Java tutorial
/* * This file is part of the LIRE project: http://www.semanticmetadata.net/lire * LIRE 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. * * LIRE is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License * along with LIRE; if not, write to the Free Software * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA * * We kindly ask you to refer the any or one of the following publications in * any publication mentioning or employing Lire: * * Lux Mathias, Savvas A. Chatzichristofis. Lire: Lucene Image Retrieval * An Extensible Java CBIR Library. In proceedings of the 16th ACM International * Conference on Multimedia, pp. 1085-1088, Vancouver, Canada, 2008 * URL: http://doi.acm.org/10.1145/1459359.1459577 * * Lux Mathias. Content Based Image Retrieval with LIRE. In proceedings of the * 19th ACM International Conference on Multimedia, pp. 735-738, Scottsdale, * Arizona, USA, 2011 * URL: http://dl.acm.org/citation.cfm?id=2072432 * * Mathias Lux, Oge Marques. Visual Information Retrieval using Java and LIRE * Morgan & Claypool, 2013 * URL: http://www.morganclaypool.com/doi/abs/10.2200/S00468ED1V01Y201301ICR025 * * Copyright statement: * -------------------- * (c) 2002-2013 by Mathias Lux (mathias@juggle.at) * http://www.semanticmetadata.net/lire, http://www.lire-project.net */ package net.semanticmetadata.lire.solr; import com.google.common.collect.Lists; import net.semanticmetadata.lire.imageanalysis.EdgeHistogram; import net.semanticmetadata.lire.imageanalysis.LireFeature; import net.semanticmetadata.lire.impl.SimpleResult; import net.semanticmetadata.lire.indexing.hashing.BitSampling; import net.semanticmetadata.lire.utils.ImageUtils; import org.apache.commons.codec.binary.Base64; import org.apache.lucene.document.Document; import org.apache.lucene.index.*; import org.apache.lucene.queries.TermsFilter; import org.apache.lucene.search.*; import org.apache.lucene.util.BytesRef; import org.apache.solr.common.params.SolrParams; import org.apache.solr.common.util.NamedList; import org.apache.solr.handler.RequestHandlerBase; import org.apache.solr.request.SolrQueryRequest; import org.apache.solr.response.SolrQueryResponse; import org.apache.solr.search.SolrIndexSearcher; import javax.imageio.ImageIO; import java.awt.image.BufferedImage; import java.io.IOException; import java.net.URL; import java.util.*; /** * This is the main LIRE RequestHandler for the Solr Plugin. It supports query by example using the indexed id, * an url or a feature vector. Furthermore, feature extraction and random selection of images are supported. * * @author Mathias Lux, mathias@juggle.at, 07.07.13 */ public class FastLireRequestHandler extends RequestHandlerBase { // private static HashMap<String, Class> fieldToClass = new HashMap<String, Class>(5); private long time = 0; private int countRequests = 0; private int defaultNumberOfResults = 60; /** * number of candidate results retrieved from the index. The higher this number, the slower, * the but more accurate the retrieval will be. 10k is a good value for starters. */ private int numberOfCandidateResults = 10000; private static final int DEFAULT_NUMBER_OF_CANDIDATES = 10000; /** * The number of query terms that go along with the TermsFilter search. We need some to get a * score, the less the faster. I put down a minimum of three in the method, this value gives * the percentage of the overall number used (selected randomly). */ private double numberOfQueryTerms = 0.33; private static final double DEFAULT_NUMBER_OF_QUERY_TERMS = 0.33; static { // one time hash function read ... try { BitSampling.readHashFunctions(); } catch (IOException e) { e.printStackTrace(); } } @Override public void init(NamedList args) { super.init(args); } /** * Handles three types of requests. * <ol> * <li>search by already extracted images.</li> * <li>search by an image URL.</li> * <li>Random results.