Tesseract's interaction with JavaFX

[nagygr]

I'm developing a simple GUI application based on the Tesseract library. My application is working fine with Java Swing but when I change the GUI around it to use JavaFX (and leave all the OCR code unchanged), I get the following error:

libpng warning: Application built with libpng-1.5.30 but running with 1.6.37
Error in pixReadStreamPng: png_ptr not made
Error in pixReadStream: png: no pix returned
Error in pixRead: pix not read
Error in pixGetSpp: pix not defined
Error in pixGetDimensions: pix not defined
Error in pixGetColormap: pix not defined
Error in pixCopy: pixs not defined
Error in pixGetDepth: pix not defined
Error in pixGetWpl: pix not defined
Error in pixGetYRes: pix not defined
Error in pixClone: pixs not defined
Please call SetImage before attempting recognition.

The code that is invoking Tesseract is the following (in both implementations):

BytePointer outText;

TessBaseAPI api = new TessBaseAPI();
if (api.Init("tessdata", languageCode) != 0) {
	throw new RuntimeException("Could not initialize tesseract.");
}

updateMessage("Successfully initialized tesseract");
updateMessage("Starting OCR...");

int page = 0;
for (var imageFile: imageFiles) {
	updateMessage(String.format("Performing OCR on %s", imageFile));

	PIX image = pixRead(imageFile);
	api.SetImage(image);

	outText = api.GetUTF8Text();
	documentLines.addAll(
		Arrays.asList(
			new String(outText.getStringBytes(), StandardCharsets.UTF_8)
				.split("\n")
		)
	);

	outText.deallocate();
	pixDestroy(image);

	if (isCancelled()) {
		updateMessage("Cancelled");
		return null;
	}

	updateProgress(300 + (300 * (page + 1) / documentPages), 1000);
}

api.End();

where imageFiles is an ArrayList<String> with the paths of the image files and documentPages is the size of this list.

The relevant parts of the build.gradle file I'm using:

dependencies {
	// ...

	// OCR
	implementation 'org.bytedeco:tesseract-platform:4.1.1-1.5.4'

	// JavaFX
	runtimeOnly "org.openjfx:javafx-base:11:linux"
	runtimeOnly "org.openjfx:javafx-controls:11:linux"

	runtimeOnly "org.openjfx:javafx-base:11:win"
	runtimeOnly "org.openjfx:javafx-controls:11:win"
	
	runtimeOnly "org.openjfx:javafx-base:11:mac"
	runtimeOnly "org.openjfx:javafx-controls:11:mac"
}

javafx {
	modules = [ "javafx.controls" ]
	version = "11"
}

I guess, this is because JavaFX also bundles libpng, the two are of different versions and the linker finds the wrong version from Tesseract's perspective.

Is there a way to counter this problem?

Thank you very much in advance,
Greg

[saudet]

To fix this, we'll need to build Leptonica with libpng 1.6.x, but we'll lose compatibility with CentOS 6, see issue #680. However, since CentOS 6 has reached EOL, that's not so relevant anymore. Could you please send a pull request with an updated version of LIBPNG here?
https://github.com/bytedeco/javacpp-presets/blob/master/leptonica/cppbuild.sh#L14

[nagygr]

Thank you very much for the answer.

I have created a draft PR with the change: #1007

[nagygr]

Thank you very much for all the help. Everything works now with the SNAPSHOT build.

repositories {
	...
	maven { url 'https://oss.sonatype.org/content/repositories/snapshots' }
}

dependencies {
	...
	implementation "org.bytedeco:tesseract-platform:4.1.1-1.5.5-SNAPSHOT"
}