/**
* Cluster the blocks following the font, style and size aspects
*
* <p>-> not used at this stage, but could be an interesting feature in the full text model in the
* future
*
* @param b integer
* @param doc a document
*/
private static void addBlockToCluster(Integer b, Document doc) {
// get block features
Block block = doc.getBlocks().get(b);
String font = block.getFont();
boolean bold = block.getBold();
boolean italic = block.getItalic();
double fontSize = block.getFontSize();
boolean found = false;
if (font == null) {
font = "unknown";
}
// System.out.println(font + " " + bold + " " + italic + " " + fontSize );
if (doc.getClusters() == null) {
doc.setClusters(new ArrayList<Cluster>());
} else {
for (Cluster cluster : doc.getClusters()) {
String font2 = cluster.getFont();
if (font2 == null) font2 = "unknown";
if (font.equals(font2)
&& (bold == cluster.getBold())
& (italic == cluster.getItalic())
& (fontSize == cluster.getFontSize())) {
cluster.addBlock2(b);
found = true;
}
}
}
if (!found) {
Cluster cluster = new Cluster();
cluster.setFont(font);
cluster.setBold(bold);
cluster.setItalic(italic);
cluster.setFontSize(fontSize);
cluster.addBlock2(b);
doc.getClusters().add(cluster);
}
}
/**
* Filter out line numbering possibly present in the document. This can be frequent for document
* in a review/submission format and degrades strongly the machine learning extraction results.
*
* <p>-> Not used !
*
* @param doc a document
* @return if found numbering
*/
public boolean filterLineNumber(Document doc) {
// we first test if we have a line numbering by checking if we have an increasing integer
// at the begin or the end of each block
boolean numberBeginLine = false;
boolean numberEndLine = false;
boolean foundNumbering = false;
int currentNumber = -1;
int lastNumber = -1;
int i = 0;
for (Block block : doc.getBlocks()) {
// Integer ii = i;
String localText = block.getText();
List<LayoutToken> tokens = block.tokens;
if ((localText != null) && (tokens != null)) {
if (tokens.size() > 0) {
// we get the first and last token iof the block
// String tok1 = tokens.get(0).getText();
// String tok2 = tokens.get(tokens.size()).getText();
localText = localText.trim();
Matcher ma1 = startNum.matcher(localText);
Matcher ma2 = endNum.matcher(localText);
if (ma1.find()) {
String groupStr = ma1.group(0);
try {
currentNumber = Integer.parseInt(groupStr);
numberBeginLine = true;
} catch (NumberFormatException e) {
currentNumber = -1;
}
} else if (ma2.find()) {
String groupStr = ma2.group(0);
try {
currentNumber = Integer.parseInt(groupStr);
numberEndLine = true;
} catch (NumberFormatException e) {
currentNumber = -1;
}
}
if (lastNumber != -1) {
if (currentNumber == lastNumber + 1) {
foundNumbering = true;
break;
}
} else lastNumber = currentNumber;
}
}
i++;
if (i > 5) {
break;
}
}
i = 0;
if (foundNumbering) {
// we have a line numbering, so we filter them
int counter = 1; // we start at 1, if the actual start is 0,
// it will remain (as it is negligeable)
for (Block block : doc.getBlocks()) {
String localText = block.getText();
List<LayoutToken> tokens = block.tokens;
if ((localText != null) && (tokens.size() > 0)) {
if (numberEndLine) {
Matcher ma2 = endNum.matcher(localText);
if (ma2.find()) {
String groupStr = ma2.group(0);
if (groupStr.trim().equals("" + counter)) {
localText = localText.substring(0, localText.length() - groupStr.length());
block.setText(localText);
tokens.remove(tokens.size() - 1);
counter++;
}
}
} else if (numberBeginLine) {
Matcher ma1 = endNum.matcher(localText);
if (ma1.find()) {
String groupStr = ma1.group(0);
if (groupStr.trim().equals("" + counter)) {
localText = localText.substring(groupStr.length(), localText.length() - 1);
block.setText(localText);
tokens.remove(0);
counter++;
}
}
}
}
i++;
}
}
return foundNumbering;
}
public static Document generalResultSegmentation(
Document doc, String labeledResult, List<LayoutToken> documentTokens) {
List<Pair<String, String>> labeledTokens = GenericTaggerUtils.getTokensAndLabels(labeledResult);
SortedSetMultimap<String, DocumentPiece> labeledBlocks = TreeMultimap.create();
doc.setLabeledBlocks(labeledBlocks);
/*try {
FileUtils.writeStringToFile(new File("/tmp/x1.txt"), labeledResult);
FileUtils.writeStringToFile(new File("/tmp/x2.txt"), documentTokens.toString());
}
catch(Exception e) {
e.printStackTrace();
}*/
List<Block> docBlocks = doc.getBlocks();
int indexLine = 0;
int blockIndex = 0;
int p = 0; // position in the labeled result
int currentLineEndPos = 0; // position in the global doc. tokenization of the last
// token of the current line
int currentLineStartPos = 0; // position in the global doc.