</li> * </ol> * * @param req * @param rsp * @throws Exception */ @Override public void handleRequestBody(SolrQueryRequest req, SolrQueryResponse rsp) throws Exception { // (1) check if the necessary parameters are here if (req.getParams().get("hashes") != null) { // we are searching for hashes ... handleHashSearch(req, rsp); } else if (req.getParams().get("url") != null) { // we are searching for an image based on an URL handleUrlSearch(req, rsp); } else if (req.getParams().get("id") != null) { // we are searching for an image based on an URL handleIdSearch(req, rsp); } else if (req.getParams().get("extract") != null) { // we are trying to extract from an image URL. handleExtract(req, rsp); } else { // lets return random results. handleRandomSearch(req, rsp); } } /** * Handles the get parameters id, field and rows. * * @param req * @param rsp * @throws java.io.IOException * @throws InstantiationException * @throws IllegalAccessException */ private void handleIdSearch(SolrQueryRequest req, SolrQueryResponse rsp) throws IOException, InstantiationException, IllegalAccessException { SolrIndexSearcher searcher = req.getSearcher(); try { TopDocs hits = searcher.search(new TermQuery(new Term("id", req.getParams().get("id"))), 1); String paramField = "cl_ha"; if (req.getParams().get("field") != null) paramField = req.getParams().get("field"); LireFeature queryFeature = (LireFeature) FeatureRegistry.getClassForHashField(paramField).newInstance(); rsp.add("QueryField", paramField); rsp.add("QueryFeature", queryFeature.getClass().getName()); numberOfQueryTerms = req.getParams().getDouble("accuracy", DEFAULT_NUMBER_OF_QUERY_TERMS); numberOfCandidateResults = req.getParams().getInt("candidates", DEFAULT_NUMBER_OF_CANDIDATES); if (hits.scoreDocs.length > 0) { // Using DocValues to get the actual data from the index. // BinaryDocValues binaryValues = MultiDocValues.getBinaryValues(searcher.getIndexReader(), FeatureRegistry.getFeatureFieldName(paramField)); // *** # // if (binaryValues == null) // System.err.println("Could not find the DocValues of the query document. Are they in the index?"); // BytesRef bytesRef = new BytesRef(); SortedSetDocValues sortedSetValues = MultiDocValues.getSortedSetValues(searcher.getIndexReader(), paramField); long valueCount = sortedSetValues.getValueCount(); List<Term> termFilter = Lists.newArrayList(); for (int i = 0; i < valueCount; i++) { BytesRef v = new BytesRef(); sortedSetValues.lookupOrd(i, v); termFilter.add(new Term(paramField, v)); } int[] hashes = null; int paramRows = defaultNumberOfResults; if (req.getParams().getInt("rows") != null) paramRows = req.getParams().getInt("rows"); /** // bytesRef = binaryValues.get(hits.scoreDocs[0].doc); binaryValues.get(hits.scoreDocs[0].doc,bytesRef); // Document d = searcher.getIndexReader().document(hits.scoreDocs[0].doc); // String histogramFieldName = paramField.replace("_ha", "_hi"); queryFeature.setByteArrayRepresentation(bytesRef.bytes, bytesRef.offset, bytesRef.length); int paramRows = defaultNumberOfResults; if (req.getParams().getInt("rows") != null) paramRows = req.getParams().getInt("rows"); // Re-generating the hashes to save sgenerateHashespace (instead of storing them in the index) int[] hashes = BitSampling.generateHashes(queryFeature.getDoubleHistogram()); List<Term> termFilter = createTermFilter(hashes, paramField); **/ doSearch(req, rsp, searcher, paramField, paramRows, termFilter, createQuery(hashes, paramField, numberOfQueryTerms), queryFeature); } else { rsp.add("Error", "Did not find an image with the given id " + req.getParams().get("id")); } } catch (Exception e) { rsp.add("Error", "There was an error with your search for the image with the id " + req.getParams().get("id") + ": " + e.getMessage()); } } /** * Returns a random set of documents from the index. Mainly for testing purposes. * * @param req * @param rsp * @throws java.io.IOException */ private void handleRandomSearch(SolrQueryRequest req, SolrQueryResponse rsp) throws IOException { SolrIndexSearcher searcher = req.getSearcher(); DirectoryReader indexReader = searcher.getIndexReader(); double maxDoc = indexReader.maxDoc(); int paramRows = defaultNumberOfResults; if (req.getParams().getInt("rows") != null) paramRows = req.getParams().getInt("rows"); LinkedList list = new LinkedList(); while (list.size() < paramRows) { HashMap m = new HashMap(2); Document d = indexReader.document((int) Math.floor(Math.random() * maxDoc)); m.put("id", d.getValues("id")[0]); m.put("title", d.getValues("title")[0]); list.add(m); } rsp.add("docs", list); } /** * Searches for an image given by an URL. Note that (i) extracting image features takes time and * (ii) not every image is readable by Java. * * @param req * @param rsp * @throws java.io.IOException * @throws InstantiationException * @throws