001/**
002 * Copyright (c) 2011, The University of Southampton and the individual contributors.
003 * All rights reserved.
004 *
005 * Redistribution and use in source and binary forms, with or without modification,
006 * are permitted provided that the following conditions are met:
007 *
008 *   *  Redistributions of source code must retain the above copyright notice,
009 *      this list of conditions and the following disclaimer.
010 *
011 *   *  Redistributions in binary form must reproduce the above copyright notice,
012 *      this list of conditions and the following disclaimer in the documentation
013 *      and/or other materials provided with the distribution.
014 *
015 *   *  Neither the name of the University of Southampton nor the names of its
016 *      contributors may be used to endorse or promote products derived from this
017 *      software without specific prior written permission.
018 *
019 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
020 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
021 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
022 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
023 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
024 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
025 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
026 * ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
027 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
028 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
029 */
030package org.openimaj.feature.normalisation;
031
032import org.openimaj.feature.FloatFV;
033
034/**
035 * This {@link Normaliser} normalises vectors such that the Euclidean distance
036 * between normalised vectors is equivalent to computing the similarity using
037 * the Hellinger kernel on the un-normalised vectors.
038 * <p>
039 * The normalisation works by optionally adding an offset to the vectors (to
040 * deal with input vectors that have negative values), L1 normalising the
041 * vectors and finally performing an element-wise sqrt.
042 * 
043 * @author Jonathon Hare (jsh2@ecs.soton.ac.uk)
044 * 
045 */
046public class HellingerNormaliser implements Normaliser<FloatFV> {
047        protected int offset;
048
049        /**
050         * Construct with no offset
051         */
052        public HellingerNormaliser() {
053                this.offset = 0;
054        }
055
056        /**
057         * Construct with the given offset
058         * 
059         * @param offset
060         *            the offset
061         */
062        public HellingerNormaliser(int offset) {
063                this.offset = offset;
064        }
065
066        @Override
067        public void normalise(FloatFV feature) {
068                normalise(feature.values, offset);
069        }
070
071        /**
072         * Static utility function to perform Hellinger normalisation.
073         * 
074         * @param values
075         *            the values to normalise
076         * @param offset
077         *            the offset to add to the values before normalisation (to
078         *            ensure they are +ve).
079         */
080        public static void normalise(float[] values, int offset) {
081                double sum = 0;
082
083                for (int i = 0; i < values.length; i++) {
084                        values[i] += offset;
085                        sum += values[i];
086                }
087
088                for (int i = 0; i < values.length; i++) {
089                        values[i] = (float) Math.sqrt(values[i] / sum);
090                }
091        }
092}