// tokenization of the first token of the current line
String line = null;
DocumentPointer pointerA = DocumentPointer.START_DOCUMENT_POINTER;
DocumentPointer currentPointer = null;
DocumentPointer lastPointer = null;
String curLabel;
String curPlainLabel = null;
String lastPlainLabel = null;
int lastTokenInd = -1;
for (int i = docBlocks.size() - 1; i >= 0; i--) {
int endToken = docBlocks.get(i).getEndToken();
if (endToken != -1) {
lastTokenInd = endToken;
break;
}
}
// we do this concatenation trick so that we don't have to process stuff after the main loop
// no copying of lists happens because of this, so it's ok to concatenate
String ignoredLabel = "@IGNORED_LABEL@";
for (Pair<String, String> labeledTokenPair :
Iterables.concat(
labeledTokens,
Collections.singleton(new Pair<String, String>("IgnoredToken", ignoredLabel)))) {
if (labeledTokenPair == null) {
p++;
continue;
}
// as we process the document segmentation line by line, we don't use the usual
// tokenization to rebuild the text flow, but we get each line again from the
// text stored in the document blocks (similarly as when generating the features)
line = null;
while ((line == null) && (blockIndex < docBlocks.size())) {
Block block = docBlocks.get(blockIndex);
List<LayoutToken> tokens = block.getTokens();
String localText = block.getText();
if ((tokens == null) || (localText == null) || (localText.trim().length() == 0)) {
blockIndex++;
indexLine = 0;
if (blockIndex < docBlocks.size()) {
block = docBlocks.get(blockIndex);
currentLineStartPos = block.getStartToken();
}
continue;
}
String[] lines = localText.split("[\\n\\r]");
if ((lines.length == 0) || (indexLine >= lines.length)) {
blockIndex++;
indexLine = 0;
if (blockIndex < docBlocks.size()) {
block = docBlocks.get(blockIndex);
currentLineStartPos = block.getStartToken();
}
continue;
} else {
line = lines[indexLine];
indexLine++;
if ((line.trim().length() == 0) || (TextUtilities.filterLine(line))) {
line = null;
continue;
}
if (currentLineStartPos > lastTokenInd) continue;
// adjust the start token position in documentTokens to this non trivial line
// first skip possible space characters and tabs at the beginning of the line
while ((documentTokens.get(currentLineStartPos).t().equals(" ")
|| documentTokens.get(currentLineStartPos).t().equals("\t"))
&& (currentLineStartPos != lastTokenInd)) {
currentLineStartPos++;
}
if (!labeledTokenPair.a.startsWith(documentTokens.get(currentLineStartPos).getText())) {
while (currentLineStartPos < block.getEndToken()) {
if (documentTokens.get(currentLineStartPos).t().equals("\n")
|| documentTokens.get(currentLineStartPos).t().equals("\r")) {
// move to the start of the next line, but ignore space characters and tabs
currentLineStartPos++;
while ((documentTokens.get(currentLineStartPos).t().equals(" ")
|| documentTokens.get(currentLineStartPos).t().equals("\t"))
&& (currentLineStartPos != lastTokenInd)) {
currentLineStartPos++;
}
if ((currentLineStartPos != lastTokenInd)
&& labeledTokenPair.a.startsWith(
documentTokens.get(currentLineStartPos).getText())) {
break;
}
}
currentLineStartPos++;
}
}
// what is then the position of the last token of this line?
currentLineEndPos = currentLineStartPos;
while (currentLineEndPos < block.getEndToken()) {
if (documentTokens.get(currentLineEndPos).t().equals("\n")
|| documentTokens.get(currentLineEndPos).t().equals("\r")) {
currentLineEndPos--;
break;
}
currentLineEndPos++;
}
}
}
curLabel = labeledTokenPair.b;
curPlainLabel = GenericTaggerUtils.getPlainLabel(curLabel);
/*System.out.println("-------------------------------");
System.out.println("block: " + blockIndex);
System.out.println("line: " + line);
System.out.println("token: " + labeledTokenPair.a);
System.out.println("curPlainLabel: " + curPlainLabel);
System.out.println("lastPlainLabel: " + lastPlainLabel);
if ((currentLineStartPos < lastTokenInd) && (currentLineStartPos != -1))
System.out.println("currentLineStartPos: " + currentLineStartPos +
" (" + documentTokens.get(currentLineStartPos) + ")");
if ((currentLineEndPos < lastTokenInd) && (currentLineEndPos != -1))
System.out.println("currentLineEndPos: " + currentLineEndPos +
" (" + documentTokens.get(currentLineEndPos) + ")");*/
if (blockIndex == docBlocks.size()) {
break;
}
currentPointer = new DocumentPointer(doc, blockIndex, currentLineEndPos);
// either a new entity starts or a new beginning of the same type of entity
if ((!curPlainLabel.equals(lastPlainLabel)) && (lastPlainLabel != null)) {
if ((pointerA.getTokenDocPos() <= lastPointer.getTokenDocPos())
&& (pointerA.getTokenDocPos() != -1)) {
labeledBlocks.put(lastPlainLabel, new DocumentPiece(pointerA, lastPointer));
}
pointerA = new DocumentPointer(doc, blockIndex, currentLineStartPos);
// System.out.println("add segment for: " + lastPlainLabel + ", until " +
// (currentLineStartPos-2));
}
// updating stuff for next iteration
lastPlainLabel = curPlainLabel;
lastPointer = currentPointer;
currentLineStartPos = currentLineEndPos + 2; // one shift for the EOL, one for the next line
p++;
}
if (blockIndex == docBlocks.size()) {
// the last labelled piece has still to be added
if ((!curPlainLabel.equals(lastPlainLabel)) && (lastPlainLabel != null)) {
if ((pointerA.getTokenDocPos() <= lastPointer.getTokenDocPos())
&& (pointerA.getTokenDocPos() != -1)) {
labeledBlocks.put(lastPlainLabel, new DocumentPiece(pointerA, lastPointer));
// System.out.println("add segment for: " + lastPlainLabel + ", until " +
// (currentLineStartPos-2));
}
}
}
return doc;
}
/**
* First pass to detect basic structures: remove page header/footer, identify section numbering,
* identify Figure and table blocks.
