从具有不同高度的 table 行中提取 pdf 文本（java 使用 pdfbox 库）
Extracting text from pdf (java using pdfbox library) from a table's rows with different heights
黑色形状是需要提取的文本：
到目前为止，我已经从列中提取了文本，但是是手动提取的，因为只有 5 个（使用矩形 class 作为区域）。我的问题是：有没有办法对行执行此操作，因为矩形的大小（高度）不同并且手动将其执行到 50 多行将是一种暴行？更具体地说，我可以使用函数根据每一行的高度更改矩形吗？或者任何可能有帮助的建议？
如评论中所建议，您可以通过解析页面的矢量图形指令自动识别示例 PDF 的 table 单元格区域。
对于这样的任务，您可以扩展 PDFBox PDFGraphicsStreamEngine，它提供了调用路径构建和绘图指令的抽象方法。
注意：我在这里展示的流引擎 class 专门用于识别 table 单元格框架线绘制为示例文档中使用的黑色填充的长小矩形。对于一般解决方案，您至少还应该识别绘制为矢量图形线段或描边矩形的框架线。
流引擎classPdfBoxFinder
这个流引擎class收集水平线的y坐标范围和垂直线的x坐标范围和然后提供由这些坐标范围定义的网格框。特别是这意味着不支持行跨度或列跨度；在手头的情况下，这没关系，因为没有这样的跨度。
public class PdfBoxFinder extends PDFGraphicsStreamEngine {
/**
* Supply the page to analyze here; to analyze multiple pages
* create multiple {@link PdfBoxFinder} instances.
*/
public PdfBoxFinder(PDPage page) {
super(page);
}
/**
* The boxes ({@link Rectangle2D} instances with coordinates according to
* the PDF coordinate system, e.g. for decorating the table cells) the
* {@link PdfBoxFinder} has recognized on the current page.
*/
public Map<String, Rectangle2D> getBoxes() {
consolidateLists();
Map<String, Rectangle2D> result = new HashMap<>();
if (!horizontalLines.isEmpty() && !verticalLines.isEmpty())
{
Interval top = horizontalLines.get(horizontalLines.size() - 1);
char rowLetter = 'A';
for (int i = horizontalLines.size() - 2; i >= 0; i--, rowLetter++) {
Interval bottom = horizontalLines.get(i);
Interval left = verticalLines.get(0);
int column = 1;
for (int j = 1; j < verticalLines.size(); j++, column++) {
Interval right = verticalLines.get(j);
String name = String.format("%s%s", rowLetter, column);
Rectangle2D rectangle = new Rectangle2D.Float(left.from, bottom.from, right.to - left.from, top.to - bottom.from);
result.put(name, rectangle);
left = right;
}
top = bottom;
}
}
return result;
}
/**
* The regions ({@link Rectangle2D} instances with coordinates according
* to the PDFBox text extraction API, e.g. for initializing the regions of
* a {@link PDFTextStripperByArea}) the {@link PdfBoxFinder} has recognized
* on the current page.
*/
public Map<String, Rectangle2D> getRegions() {
PDRectangle cropBox = getPage().getCropBox();
float xOffset = cropBox.getLowerLeftX();
float yOffset = cropBox.getUpperRightY();
Map<String, Rectangle2D> result = getBoxes();
for (Map.Entry<String, Rectangle2D> entry : result.entrySet()) {
Rectangle2D box = entry.getValue();
Rectangle2D region = new Rectangle2D.Float(xOffset + (float)box.getX(), yOffset - (float)(box.getY() + box.getHeight()), (float)box.getWidth(), (float)box.getHeight());
entry.setValue(region);
}
return result;
}
/**
* <p>
* Processes the path elements currently in the {@link #path} list and
* eventually clears the list.
* </p>
* <p>
* Currently only elements are considered which
* </p>
* <ul>
* <li>are {@link Rectangle} instances;
* <li>are filled fairly black;
* <li>have a thin and long form; and
* <li>have sides fairly parallel to the coordinate axis.
