Train to recognize additional symbols

Train user patterns

To train a custom symbol pattern, you need a symbol image.

You should set the smallSymbolHeight and BaseLine properties to correct values. Otherwise, the trained pattern fails to function properly.

• smallSymbolHeight specifies the height of small characters in pixels on the source image. By default, the value is 0.

• BaseLine contains the distance from the baseline to the top edge of the cropped image of the character. The baseline is the line on which the characters are located. The top edge of the image is determined by the character orientation. By default, the value is 0.

Example:

In the currency image below, BaseLine=31, smallSymbolHeight=22.

Sample code

.....
private void processWithEngine() {
    try {
        displayMessage( "Training user pattern..." );
        // Setup FREngine
        setupFREngine();
        //create pattern
        ITrainingImagesCollection imagesCollection = engine.CreateTrainingImagesCollection();
        ITrainingImage image = loadImage("temp_images/costa_rica_currency.bmp", 31, 22);
        imagesCollection.Add(image);
        String filename = "costa_rica_currency.ptn";
        engine.CreateEmptyUserPattern(filename);
        engine.TrainUserPattern(filename, imagesCollection, "₡", 0, TextTypeEnum.TT_Normal);
        image.Release();
        imagesCollection.Release();
    } catch( Exception ex ) {
        displayMessage( ex.getMessage() );
    }
}

private ITrainingImage loadImage(String filename, int baseLine, int smallSymbolHeight){
    ITrainingImage image = engine.CreateTrainingImage();
    image.setBaseLine(baseLine);
    image.setsmallSymbolHeight(smallSymbolHeight);
    IImageDocument imageDoc = engine.OpenImageFile(filename, null, null, null);
    IImage bwImage = imageDoc.getBlackWhiteImage();
    IRegion region = engine.CreateRegion();
    region.AddRect(0, 0, bwImage.getWidth(), bwImage.getHeight());
    image.SetImageData(imageDoc, region);

    region.Release();
    imageDoc.Release();
    return image;
}
.....

Test code

You can look at samples in /opt/ABBYY/FREngine11/Samples/Java on the OCR server:

1. Create a sample folder with a Java file.
2. Export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/ABBYY/FREngine11/Bin.
3. Execute build.sh.
4. Execute run.sh.

Look at the /ocr/ABBYY/FREngine11/Samples/Java/TrainUserPattern2 example.

Prevent memory leaks

The ABBYY FineReader Engine provides Java support through JNI (Java Native Interface). JNI allows running native code from Java using a special wrapper in C++.

In this case, pay attention to releasing the objects created with engine methods.

Example:

IFRDocument document = engine.CreateFRDocument();
...
FRPages pages = document.getPages();
for(int i=0; i < pages.getCount(); i++) {
    FRPage page = pages.getElement(i);
    ILayout layout = page.getLayout();
    ILayoutBlocks blocks = layout.getBlocks();
    .....
    .....
    .....
    page.Release();
    layout.Release();
    blocks.Release();
}
pages.Release();
.....
ITextExportParams txtParams = engine.CreateTextExportParams();
.....
txtParams.Release();
.....
document.Release();

To make sure all objects are releasing correctly, enable the debug mode:

engine.StartLogging("filename.log", true); 

Then, check the log file to check that the following message does not appear there:

Not all objects were released before DeinitializeEngine()