*
* <p>-> to be removed at some point!
*
* @param doc a document
*/
public static void firstPass(Document doc) {
if (doc == null) {
throw new NullPointerException();
}
if (doc.getBlocks() == null) {
throw new NullPointerException();
}
int i = 0;
List<Integer> blockHeaders = new ArrayList<Integer>();
List<Integer> blockFooters = new ArrayList<Integer>();
List<Integer> blockSectionTitles = new ArrayList<Integer>();
List<Integer> acknowledgementBlocks = new ArrayList<Integer>();
List<Integer> blockTables = new ArrayList<Integer>();
List<Integer> blockFigures = new ArrayList<Integer>();
List<Integer> blockHeadTables = new ArrayList<Integer>();
List<Integer> blockHeadFigures = new ArrayList<Integer>();
List<Integer> blockDocumentHeaders = new ArrayList<Integer>();
doc.setTitleMatchNum(false);
try {
for (Block block : doc.getBlocks()) {
String localText = block.getText().trim();
localText = localText.replace("\n", " ");
localText = localText.replace(" ", " ");
localText = localText.trim();
Matcher ma1 = BasicStructureBuilder.introduction.matcher(localText);
Matcher ma2 = BasicStructureBuilder.references.matcher(localText);
if ((ma1.find()) || (ma2.find())) {
if (((localText.startsWith("1.")) || (localText.startsWith("1 ")))
|| ((localText.startsWith("2.")) || (localText.startsWith("2 ")))
|| (localText.startsWith("Contents"))) doc.setTitleMatchNum(true);
// System.out.println("Title section identified: block " + i + ", " + localText);
blockSectionTitles.add(i);
} else {
StringTokenizer st = new StringTokenizer(localText, "\n");
while (st.hasMoreTokens()) {
String token = st.nextToken();
if (token.startsWith("@PAGE")) {
// current block should give the header/footors
if (i > 4) {
if (doc.getBlocks().get(i - 5).getNbTokens() < 20) {
Integer i2 = i - 5;
if (!blockFooters.contains(i2)) blockFooters.add(i2);
}
}
if (i > 3) {
if (doc.getBlocks().get(i - 4).getNbTokens() < 20) {
Integer i2 = i - 4;
if (!blockFooters.contains(i2)) blockFooters.add(i2);
}
}
if (i > 2) {
if (doc.getBlocks().get(i - 3).getNbTokens() < 20) {
Integer i2 = i - 3;
if (!blockFooters.contains(i2)) blockFooters.add(i2);
}
}
if (i > 1) {
if (doc.getBlocks().get(i - 2).getNbTokens() < 20) {
Integer i2 = i - 2;
if (!blockFooters.contains(i2)) blockFooters.add(i2);
}
}
if (i > 0) {
if (doc.getBlocks().get(i - 1).getNbTokens() < 20) {
Integer i2 = i - 1;
if (!blockFooters.contains(i2)) blockFooters.add(i2);
}
}
blockFooters.add(i);
// page header candidates
blockHeaders.add(i);
if (i < doc.getBlocks().size() - 1) {
if (doc.getBlocks().get(i + 1).getNbTokens() < 20) {
Integer i2 = i + 1;
if (!blockHeaders.contains(i2)) blockHeaders.add(i + 1);
}
}
if (i < doc.getBlocks().size() - 2) {
if (doc.getBlocks().get(i + 2).getNbTokens() < 20) {
Integer i2 = i + 2;
if (!blockHeaders.contains(i2)) blockHeaders.add(i + 2);
}
}
if (i < doc.getBlocks().size() - 3) {
if (doc.getBlocks().get(i + 3).getNbTokens() < 20) {
Integer i2 = i + 3;
if (!blockHeaders.contains(i2)) blockHeaders.add(i + 3);
}
}
if (i < doc.getBlocks().size() - 4) {
if (doc.getBlocks().get(i + 4).getNbTokens() < 20) {
Integer i2 = i + 4;
if (!blockHeaders.contains(i2)) blockHeaders.add(i + 4);
}
}
// more ??