* </ul>
*/
void processPath() throws IOException {
PDColor color = getGraphicsState().getNonStrokingColor();
if (!isBlack(color)) {
logger.debug("Dropped path due to non-black fill-color.");
return;
}
for (PathElement pathElement : path) {
if (pathElement instanceof Rectangle) {
Rectangle rectangle = (Rectangle) pathElement;
double p0p1 = rectangle.p0.distance(rectangle.p1);
double p1p2 = rectangle.p1.distance(rectangle.p2);
boolean p0p1small = p0p1 < 3;
boolean p1p2small = p1p2 < 3;
if (p0p1small) {
if (p1p2small) {
logger.debug("Dropped rectangle too small on both sides.");
} else {
processThinRectangle(rectangle.p0, rectangle.p1, rectangle.p2, rectangle.p3);
}
} else if (p1p2small) {
processThinRectangle(rectangle.p1, rectangle.p2, rectangle.p3, rectangle.p0);
} else {
logger.debug("Dropped rectangle too large on both sides.");
}
}
}
path.clear();
}
/**
* The argument points shall be sorted to have (p0, p1) and (p2, p3) be the small
* edges and (p1, p2) and (p3, p0) the long ones.
*/
void processThinRectangle(Point2D p0, Point2D p1, Point2D p2, Point2D p3) {
float longXDiff = (float)Math.abs(p2.getX() - p1.getX());
float longYDiff = (float)Math.abs(p2.getY() - p1.getY());
boolean longXDiffSmall = longXDiff * 10 < longYDiff;
boolean longYDiffSmall = longYDiff * 10 < longXDiff;
if (longXDiffSmall) {
verticalLines.add(new Interval(p0.getX(), p1.getX(), p2.getX(), p3.getX()));
} else if (longYDiffSmall) {
horizontalLines.add(new Interval(p0.getY(), p1.getY(), p2.getY(), p3.getY()));
} else {
logger.debug("Dropped rectangle too askew.");
}
}
/**
* Sorts the {@link #horizontalLines} and {@link #verticalLines} lists and
* merges fairly identical entries.
*/
void consolidateLists() {
for (List<Interval> intervals : Arrays.asList(horizontalLines, verticalLines)) {
intervals.sort(null);
for (int i = 1; i < intervals.size();) {
if (intervals.get(i-1).combinableWith(intervals.get(i))) {
Interval interval = intervals.get(i-1).combineWith(intervals.get(i));
intervals.set(i-1, interval);
intervals.remove(i);
} else {
i++;
}
}
}
}
/**
* Checks whether the given color is black'ish.
*/
boolean isBlack(PDColor color) throws IOException {
int value = color.toRGB();
for (int i = 0; i < 2; i++) {
int component = value & 0xff;
if (component > 5)
return false;
value /= 256;
}
return true;
}
//
// PDFGraphicsStreamEngine overrides
//
@Override
public void appendRectangle(Point2D p0, Point2D p1, Point2D p2, Point2D p3) throws IOException {
path.add(new Rectangle(p0, p1, p2, p3));
}
@Override
public void endPath() throws IOException {
path.clear();
}
@Override
public void strokePath() throws IOException {
path.clear();
}
@Override
public void fillPath(int windingRule) throws IOException {
processPath();
}
@Override
public void fillAndStrokePath(int windingRule) throws IOException {
processPath();
}
@Override public void drawImage(PDImage pdImage) throws IOException { }
@Override public void clip(int windingRule) throws IOException { }
@Override public void moveTo(float x, float y) throws IOException { }
@Override public void lineTo(float x, float y) throws IOException { }
@Override public void curveTo(float x1, float y1, float x2, float y2, float x3, float y3) throws IOException { }
@Override public Point2D getCurrentPoint() throws IOException { return null; }
@Override public void closePath() throws IOException { }
@Override public void shadingFill(COSName shadingName) throws IOException { }
//
// inner classes
//
class Interval implements Comparable<Interval> {
