android - recognized text from tess-two library is wrong

Question

I am trying to use the tess-two library to recognize text from imagae.

Here is my code:

load.setOnClickListener(new View.OnClickListener() {

            @Override
            public void onClick(View v) {
                // recognize text
                Bitmap temp = loadJustTakenImage(); //loads taken image from sdcard
                Bitmap rotatedImage = rotateIfNeeded(temp); // rotate method i found in some tutorial
                String text1 = recognizeText(rotatedImage);
            }

        });

Recognize text method:

(tessdata folder is in Download with the eng.traineddata and other files)

private String recognizeText(Bitmap bitmap) {
        // TODO Auto-generated method stub
        TessBaseAPI baseApi = new TessBaseAPI();
        // DATA_PATH = Path to the storage
        // lang = for which the language data exists, usually "eng"
        baseApi.init(Environment.getExternalStorageDirectory().toString()
                + "/Download/", "eng");
        // Eg. baseApi.init("/mnt/sdcard/tesseract/tessdata/eng.traineddata",
        // "eng");
        baseApi.setImage(bitmap);
        String recognizedText = baseApi.getUTF8Text();
        baseApi.end();
        return recognizedText;
    }

rotate image method:

private Bitmap rotateIfNeeded(Bitmap bitmap) {
        ExifInterface exif = null;
        try {
            exif = new ExifInterface(directoryPath+"/"+currentFileName+".jpg");
        } catch (IOException e) {
            // TODO Auto-generated catch block
            e.printStackTrace();
        }
        int exifOrientation = exif
                .getAttributeInt(ExifInterface.TAG_ORIENTATION,
                        ExifInterface.ORIENTATION_NORMAL);

        int rotate = 0;

        switch (exifOrientation) {
        case ExifInterface.ORIENTATION_ROTATE_90:
            rotate = 90;
            break;
        case ExifInterface.ORIENTATION_ROTATE_180:
            rotate = 180;
            break;
        case ExifInterface.ORIENTATION_ROTATE_270:
            rotate = 270;
            break;
        }

        if (rotate != 0) {
            int w = bitmap.getWidth();
            int h = bitmap.getHeight();

            // Setting pre rotate
            Matrix mtx = new Matrix();
            mtx.preRotate(rotate);

            // Rotating Bitmap & convert to ARGB_8888, required by tess
            bitmap = Bitmap.createBitmap(bitmap, 0, 0, w, h, mtx, false);
        }
        bitmap = bitmap.copy(Bitmap.Config.ARGB_8888, true);
        return bitmap;
    }

the text I am getting is a real mess, for example

for this image:

enter image description here

I got this text:

,7‘

Sometimes I just get an empty String.

What am I doing wrong?

Did you make sure the rotation is producing a correct image? Paste the image to the screen and make sure. — Vaiden, May 01 '15 at 16:02
actually, I don't think the rotation do anything.. I try to recognize the text both from the original image and from the rotated image, and both Strings are the same weird text.. either weird text or empty.. is there s working code sample for rotation? — Ofek Agmon, May 02 '15 at 12:03
so, after testing again, the exifOrientation and rotate variables are always 0, even when I take a 45 angles picture. so rotation I think does nothing. but even when I take a straight photo, (when I think the rotation isn't needed), I still get weird text. ideas? — Ofek Agmon, May 02 '15 at 12:23
@OfekAgmon : Hi, Did you get any solution for the same? I'm having same problem — Jai, Sep 10 '15 at 09:42

score 4 · Answer 1 · answered Oct 05 '15 at 18:09

Based on the type of image on which you are trying to detect the characters, setting an appropriate Page segmentation mode will help detect the characters.

For example :

baseAPI.setPageSegMode(TessBaseAPI.PageSegMode.PSM_AUTO_ONLY);

The various other Page segmentation values are present in TessBaseApi.java :

/** Page segmentation mode. */
public static final class PageSegMode {
    /** Orientation and script detection only. */
    public static final int PSM_OSD_ONLY = 0;

    /** Automatic page segmentation with orientation and script detection. (OSD) */
    public static final int PSM_AUTO_OSD = 1;

    /** Fully automatic page segmentation, but no OSD, or OCR. */
    public static final int PSM_AUTO_ONLY = 2;

    /** Fully automatic page segmentation, but no OSD. */
    public static final int PSM_AUTO = 3;

    /** Assume a single column of text of variable sizes. */
    public static final int PSM_SINGLE_COLUMN = 4;

    /** Assume a single uniform block of vertically aligned text. */
    public static final int PSM_SINGLE_BLOCK_VERT_TEXT = 5;

    /** Assume a single uniform block of text. (Default.) */
    public static final int PSM_SINGLE_BLOCK = 6;

    /** Treat the image as a single text line. */
    public static final int PSM_SINGLE_LINE = 7;

    /** Treat the image as a single word. */
    public static final int PSM_SINGLE_WORD = 8;

    /** Treat the image as a single word in a circle. */
    public static final int PSM_CIRCLE_WORD = 9;

    /** Treat the image as a single character. */
    public static final int PSM_SINGLE_CHAR = 10;

    /** Find as much text as possible in no particular order. */
    public static final int PSM_SPARSE_TEXT = 11;

    /** Sparse text with orientation and script detection. */
    public static final int PSM_SPARSE_TEXT_OSD = 12;

    /** Number of enum entries. */
    public static final int PSM_COUNT = 13;
}

You can experiment with different page segmentation enum values and see which gives the best result.

For the above image, it seems like setting page segmentation to 'PSM_SINGLE_LINE' should yield the result you are looking for.

android - recognized text from tess-two library is wrong

1 Answers1

Linked