}
}
}
// clustering of blocks per font (for section header and figure/table detections)
addBlockToCluster(i, doc);
i++;
}
// try to find the cluster of section titles
Cluster candidateCluster = null;
// System.out.println("nb clusters: " + clusters.size());
for (Cluster cluster : doc.getClusters()) {
if ((cluster.getNbBlocks() < (doc.getBlocks().size() / 5))
&& (cluster.getNbBlocks() < 20)) {
List<Integer> blo = cluster.getBlocks2();
for (Integer b : blo) {
if (blockSectionTitles.contains(b)) {
if (candidateCluster == null) {
candidateCluster = cluster;
break;
}
// else if (cluster.getFontSize() >= candidateCluster.getFontSize())
// candidateCluster = cluster;
}
}
}
}
if (candidateCluster != null) {
List<Integer> newBlockSectionTitles = new ArrayList<Integer>();
for (Integer bl : blockSectionTitles) {
if (!newBlockSectionTitles.contains(bl)) newBlockSectionTitles.add(bl);
}
List<Integer> blockClusterTitles = candidateCluster.getBlocks2();
if (blockClusterTitles.size() < 20) {
for (Integer bl : blockClusterTitles) {
if (!newBlockSectionTitles.contains(bl)) newBlockSectionTitles.add(bl);
}
}
blockSectionTitles = newBlockSectionTitles;
}
// aknowledgement section recognition
boolean ackn = false;
i = 0;
for (Block block : doc.getBlocks()) {
String localText = block.getText().trim();
localText = localText.replace("\n", " ");
localText = localText.replace(" ", " ");
localText = localText.trim();
// System.out.println(i + ": " + localText+"\n");
Integer iii = i;
Matcher m3 = BasicStructureBuilder.acknowledgement.matcher(localText);
if ((m3.find()) && (blockSectionTitles.contains(iii))) {
acknowledgementBlocks.add(iii);
ackn = true;
// int index = blockSectionTitles.indexOf(iii);
// blockSectionTitles.remove(index);
} else if ((ackn) && (blockSectionTitles.contains(iii))) {
ackn = false;
break;
} else if (ackn) {
Matcher m4 = BasicStructureBuilder.references.matcher(localText);
if ((ackn) && (!blockFooters.contains(iii)) && (!m4.find())) {
acknowledgementBlocks.add(iii);
} else if (m4.find()) {
ackn = false;
break;
}
}
i++;
}
// we remove references headers in blockSectionTitles
int index = -1;
for (Integer ii : blockSectionTitles) {
Block block = doc.getBlocks().get(ii);
String localText = block.getText().trim();
localText = localText.replace("\n", " ");
localText = localText.replace(" ", " ");
localText = localText.trim();
Matcher m4 = BasicStructureBuilder.references.matcher(localText);
if (m4.find()) {
index = blockSectionTitles.indexOf(ii);
break;
}
}
if (index != -1) {
blockSectionTitles.remove(index);
}
// we check headers repetition from page to page to decide if it is an header or not
ArrayList<Integer> toRemove = new ArrayList<Integer>();
for (Integer ii : blockHeaders) {
String localText = (doc.getBlocks().get(ii)).getText().trim();
localText = TextUtilities.shadowNumbers(localText);
int length = localText.length();
if (length > 160) toRemove.add(ii);
else {
// System.out.println("header candidate: " + localText);
// evaluate distance with other potential headers
boolean valid = false;
for (Integer ii2 : blockHeaders) {
if (ii.intValue() != ii2.intValue()) {
String localText2 = doc.getBlocks().get(ii2).getText().trim();
if (localText2.length() < 160) {
localText2 = TextUtilities.shadowNumbers(localText2);
double dist =
(double) TextUtilities.getLevenshteinDistance(localText, localText2) / length;
// System.out.println("dist with " + localText2 + " : " + dist);
if (dist < 0.25) {
valid = true;
break;
}
}
}
}
if (!valid) {
toRemove.add(ii);
}
}
}
for (Integer ii : toRemove) {
blockHeaders.remove(ii);
}
// same for footers
toRemove = new ArrayList<Integer>();
for (Integer ii : blockFooters) {
String localText = (doc.getBlocks().get(ii)).getText().trim();
localText = TextUtilities.shadowNumbers(localText);
int length = localText.length();
if (length > 160) toRemove.add(ii);
else {
// System.out.println("footer candidate: " + localText);
// evaluate distance with other potential headers
boolean valid = false;
for (Integer ii2 : blockFooters) {
if (ii.intValue() != ii2.intValue()) {
String localText2 = doc.getBlocks().get(ii2).getText().trim();
if (localText2.length() < 160) {
localText2 = TextUtilities.shadowNumbers(localText2);
double dist =
(double) TextUtilities.getLevenshteinDistance(localText, localText2) / length;
if (dist < 0.25) {
valid = true;
break;
}
}
}
}
if (!valid) {
toRemove.add(ii);
}
}
}
for (Integer ii : toRemove) {
blockFooters.remove(ii);
}
// a special step for added banner repositoryies such HAL
i = 0;
for (Block block : doc.getBlocks()) {
String localText = block.getText().trim();
localText = localText.replace("\n", " ");
localText = localText.replace(" ", " ");
localText = localText.trim();
// HAL
if (localText.startsWith("Author manuscript, published in")) {
Double y = block.getY();
// System.out.println("HAL banner found, " + "block " + i + ", y = " + y);