final float from;
final float to;
Interval(float... values) {
Arrays.sort(values);
this.from = values[0];
this.to = values[values.length - 1];
}
Interval(double... values) {
Arrays.sort(values);
this.from = (float) values[0];
this.to = (float) values[values.length - 1];
}
boolean combinableWith(Interval other) {
if (this.from > other.from)
return other.combinableWith(this);
if (this.to < other.from)
return false;
float intersectionLength = Math.min(this.to, other.to) - other.from;
float thisLength = this.to - this.from;
float otherLength = other.to - other.from;
return (intersectionLength >= thisLength * .9f) || (intersectionLength >= otherLength * .9f);
}
Interval combineWith(Interval other) {
return new Interval(this.from, this.to, other.from, other.to);
}
@Override
public int compareTo(Interval o) {
return this.from == o.from ? Float.compare(this.to, o.to) : Float.compare(this.from, o.from);
}
@Override
public String toString() {
return String.format("[%3.2f, %3.2f]", from, to);
}
}
interface PathElement {
}
class Rectangle implements PathElement {
final Point2D p0, p1, p2, p3;
Rectangle(Point2D p0, Point2D p1, Point2D p2, Point2D p3) {
this.p0 = p0;
this.p1 = p1;
this.p2 = p2;
this.p3 = p3;
}
}
//
// members
//
final List<PathElement> path = new ArrayList<>();
final List<Interval> horizontalLines = new ArrayList<>();
final List<Interval> verticalLines = new ArrayList<>();
final Logger logger = LoggerFactory.getLogger(PdfBoxFinder.class);
}
示例使用
您可以像这样使用 PdfBoxFinder 从位于 FILE_PATH:
的示例文档的 table 单元格中提取文本
try ( PDDocument document = PDDocument.load(FILE_PATH) ) {
for (PDPage page : document.getDocumentCatalog().getPages()) {
PdfBoxFinder boxFinder = new PdfBoxFinder(page);
boxFinder.processPage(page);
PDFTextStripperByArea stripperByArea = new PDFTextStripperByArea();
for (Map.Entry<String, Rectangle2D> entry : boxFinder.getRegions().entrySet()) {
stripperByArea.addRegion(entry.getKey(), entry.getValue());
}
stripperByArea.extractRegions(page);
List<String> names = stripperByArea.getRegions();
names.sort(null);
for (String name : names) {
System.out.printf("[%s] %s\n", name, stripperByArea.getTextForRegion(name));
}
}
}
(ExtractBoxedText 测试 testExtractBoxedTexts)
输出开始：
[A1] Nr.
crt.
[A2] Nume şi prenume
[A3] Titlul lucrării
[A4] Coordonator ştiinţific
[A5] Ora
[B1] 1.
[B2] SFETCU I. JESSICA-
LARISA
[B3] Analiza fluxurilor de date twitter
[B4] Conf. univ. dr. Frîncu Marc
Eduard
[B5] 8:00
[C1] 2.
[C2] TARBA V. IONUȚ-
ADRIAN
[C3] Test me - rest api folosind java şi
play framework
[C4] Conf.univ.dr. Fortiş Teodor
Florin
[C5] 8:12
文档第一页：
黑色形状是需要提取的文本：
到目前为止，我已经从列中提取了文本，但是是手动提取的，因为只有 5 个（使用矩形 class 作为区域）。我的问题是：有没有办法对行执行此操作，因为矩形的大小（高度）不同并且手动将其执行到 50 多行将是一种暴行？更具体地说，我可以使用函数根据每一行的高度更改矩形吗？或者任何可能有帮助的建议？
如评论中所建议，您可以通过解析页面的矢量图形指令自动识别示例 PDF 的 table 单元格区域。
对于这样的任务，您可以扩展 PDFBox PDFGraphicsStreamEngine，它提供了调用路径构建和绘图指令的抽象方法。
注意：我在这里展示的流引擎 class 专门用于识别 table 单元格框架线绘制为示例文档中使用的黑色填充的长小矩形。对于一般解决方案，您至少还应该识别绘制为矢量图形线段或描边矩形的框架线。
流引擎classPdfBoxFinder
这个流引擎class收集水平线的y坐标范围和垂直线的x坐标范围和然后提供由这些坐标范围定义的网格框。特别是这意味着不支持行跨度或列跨度；在手头的情况下，这没关系，因为没有这样的跨度。
public class PdfBoxFinder extends PDFGraphicsStreamEngine {
/**
* Supply the page to analyze here; to analyze multiple pages
* create multiple {@link PdfBoxFinder} instances.
*/
public PdfBoxFinder(PDPage page) {
super(page);
}
/**
* The boxes ({@link Rectangle2D} instances with coordinates according to
* the PDF coordinate system, e.g. for decorating the table cells) the
* {@link PdfBoxFinder} has recognized on the current page.
