instead of f1[freqMapSize] which used to
        // store the total energy:
        //	double weight = (5 + Math.log(f1[freqMapSize] + f2[freqMapSize]))/10.0;

        double weight = (8 + log(sum)) / 10.0;
    
        if (weight < 0) weight = 0;
        else if (weight > 1) weight = 1;

        distance = d / sum * weight;

    } else {

        distance = d;
    }
    
    return distance;
}