if (Math.abs(y - 12.538) < 2) { // reference position
// blockHeaders.add(new Integer(i));
blockDocumentHeaders.add(i);
// System.out.println("HAL banner added as header block");
break;
}
}
// ACM publications
// System.out.println("test ACM " + i);
// System.out.println(localText);
if (localText.startsWith("Permission to make digital or hard copies")) {
blockFooters.add(i);
break;
}
// arXiv, etc. put here
// IOP
if (localText.startsWith("Confidential: ") && (localText.contains("IOP"))) {
blockDocumentHeaders.add(i);
// System.out.println("IOP banner added as header block");
break;
}
i++;
}
// we try to recognize here table and figure blocks
// the idea is that the textual elements are not located as the normal text blocks
// this is recognized by exploiting the cluster of blocks starting up and down front the block
// containing a table or a figure marker
// two different runs, one for figures and one for tables (everything could be done in one
// step)
i = 0;
for (Block block : doc.getBlocks()) {
String localText = block.getText().trim();
localText = localText.replace("\n", " ");
localText = localText.replace(" ", " ");
localText = localText.trim();
Matcher m = BasicStructureBuilder.figure.matcher(localText);
Matcher m2 = BasicStructureBuilder.table.matcher(localText);
double width = block.getWidth();
boolean bold = block.getBold();
// table
// if ( (m2.find()) && (localText.length() < 200) ) {
if ((m2.find()) && ((bold) || (localText.length() < 200))) {
if (!blockHeadTables.contains(i)) {
blockHeadTables.add(i);
}
// we also put all the small blocks before and after the marker
int j = i - 1;
while ((j > i - 15) && (j > 0)) {
Block b = doc.getBlocks().get(j);
if (b.getText() != null) {
if ((b.getText().length() < 160) || (width < 50)) {
if ((!blockTables.contains(j))
&& (!blockSectionTitles.contains(j))
&& (!blockHeaders.contains(j))
&& (!blockFooters.contains(j))) blockTables.add(j);
} else j = 0;
}
j--;
}
j = i + 1;
while ((j < i + 15) && (j < doc.getBlocks().size())) {
Block b = doc.getBlocks().get(j);
if (b.getText() != null) {
if ((b.getText().length() < 160) || (width < 50)) {
if ((!blockTables.contains(j))
&& (!blockSectionTitles.contains(j))
&& (!blockHeaders.contains(j))
&& (!blockFooters.contains(j))) blockTables.add(j);
} else j = doc.getBlocks().size();
}
j++;
}
}
// figure
// else if ( (m.find()) && (localText.length() < 200) ) {
else if ((m.find()) && ((bold) || (localText.length() < 200))) {
if (!blockHeadFigures.contains(i)) blockHeadFigures.add(i);
// we also put all the small blocks before and after the marker
int j = i - 1;
boolean imageFound = false;
while ((j > i - 15) && (j > 0)) {
Block b = doc.getBlocks().get(j);
if (b.getText() != null) {
String localText2 = b.getText().trim();
// localText = localText.replace("\n", " ");
localText2 = localText2.replace(" ", " ");
localText2 = localText2.trim();
if ((localText2.startsWith("@IMAGE")) && (!imageFound)) {
// System.out.println(localText2);
block.setText(block.getText() + " " + localText2);
// System.out.println(block.getText());
imageFound = true;
}
if ((localText2.length() < 160) || (width < 50)) {
if ((!blockFigures.contains(j))
&& (!blockSectionTitles.contains(j))
&& (!blockHeaders.contains(j))
&& (!blockFooters.contains(j))) blockFigures.add(j);
} else j = 0;
}
j--;
}
j = i + 1;
while ((j < i + 15) && (j < doc.getBlocks().size())) {
Block b = doc.getBlocks().get(j);
if (b.getText() != null) {
if ((b.getText().trim().length() < 160) || (width < 50)) {
if ((!blockFigures.contains(j))
&& (!blockSectionTitles.contains(j))
&& (!blockHeaders.contains(j))
&& (!blockFooters.contains(j))) blockFigures.add(j);
} else j = doc.getBlocks().size();
}
j++;
}
}
i++;
}
} finally {
doc.setBlockHeaders(blockHeaders);
doc.setBlockFooters(blockFooters);
doc.setBlockSectionTitles(blockSectionTitles);
doc.setAcknowledgementBlocks(acknowledgementBlocks);
doc.setBlockTables(blockTables);
doc.setBlockFigures(blockFigures);
doc.setBlockHeadTables(blockHeadTables);
doc.setBlockHeadFigures(blockHeadFigures);
doc.setBlockDocumentHeaders(blockDocumentHeaders);
}
}
public void startElement(String namespaceURI, String localName, String qName, Attributes atts)
throws SAXException {
if (qName.equals("PAGE")) {
int length = atts.getLength();
currentPage++;
// Process each attribute
for (int i = 0; i < length; i++) {
// Get names and values for each attribute
String name = atts.getQName(i);
String value = atts.getValue(i);
if ((name != null) && (value != null)) {
if (name.equals("id")) {;
} else if (name.equals("number")) {
} else if (name.equals("width")) {
} else if (name.equals("height")) {
}
}
}
/*
* if (block != null) { blabla.append("\n");
* tokenizations.add("\n"); block.setText(blabla.toString());
* block.setNbTokens(nbTokens); doc.addBlock(block); } Block block0
* = new Block(); block0.setText("@PAGE\n"); block0.setNbTokens(0);
* doc.addBlock(block0);
*/
/*