*/
public Map<String, Rectangle2D> getBoxes() {
consolidateLists();
Map<String, Rectangle2D> result = new HashMap<>();
if (!horizontalLines.isEmpty() && !verticalLines.isEmpty())
{
Interval top = horizontalLines.get(horizontalLines.size() - 1);
char rowLetter = 'A';
for (int i = horizontalLines.size() - 2; i >= 0; i--, rowLetter++) {
Interval bottom = horizontalLines.get(i);
Interval left = verticalLines.get(0);
int column = 1;
for (int j = 1; j < verticalLines.size(); j++, column++) {
Interval right = verticalLines.get(j);
String name = String.format("%s%s", rowLetter, column);
Rectangle2D rectangle = new Rectangle2D.Float(left.from, bottom.from, right.to - left.from, top.to - bottom.from);
result.put(name, rectangle);
left = right;
}
top = bottom;
}
}
return result;
}
/**
* The regions ({@link Rectangle2D} instances with coordinates according
* to the PDFBox text extraction API, e.g. for initializing the regions of
* a {@link PDFTextStripperByArea}) the {@link PdfBoxFinder} has recognized
* on the current page.
*/
public Map<String, Rectangle2D> getRegions() {
PDRectangle cropBox = getPage().getCropBox();
float xOffset = cropBox.getLowerLeftX();
float yOffset = cropBox.getUpperRightY();
Map<String, Rectangle2D> result = getBoxes();
for (Map.Entry<String, Rectangle2D> entry : result.entrySet()) {
Rectangle2D box = entry.getValue();
Rectangle2D region = new Rectangle2D.Float(xOffset + (float)box.getX(), yOffset - (float)(box.getY() + box.getHeight()), (float)box.getWidth(), (float)box.getHeight());
entry.setValue(region);
}
return result;
}
/**
* <p>
* Processes the path elements currently in the {@link #path} list and
* eventually clears the list.
* </p>
* <p>
* Currently only elements are considered which
* </p>
* <ul>
* <li>are {@link Rectangle} instances;
* <li>are filled fairly black;
* <li>have a thin and long form; and
* <li>have sides fairly parallel to the coordinate axis.
* </ul>
*/
void processPath() throws IOException {
PDColor color = getGraphicsState().getNonStrokingColor();
if (!isBlack(color)) {
logger.debug("Dropped path due to non-black fill-color.");
return;
}
for (PathElement pathElement : path) {
if (pathElement instanceof Rectangle) {
Rectangle rectangle = (Rectangle) pathElement;
double p0p1 = rectangle.p0.distance(rectangle.p1);
double p1p2 = rectangle.p1.distance(rectangle.p2);
boolean p0p1small = p0p1 < 3;
boolean p1p2small = p1p2 < 3;
if (p0p1small) {
if (p1p2small) {
logger.debug("Dropped rectangle too small on both sides.");
} else {
processThinRectangle(rectangle.p0, rectangle.p1, rectangle.p2, rectangle.p3);
}
} else if (p1p2small) {
processThinRectangle(rectangle.p1, rectangle.p2, rectangle.p3, rectangle.p0);
} else {
logger.debug("Dropped rectangle too large on both sides.");
}
}
}
path.clear();
}
/**
* The argument points shall be sorted to have (p0, p1) and (p2, p3) be the small
* edges and (p1, p2) and (p3, p0) the long ones.
*/
void processThinRectangle(Point2D p0, Point2D p1, Point2D p2, Point2D p3) {
float longXDiff = (float)Math.abs(p2.getX() - p1.getX());
float longYDiff = (float)Math.abs(p2.getY() - p1.getY());
boolean longXDiffSmall = longXDiff * 10 < longYDiff;
boolean longYDiffSmall = longYDiff * 10 < longXDiff;
if (longXDiffSmall) {
verticalLines.add(new Interval(p0.getX(), p1.getX(), p2.getX(), p3.getX()));
} else if (longYDiffSmall) {
horizontalLines.add(new Interval(p0.getY(), p1.getY(), p2.getY(), p3.getY()));
} else {
logger.debug("Dropped rectangle too askew.");
}
}
/**
* Sorts the {@link #horizontalLines} and {@link #verticalLines} lists and
* merges fairly identical entries.
*/
void consolidateLists() {
for (List<Interval> intervals : Arrays.asList(horizontalLines, verticalLines)) {
intervals.sort(null);
for (int i = 1; i < intervals.size();) {
if (intervals.get(i-1).combinableWith(intervals.get(i))) {
Interval interval = intervals.get(i-1).combineWith(intervals.get(i));
intervals.set(i-1, interval);
intervals.remove(i);
} else {
i++;
}
}
}
}
/**
* Checks whether the given color is black'ish.