IllegalAccessException */ private void handleUrlSearch(SolrQueryRequest req, SolrQueryResponse rsp) throws IOException, InstantiationException, IllegalAccessException { SolrParams params = req.getParams(); String paramUrl = params.get("url"); String paramField = "cl_ha"; if (req.getParams().get("field") != null) paramField = req.getParams().get("field"); int paramRows = defaultNumberOfResults; if (params.get("rows") != null) paramRows = params.getInt("rows"); numberOfQueryTerms = req.getParams().getDouble("accuracy", DEFAULT_NUMBER_OF_QUERY_TERMS); numberOfCandidateResults = req.getParams().getInt("candidates", DEFAULT_NUMBER_OF_CANDIDATES); LireFeature feat = null; List<Term> termFilter = null; int[] hashes = null; // wrapping the whole part in the try try { BufferedImage img = ImageIO.read(new URL(paramUrl).openStream()); img = ImageUtils.trimWhiteSpace(img); // getting the right feature per field: if (paramField == null || FeatureRegistry.getClassForHashField(paramField) == null) // if the feature is not registered. feat = new EdgeHistogram(); else { feat = (LireFeature) FeatureRegistry.getClassForHashField(paramField).newInstance(); } feat.extract(img); hashes = BitSampling.generateHashes(feat.getDoubleHistogram()); termFilter = createTermFilter(hashes, paramField); } catch (Exception e) { rsp.add("Error", "Error reading image from URL: " + paramUrl + ": " + e.getMessage()); e.printStackTrace(); } // search if the feature has been extracted. if (feat != null) doSearch(req, rsp, req.getSearcher(), paramField, paramRows, termFilter, createQuery(hashes, paramField, numberOfQueryTerms), feat); } private void handleExtract(SolrQueryRequest req, SolrQueryResponse rsp) throws IOException, InstantiationException, IllegalAccessException { SolrParams params = req.getParams(); String paramUrl = params.get("extract"); String paramField = "cl_ha"; if (req.getParams().get("field") != null) paramField = req.getParams().get("field"); // int paramRows = defaultNumberOfResults; // if (params.get("rows") != null) // paramRows = params.getInt("rows"); LireFeature feat = null; // BooleanQuery query = null; // wrapping the whole part in the try try { BufferedImage img = ImageIO.read(new URL(paramUrl).openStream()); img = ImageUtils.trimWhiteSpace(img); // getting the right feature per field: if (paramField == null || FeatureRegistry.getClassForHashField(paramField) == null) // if the feature is not registered. feat = new EdgeHistogram(); else { feat = (LireFeature) FeatureRegistry.getClassForHashField(paramField).newInstance(); } feat.extract(img); rsp.add("histogram", Base64.encodeBase64String(feat.getByteArrayRepresentation())); int[] hashes = BitSampling.generateHashes(feat.getDoubleHistogram()); ArrayList<String> hashStrings = new ArrayList<String>(hashes.length); for (int i = 0; i < hashes.length; i++) { hashStrings.add(Integer.toHexString(hashes[i])); } Collections.shuffle(hashStrings); rsp.add("hashes", hashStrings); // just use 50% of the hashes for search ... // query = createTermFilter(hashes, paramField, 0.5d); } catch (Exception e) { // rsp.add("Error", "Error reading image from URL: " + paramUrl + ": " + e.getMessage()); e.printStackTrace(); } // search if the feature has been extracted. // if (feat != null) doSearch(rsp, req.getSearcher(), paramField, paramRows, query, feat); } /** * Search based on the given image hashes. * * @param req * @param rsp * @throws java.io.IOException * @throws IllegalAccessException * @throws InstantiationException */ private void handleHashSearch(SolrQueryRequest req, SolrQueryResponse rsp) throws IOException, IllegalAccessException, InstantiationException { SolrParams params = req.getParams(); SolrIndexSearcher searcher = req.getSearcher(); // get the params needed: // hashes=x y z ... // feature=<base64> // field=<cl_ha|ph_ha|...