* block = new Block(); blabla = new StringBuffer(); nbTokens = 0;
* //blabla.append("\n@block\n"); tokenizations.add("\n");
*/
} else if (qName.equals("BLOCK")) {
block = new Block();
blabla = new StringBuffer();
nbTokens = 0;
block.setPage(currentPage);
// blabla.append("\n@block\n");
} else if (qName.equals("IMAGE")) {
int length = atts.getLength();
// Process each attribute
for (int i = 0; i < length; i++) {
// Get names and values for each attribute
String name = atts.getQName(i);
String value = atts.getValue(i);
if ((name != null) && (value != null)) {
if (name.equals("href")) {
// if (images == null)
// images = new ArrayList<String>();
images.add(value);
} else if (name.equals("x")) {
double x = Double.parseDouble(value);
if (x != currentX) {
currentX = x;
}
} else if (name.equals("y")) {
double y = Double.parseDouble(value);
if (y != currentY) {
currentY = y;
}
} else if (name.equals("width")) {
double width = Double.parseDouble(value);
if (width != currentWidth) {
currentWidth = width;
}
} else if (name.equals("height")) {
double height = Double.parseDouble(value);
if (height != currentHeight) {
currentHeight = height;
}
}
}
}
} else if (qName.equals("TEXT")) {
int length = atts.getLength();
// Process each attribute
for (int i = 0; i < length; i++) {
// Get names and values for each attribute
String name = atts.getQName(i);
String value = atts.getValue(i);
if ((name != null) && (value != null)) {
if (name.equals("id")) {
} else if (name.equals("x")) {
} else if (name.equals("y")) {
} else if (name.equals("width")) {
} else if (name.equals("height")) {
}
}
}
} else if (qName.equals("TOKEN")) {
int length = atts.getLength();
// Process each attribute
for (int i = 0; i < length; i++) {
// Get names and values for each attribute
String name = atts.getQName(i);
String value = atts.getValue(i);
if ((name != null) && (value != null)) {
if (name.equals("id")) {;
} else if (name.equals("font-name")) {
if (!value.equals(currentFont)) {
currentFont = value;
blabla.append(" ");
}
} else if (name.equals("font-size")) {
double fontSize = Double.parseDouble(value);
if (fontSize != currentFontSize) {
currentFontSize = fontSize;
blabla.append(" ");
}
} else if (name.equals("bold")) {
if (value.equals("yes")) {
currentBold = true;
} else {
currentBold = false;
}
} else if (name.equals("italic")) {
if (value.equals("yes")) {
currentItalic = true;
} else {
currentItalic = false;
}
} else if (name.equals("font-color")) {
if (!value.equals(colorFont)) {
colorFont = value;
}
} else if (name.equals("rotation")) {
if (value.equals("0")) currentRotation = false;
else currentRotation = true;
} else if (name.equals("x")) {
double x = Double.parseDouble(value);
if (x != currentX) {
currentX = x;
}
} else if (name.equals("y")) {
double y = Double.parseDouble(value);
if (y != currentY) {
currentY = y;
}
} else if (name.equals("base")) {
double base = Double.parseDouble(value);
} else if (name.equals("width")) {
double width = Double.parseDouble(value);
if (width != currentWidth) {
currentWidth = width;
}
} else if (name.equals("height")) {
double height = Double.parseDouble(value);
if (height != currentHeight) {
currentHeight = height;
}
}
}
}
} else if (qName.equals("xi:include")) {
int length = atts.getLength();
// Process each attribute
for (int i = 0; i < length; i++) {
// Get names and values for each attribute
String name = atts.getQName(i);
String value = atts.getValue(i);
if ((name != null) && (value != null)) {
if (name.equals("href")) {
// if (images == null)
// images = new ArrayList<String>();
images.add(value);
}
}
}
}
// accumulator.setLength(0);
}
public void endElement(java.lang.String uri, java.lang.String localName, java.lang.String qName)
throws SAXException {
// if (!qName.equals("TOKEN") && !qName.equals("BLOCK") &&
// !qName.equals("TEXT"))
// System.out.println(qName);
if (qName.equals("TEXT")) {
blabla.append("\n");
LayoutToken token = new LayoutToken();
token.setText("\n");
block.addToken(token);
accumulator.setLength(0);
tokenizations.add("\n");
} else if (qName.equals("METADATA")) {
accumulator.setLength(0);
} else if (qName.equals("TOKEN")) {
String tok0 = TextUtilities.clean(getText());
if (block.getStartToken() == -1) {
block.setStartToken(tokenizations.size());
}
if (tok0.length() > 0) {
StringTokenizer st = new StringTokenizer(tok0, TextUtilities.fullPunctuations, true);
boolean diaresis = false;
boolean accent = false;
boolean keepLast = false;
while (st.hasMoreTokens()) {
diaresis = false;
accent = false;
keepLast = false;
String tok = st.nextToken();
if (tok.length() > 0) {
LayoutToken token = new LayoutToken();
if ((previousToken != null)
&& (tok != null)
&& (previousToken.length() > 0)
&& (tok.length() > 0)
&& blabla.length() > 0) {
Character leftChar = previousTok.getText().charAt(previousTok.getText().length() - 1);
Character rightChar = tok.charAt(0);
ModifierClass leftClass = classifyChar(leftChar);