*/
boolean isBlack(PDColor color) throws IOException {
int value = color.toRGB();
for (int i = 0; i < 2; i++) {
int component = value & 0xff;
if (component > 5)
return false;
value /= 256;
}
return true;
}
//
// PDFGraphicsStreamEngine overrides
//
@Override
public void appendRectangle(Point2D p0, Point2D p1, Point2D p2, Point2D p3) throws IOException {
path.add(new Rectangle(p0, p1, p2, p3));
}
@Override
public void endPath() throws IOException {
path.clear();
}
@Override
public void strokePath() throws IOException {
path.clear();
}
@Override
public void fillPath(int windingRule) throws IOException {
processPath();
}
@Override
public void fillAndStrokePath(int windingRule) throws IOException {
processPath();
}
@Override public void drawImage(PDImage pdImage) throws IOException { }
@Override public void clip(int windingRule) throws IOException { }
@Override public void moveTo(float x, float y) throws IOException { }
@Override public void lineTo(float x, float y) throws IOException { }
@Override public void curveTo(float x1, float y1, float x2, float y2, float x3, float y3) throws IOException { }
@Override public Point2D getCurrentPoint() throws IOException { return null; }
@Override public void closePath() throws IOException { }
@Override public void shadingFill(COSName shadingName) throws IOException { }
//
// inner classes
//
class Interval implements Comparable<Interval> {
final float from;
final float to;
Interval(float... values) {
Arrays.sort(values);
this.from = values[0];
this.to = values[values.length - 1];
}
Interval(double... values) {
Arrays.sort(values);
this.from = (float) values[0];
this.to = (float) values[values.length - 1];
}
boolean combinableWith(Interval other) {
if (this.from > other.from)
return other.combinableWith(this);
if (this.to < other.from)
return false;
float intersectionLength = Math.min(this.to, other.to) - other.from;
float thisLength = this.to - this.from;
float otherLength = other.to - other.from;
return (intersectionLength >= thisLength * .9f) || (intersectionLength >= otherLength * .9f);
}
Interval combineWith(Interval other) {
return new Interval(this.from, this.to, other.from, other.to);
}
@Override
public int compareTo(Interval o) {
return this.from == o.from ? Float.compare(this.to, o.to) : Float.compare(this.from, o.from);
}
@Override
public String toString() {
return String.format("[%3.2f, %3.2f]", from, to);
}
}
interface PathElement {
}
class Rectangle implements PathElement {
final Point2D p0, p1, p2, p3;
Rectangle(Point2D p0, Point2D p1, Point2D p2, Point2D p3) {
this.p0 = p0;
this.p1 = p1;
this.p2 = p2;
this.p3 = p3;
}
}
//
// members
//
final List<PathElement> path = new ArrayList<>();
final List<Interval> horizontalLines = new ArrayList<>();
final List<Interval> verticalLines = new ArrayList<>();
final Logger logger = LoggerFactory.getLogger(PdfBoxFinder.class);
}
示例使用
您可以像这样使用 PdfBoxFinder 从位于 FILE_PATH:
try ( PDDocument document = PDDocument.load(FILE_PATH) ) {
for (PDPage page : document.getDocumentCatalog().getPages()) {
PdfBoxFinder boxFinder = new PdfBoxFinder(page);
boxFinder.processPage(page);
PDFTextStripperByArea stripperByArea = new PDFTextStripperByArea();
for (Map.Entry<String, Rectangle2D> entry : boxFinder.getRegions().entrySet()) {
stripperByArea.addRegion(entry.getKey(), entry.getValue());
}
stripperByArea.extractRegions(page);
List<String> names = stripperByArea.getRegions();
names.sort(null);
for (String name : names) {
System.out.printf("[%s] %s\n", name, stripperByArea.getTextForRegion(name));
}
}
}
(ExtractBoxedText 测试 testExtractBoxedTexts)
输出开始：
[A1] Nr.
crt.
[A2] Nume şi prenume
[A3] Titlul lucrării
[A4] Coordonator ştiinţific
[A5] Ora
[B1] 1.
[B2] SFETCU I. JESSICA-
LARISA
[B3] Analiza fluxurilor de date twitter
[B4] Conf. univ. dr. Frîncu Marc
Eduard
[B5] 8:00
[C1] 2.
[C2] TARBA V. IONUȚ-
ADRIAN
[C3] Test me - rest api folosind java şi
play framework
[C4] Conf.univ.dr. Fortiş Teodor
Florin
[C5] 8:12
文档第一页：