> String[] hashStrings = params.get("hashes").trim().split(" "); byte[] featureVector = Base64.decodeBase64(params.get("feature")); String paramField = "cl_ha"; if (req.getParams().get("field") != null) paramField = req.getParams().get("field"); int paramRows = defaultNumberOfResults; if (params.getInt("rows") != null) paramRows = params.getInt("rows"); numberOfQueryTerms = req.getParams().getDouble("accuracy", DEFAULT_NUMBER_OF_QUERY_TERMS); numberOfCandidateResults = req.getParams().getInt("candidates", DEFAULT_NUMBER_OF_CANDIDATES); // create boolean query: // System.out.println("** Creating query."); LinkedList<Term> termFilter = new LinkedList<Term>(); BooleanQuery query = new BooleanQuery(); for (int i = 0; i < hashStrings.length; i++) { // be aware that the hashFunctionsFileName of the field must match the one you put the hashes in before. hashStrings[i] = hashStrings[i].trim(); if (hashStrings[i].length() > 0) { termFilter.add(new Term(paramField, hashStrings[i].trim())); // System.out.println("** " + field + ": " + hashes[i].trim()); } } Collections.shuffle(termFilter); for (int k = 0; k < termFilter.size() * numberOfQueryTerms; k++) { query.add(new BooleanClause(new TermQuery(termFilter.get(k)), BooleanClause.Occur.SHOULD)); } // System.out.println("** Doing search."); // query feature LireFeature queryFeature = (LireFeature) FeatureRegistry.getClassForHashField(paramField).newInstance(); queryFeature.setByteArrayRepresentation(featureVector); // get results: doSearch(req, rsp, searcher, paramField, paramRows, termFilter, new MatchAllDocsQuery(), queryFeature); } /** * Actual search implementation based on (i) hash based retrieval and (ii) feature based re-ranking. * * @param rsp * @param searcher * @param hashFieldName the hash field name * @param maximumHits * @param terms * @param queryFeature * @throws java.io.IOException * @throws IllegalAccessException * @throws InstantiationException */ private void doSearch(SolrQueryRequest req, SolrQueryResponse rsp, SolrIndexSearcher searcher, String hashFieldName, int maximumHits, List<Term> terms, Query query, LireFeature queryFeature) throws IOException, IllegalAccessException, InstantiationException { // temp feature instance LireFeature tmpFeature = queryFeature.getClass().newInstance(); // Taking the time of search for statistical purposes. time = System.currentTimeMillis(); Filter filter = null; // if the request contains a filter: if (req.getParams().get("fq") != null) { // only filters with [<field>:<value> ]+ are supported StringTokenizer st = new StringTokenizer(req.getParams().get("fq"), " "); LinkedList<Term> filterTerms = new LinkedList<Term>(); while (st.hasMoreElements()) { String[] tmpToken = st.nextToken().split(":"); if (tmpToken.length > 1) { filterTerms.add(new Term(tmpToken[0], tmpToken[1])); } } if (filterTerms.size() > 0) filter = new TermsFilter(filterTerms); } TopDocs docs; // with query only. if (filter == null) { docs = searcher.search(query, numberOfCandidateResults); } else { docs = searcher.search(query, filter, numberOfCandidateResults); } // TopDocs docs = searcher.search(query, new TermsFilter(terms), numberOfCandidateResults); // with TermsFilter and boosting by simple query // TopDocs docs = searcher.search(new ConstantScoreQuery(new TermsFilter(terms)), numberOfCandidateResults); // just with TermsFilter time = System.currentTimeMillis() - time; rsp.add("RawDocsCount", docs.scoreDocs.length + ""); rsp.add("RawDocsSearchTime", time + ""); // re-rank time = System.currentTimeMillis(); TreeSet<SimpleResult> resultScoreDocs = new TreeSet<SimpleResult>(); float maxDistance = -1f; float tmpScore; String featureFieldName = FeatureRegistry.getFeatureFieldName(hashFieldName); // iterating and re-ranking the documents. BinaryDocValues binaryValues = MultiDocValues.getBinaryValues(searcher.getIndexReader(), featureFieldName); // *** # BytesRef bytesRef = new BytesRef(); for (int i = 0; i < docs.scoreDocs.length; i++) { // using DocValues to retrieve the field values ... binaryValues.get(docs.scoreDocs[i].doc, bytesRef); tmpFeature.setByteArrayRepresentation(bytesRef.bytes, bytesRef.offset, bytesRef.length); // Getting the document from the index. // This is the slow step based on the field compression of stored fields. // tmpFeature.setByteArrayRepresentation(d.getBinaryValue(name).bytes, d.getBinaryValue(name).offset, d.getBinaryValue(name).length); tmpScore = queryFeature.getDistance(tmpFeature); if (resultScoreDocs.size() < maximumHits) { // todo: There's potential here for a memory saver, think of a clever data structure that can do the trick without creating a new SimpleResult for each result. resultScoreDocs.add( new SimpleResult(tmpScore, searcher.doc(docs.scoreDocs[i].doc), docs.scoreDocs[i].doc)); maxDistance = resultScoreDocs.last().getDistance(); } else if (tmpScore < maxDistance) { // if it is