ModifierClass rightClass = classifyChar(rightChar);
ModifierClass modifierClass = ModifierClass.NOT_A_MODIFIER;
if (leftClass != ModifierClass.NOT_A_MODIFIER
|| rightClass != ModifierClass.NOT_A_MODIFIER) {
Character baseChar = null;
Character modifierChar = null;
if (leftClass != ModifierClass.NOT_A_MODIFIER) {
if (rightClass != ModifierClass.NOT_A_MODIFIER) {
// assert false;
// keeping characters, but setting class
// to not a modifier
baseChar = leftChar;
modifierChar = rightChar;
modifierClass = ModifierClass.NOT_A_MODIFIER;
} else {
baseChar = rightChar;
modifierChar = leftChar;
modifierClass = leftClass;
}
} else {
baseChar = leftChar;
modifierChar = rightChar;
modifierClass = rightClass;
}
String updatedChar = modifyCharacter(baseChar, modifierChar);
tokenizations.remove(tokenizations.size() - 1);
if (tokenizations.size() > 0) {
tokenizations.remove(tokenizations.size() - 1);
}
blabla.deleteCharAt(blabla.length() - 1);
if (blabla.length() > 0) {
blabla.deleteCharAt(blabla.length() - 1);
}
removeLastCharacterIfPresent(previousTok);
if (updatedChar != null) {
blabla.append(updatedChar);
previousTok.setText(previousTok.getText() + updatedChar);
}
blabla.append(tok.substring(1, tok.length()));
previousTok.setText(previousTok.getText() + tok.substring(1, tok.length()));
tokenizations.add(previousTok.getText());
diaresis =
(modifierClass == ModifierClass.DIAERESIS
|| modifierClass == ModifierClass.NORDIC_RING
|| modifierClass == ModifierClass.CZECH_CARON
|| modifierClass == ModifierClass.TILDE
|| modifierClass == ModifierClass.CEDILLA);
accent =
(modifierClass == ModifierClass.ACUTE_ACCENT
|| modifierClass == ModifierClass.CIRCUMFLEX
|| modifierClass == ModifierClass.GRAVE_ACCENT);
if (rightClass != ModifierClass.NOT_A_MODIFIER) {
tok = ""; // resetting current token as it
// is a single-item
}
}
}
if (tok != null) {
// actually in certain cases, the extracted string under token can be a chunk of text
// with separators that need to be preserved
// tok = tok.replace(" ", "");
}
if ((!diaresis) && (!accent)) {
// blabla.append(" ");
blabla.append(tok);
token.setText(tok);
tokenizations.add(tok);
} else {
tok = "";
keepLast = true;
}
/*
* StringTokenizer st0 = new StringTokenizer(tok0,
* TextUtilities.fullPunctuations, true);
* while(st0.hasMoreTokens()) { String tok =
* st0.nextToken(); tokenizations.add(tok); }
* tokenizations.add(" ");
*/
/*
* boolean punct1 = false; boolean punct2 = false;
* boolean punct3 = false; String content = null; int i
* = 0; for(; i<TextUtilities.punctuations.length();
* i++) { if (tok.length() > 0) { if
* (tok.charAt(tok.length()-1) ==
* TextUtilities.punctuations.charAt(i)) { punct1 =
* true; content = tok.substring(0, tok.length()-1); if
* (tok.length() > 1) { int j = 0; for(;
* j<TextUtilities.punctuations.length(); j++) { if
* (tok.charAt(tok.length()-2) ==
* TextUtilities.punctuations.charAt(j)) { punct3 =
* true; content = tok.substring(0, tok.length()-2); } }
* } break; } } } if (tok.length() > 0) { if (
* (tok.startsWith("(")) && (tok.length() > 1) ) { if
* ((punct3) && (tok.length() > 2)) content =
* tok.substring(1, tok.length()-2); else if (punct1)
* content = tok.substring(1, tok.length()-1); else
* content = tok.substring(1, tok.length()); punct2 =
* true; token.setText("("); } else if (
* (tok.startsWith("[")) && (tok.length() > 1) ) { if
* ((punct3) && (tok.length() > 2)) content =
* tok.substring(1, tok.length()-2); else if (punct1)
* content = tok.substring(1, tok.length()-1); else
* content = tok.substring(1, tok.length()); punct2 =
* true; token.setText("["); } else if (
* (tok.startsWith("\"")) && (tok.length() > 1) ) { if
* ((punct3) && (tok.length() > 2)) content =
* tok.substring(1, tok.length()-2); else if (punct1)
* content = tok.substring(1, tok.length()-1); else
* content = tok.substring(1, tok.length()); punct2 =
* true; token.setText("\""); } }
*/
if (currentRotation) currentFontSize = currentFontSize / 2;
/*
* if (punct2) { if (currentFont != null)
* token.setFont(currentFont.toLowerCase()); else
* token.setFont("default");
* token.setItalic(currentItalic);
* token.setBold(currentBold);
* token.setRotation(currentRotation);
* token.setColorFont(colorFont); token.setX(currentX);
* token.setY(currentY); token.setWidth(currentWidth);
* token.setHeight(currentHeight);
* token.setFontSize(currentFontSize);
* block.addToken(token);
*
* token = new LayoutToken(); token.setText(content); }
* if (punct1) { token.setText(content); if (currentFont
* != null) token.setFont(currentFont.toLowerCase());
* else token.setFont("default");
* token.setItalic(currentItalic);
* token.setBold(currentBold);
* token.setRotation(currentRotation);
* token.setColorFont(colorFont); token.setX(currentX);
* token.setY(currentY); token.setWidth(currentWidth);
* token.setHeight(currentHeight);
* token.setFontSize(currentFontSize);
* block.addToken(token);
