Tesseract is one of the most popular open source Optical Character Recognition systems around. It supports many languages. It is written in C++ and needs a lot of other libraries as well to work. This blog assumes that you are already familiar with Tesseract and how it works.
We would use the bytedeco javacpp-presets to call Tesseract API from Java. This library comes with the needed binaries for the given platform. So, we would just declare the Maven dependency and pretty much done.
org.bytedeco tesseract-platform 4.1.1-1.5.3
Tesseract can be run in many modes. We will first see how we can detect lines in a given image.
try (TessBaseAPI api = new TessBaseAPI();) {
int returnCode = api.Init(tessDataDirectory, language);
if (returnCode != 0) {
throw new RuntimeException("could not initialize tesseract, error code: " + returnCode);
}
PIX image = pixRead(imagePath.toFile().getAbsolutePath());
LOGGER.info("The image has a width of {} and height of {}", image.w(), image.h());
api.SetImage(image);
BOXA boxes = api.GetComponentImages(tesseract.RIL_TEXTLINE, true, (PIXA) null, (IntBuffer) null);
LOGGER.info("Found {} textline image components.", boxes.n());
lines = IntStream.range(0, boxes.n()).mapToObj((int lineSequenceNumber) -> {
BOX box = boxaGetBox(boxes, lineSequenceNumber, L_CLONE);
api.SetRectangle(box.x(), box.y(), box.w(), box.h());
BytePointer ocrResult = api.GetUTF8Text();
String ocrLineText = ocrResult.getString().trim();
ocrResult.deallocate();
int confidence = api.MeanTextConf();
int x1 = box.x();
int y1 = box.y();
int width = box.w();
int height = box.h();
OcrWord lineTextBox = new OcrWord(x1, y1, x1 + width, y1 + height, confidence, ocrLineText, lineSequenceNumber + 1);
LOGGER.debug("lineTextBox: {}", lineTextBox);
return lineTextBox;
}).collect(Collectors.toList());
api.End();
api.close();
pixDestroy(image);
}
Next, we would see how to detect individual words with Tesseract.
try (TessBaseAPI api = new TessBaseAPI();) {
int returnCode = api.Init(tessDataDirectory, language);
if (returnCode != 0) {
throw new RuntimeException("could not initialize tesseract, error code: " + returnCode);
}
PIX image = pixRead(imagePath.toFile().getAbsolutePath());
api.SetImage(image);
int code = api.Recognize(new ETEXT_DESC());
if (code != 0) {
throw new IllegalArgumentException("could not recognize text");
}
try (ResultIterator ri = api.GetIterator();) {
int level = tesseract.RIL_WORD;
int wordSequenceNumber = 1;
Supplier intPointerSupplier = () -> new IntPointer(new int[1]);
do {
BytePointer ocrResult = ri.GetUTF8Text(level);
String ocrText = ocrResult.getString().trim();
float confidence = ri.Confidence(level);
IntPointer x1 = intPointerSupplier.get();
IntPointer y1 = intPointerSupplier.get();
IntPointer x2 = intPointerSupplier.get();
IntPointer y2 = intPointerSupplier.get();
boolean foundRectangle = ri.BoundingBox(level, x1, y1, x2, y2);
if (!foundRectangle) {
throw new IllegalArgumentException("Could not find any rectangle here");
}
OcrWord wordTextBox = new OcrWord(x1.get(), y1.get(), x2.get(), y2.get(), confidence, ocrText, wordSequenceNumber++);
LOGGER.info("wordTextBox: {}", wordTextBox);
words.add(wordTextBox);
x1.deallocate();
y1.deallocate();
x2.deallocate();
y2.deallocate();
ocrResult.deallocate();
} while (ri.Next(level));
ri.deallocate();
}
api.End();
api.deallocate();
pixDestroy(image);
}
https://github.com/paawak/blog/tree/master/code/tesseract-ocr/tesseract-java-demo
https://tesseract-ocr.github.io/tessdoc/APIExample
https://github.com/tesseract-ocr/tesseract
https://github.com/tesseract-ocr/tessdata
https://github.com/bytedeco/javacpp-presets
https://github.com/bytedeco/javacpp-presets/tree/master/tesseract