nearer to the sample than at least one of the current set: // remove the last one ... resultScoreDocs.remove(resultScoreDocs.last()); // add the new one ... resultScoreDocs.add( new SimpleResult(tmpScore, searcher.doc(docs.scoreDocs[i].doc), docs.scoreDocs[i].doc)); // and set our new distance border ... maxDistance = resultScoreDocs.last().getDistance(); } } // System.out.println("** Creating response."); time = System.currentTimeMillis() - time; rsp.add("ReRankSearchTime", time + ""); LinkedList list = new LinkedList(); for (Iterator<SimpleResult> it = resultScoreDocs.iterator(); it.hasNext();) { SimpleResult result = it.next(); HashMap m = new HashMap(2); m.put("d", result.getDistance()); // add fields as requested: if (req.getParams().get("fl") == null) { m.put("id", result.getDocument().get("id")); if (result.getDocument().get("title") != null) m.put("title", result.getDocument().get("title")); } else { String fieldsRequested = req.getParams().get("fl"); if (fieldsRequested.contains("score")) { m.put("score", result.getDistance()); } if (fieldsRequested.contains("*")) { // all fields for (IndexableField field : result.getDocument().getFields()) { String tmpField = field.name(); if (result.getDocument().getFields(tmpField).length > 1) { m.put(result.getDocument().getFields(tmpField)[0].name(), result.getDocument().getValues(tmpField)); } else if (result.getDocument().getFields(tmpField).length > 0) { m.put(result.getDocument().getFields(tmpField)[0].name(), result.getDocument().getFields(tmpField)[0].stringValue()); } } } else { StringTokenizer st; if (fieldsRequested.contains(",")) st = new StringTokenizer(fieldsRequested, ","); else st = new StringTokenizer(fieldsRequested, " "); while (st.hasMoreElements()) { String tmpField = st.nextToken(); if (result.getDocument().getFields(tmpField).length > 1) { m.put(result.getDocument().getFields(tmpField)[0].name(), result.getDocument().getValues(tmpField)); } else if (result.getDocument().getFields(tmpField).length > 0) { m.put(result.getDocument().getFields(tmpField)[0].name(), result.getDocument().getFields(tmpField)[0].stringValue()); } } } } // m.put(field, result.getDocument().get(field)); // m.put(field.replace("_ha", "_hi"), result.getDocument().getBinaryValue(field)); list.add(m); } rsp.add("docs", list); // rsp.add("Test-name", "Test-val"); } @Override public String getDescription() { return "LIRE Request Handler to add images to an index and search them. Search images by id, by url and by extracted features."; } @Override public String getSource() { return "http://lire-project.net"; } @Override public NamedList<Object> getStatistics() { // Change stats here to get an insight in the admin console. NamedList<Object> statistics = super.getStatistics(); statistics.add("Number of Requests", countRequests); return statistics; } private BooleanQuery createQuery(int[] hashes, String paramField, double size) { List<Integer> hList = new ArrayList<Integer>(hashes.length); for (int i = 0; i < hashes.length; i++) { hList.add(hashes[i]); } Collections.shuffle(hList); BooleanQuery query = new BooleanQuery(); int numHashes = (int) Math.min(hashes.length, Math.floor(hashes.length * size)); // a minimum of 3 hashes ... if (numHashes < 3) numHashes = 3; for (int i = 0; i < numHashes; i++) { // be aware that the hashFunctionsFileName of the field must match the one you put the hashes in before. query.add(new BooleanClause(new TermQuery(new Term(paramField, Integer.toHexString(hashes[i]))), BooleanClause.Occur.SHOULD)); } // this query is just for boosting the results with more matching hashes. We'd need to match it to all docs. // query.add(new BooleanClause(new MatchAllDocsQuery(), BooleanClause.Occur.SHOULD)); return query; } /** * This is used to create a TermsFilter ... should be used to select in the index based on many terms. * We just need to integrate a minimum query too, else we'd not get the appropriate results. * * @param hashes * @param paramField * @return */ private List<Term> createTermFilter(int[] hashes, String paramField) { LinkedList<Term> termFilter = new LinkedList<Term>(); for (int i = 0; i < hashes.length; i++) { // be aware that the hashFunctionsFileName of the field must match the one you put the hashes in before. termFilter.add(new Term(paramField, Integer.toHexString(hashes[i]))); } return termFilter; } }