*
* if (punct3) { token = new LayoutToken();
* token.setText(""+tok.charAt(tok.length()-2)); if
* (currentFont != null)
* token.setFont(currentFont.toLowerCase()); else
* token.setFont("default");
* token.setItalic(currentItalic);
* token.setBold(currentBold);
* token.setRotation(currentRotation);
* token.setColorFont(colorFont); token.setX(currentX);
* token.setY(currentY); token.setWidth(currentWidth);
* token.setHeight(currentHeight);
* token.setFontSize(currentFontSize);
* block.addToken(token); }
*
* token = new LayoutToken();
* token.setText(""+tok.charAt(tok.length()-1)); }
*/
if (currentFont != null) token.setFont(currentFont.toLowerCase());
else token.setFont("default");
token.setItalic(currentItalic);
token.setBold(currentBold);
token.setRotation(currentRotation);
token.setColorFont(colorFont);
token.setX(currentX);
token.setY(currentY);
token.setWidth(currentWidth);
token.setHeight(currentHeight);
token.setFontSize(currentFontSize);
if (!diaresis && !accent) {
block.addToken(token);
}
if (block.getFont() == null) {
if (currentFont != null) block.setFont(currentFont.toLowerCase());
else token.setFont("default");
}
if (nbTokens == 0) {
block.setItalic(currentItalic);
block.setBold(currentBold);
}
if (block.getColorFont() == null) block.setColorFont(colorFont);
if (block.getX() == 0.0) block.setX(currentX);
if (block.getY() == 0.0) block.setY(currentY);
if (block.getWidth() == 0.0) block.setWidth(currentWidth);
if (block.getHeight() == 0.0) block.setHeight(currentHeight);
if (block.getFontSize() == 0.0) block.setFontSize(currentFontSize);
if (!diaresis && !accent) {
previousToken = tok;
previousTok = token;
} else {
previousToken = previousTok.getText();
}
nbTokens++;
accumulator.setLength(0);
}
}
if (tokenizations.size() > 0) {
String justBefore = tokenizations.get(tokenizations.size() - 1);
if (!justBefore.endsWith("-")) {
tokenizations.add(" ");
blabla.append(" ");
}
}
}
block.setEndToken(tokenizations.size());
} else if (qName.equals("PAGE")) {
// page marker are usefull to detect headers (same first line(s)
// appearing on each page)
if (block != null) {
blabla.append("\n");
tokenizations.add("\n");
block.setText(blabla.toString());
block.setNbTokens(nbTokens);
doc.addBlock(block);
}
Block block0 = new Block();
block0.setText("@PAGE\n");
block0.setNbTokens(0);
block0.setPage(currentPage);
doc.addBlock(block0);
block = new Block();
block.setPage(currentPage);
blabla = new StringBuffer();
nbTokens = 0;
// blabla.append("\n@block\n");
tokenizations.add("\n");
} else if (qName.equals("IMAGE")) {
if (block != null) {
blabla.append("\n");
block.setText(blabla.toString());
block.setNbTokens(nbTokens);
doc.addBlock(block);
}
block = new Block();
block.setPage(currentPage);
blabla = new StringBuffer();
if (images.size() > 0) {
blabla.append("@IMAGE " + images.get(images.size() - 1) + "\n");
}
block.setText(blabla.toString());
block.setNbTokens(nbTokens);
if (block.getX() == 0.0) block.setX(currentX);
if (block.getY() == 0.0) block.setY(currentY);
if (block.getWidth() == 0.0) block.setWidth(currentWidth);
if (block.getHeight() == 0.0) block.setHeight(currentHeight);
doc.addBlock(block);
blabla = new StringBuffer();
nbTokens = 0;
block = new Block();
block.setPage(currentPage);
}
/*
* else if (qName.equals("VECTORIALIMAGES")) { if (block != null) {
* blabla.append("\n"); block.setText(blabla.toString());
* block.setNbTokens(nbTokens); doc.addBlock(block); } block = new
* Block(); block.setPage(currentPage); blabla = new StringBuffer();
* blabla.append("@IMAGE " + "vectorial \n");
* block.setText(blabla.toString()); block.setNbTokens(nbTokens); if
* (block.getX() == 0.0) block.setX(currentX); if (block.getY() == 0.0)
* block.setY(currentY); if (block.getWidth() == 0.0)
* block.setWidth(currentWidth); if (block.getHeight() == 0.0)
* block.setHeight(currentHeight); doc.addBlock(block); blabla = new
* StringBuffer(); nbTokens = 0; block = new Block();
* block.setPage(currentPage); }
*/
else if (qName.equals("BLOCK")) {
blabla.append("\n");
tokenizations.add("\n");
block.setText(blabla.toString());
block.setNbTokens(nbTokens);
block.setWidth(currentX - block.getX() + currentWidth);
block.setHeight(currentY - block.getY() + currentHeight);
doc.addBlock(block);
// blabla = new StringBuffer();
nbTokens = 0;
block = null;
} else if (qName.equals("xi:include")) {
if (block != null) {
blabla.append("\n");
block.setText(blabla.toString());
block.setNbTokens(nbTokens);
doc.addBlock(block);
}
block = new Block();
block.setPage(currentPage);
blabla = new StringBuffer();
blabla.append("@IMAGE " + images.get(images.size() - 1) + "\n");
block.setText(blabla.toString());
block.setNbTokens(nbTokens);
doc.addBlock(block);
blabla = new StringBuffer();
nbTokens = 0;
block = new Block();
block.setPage(currentPage);
}
/*
* else if (qName.equals("DOCUMENT")) {
* System.out.println(blabla.toString()); }
*/
}