TrainingImage.java

package com.noga.njexl.testing.ocr;

import net.sourceforge.javaocr.scanner.PixelImage;

import java.util.logging.Logger;

/**
 * Class to hold a training image for a single character.
 * A training image is a representative image for a single character, and is
 * used to determine the likelihood that a given character image is a
 * particular character.
 * @author Ronald B. Cemer
 */
public class TrainingImage extends PixelImage
{

    /**
     * This is the maximum variance of aspect ratio between a training image
     * and the actual image segment to be decoded.  It is expressed as a
     * fraction of the calculated aspect ratios.  Any training image which
     * varies by more than this fraction from the aspect ratio of the character
     * to be decoded, is not used to decode that character.
     */
    public static final float ASPECT_RATIO_TOLERANCE = 0.3f;
    /**
     * This is the maximum variance between the source character's top white
     * space fraction and the training image's top white space fraction.  Any
     * training image which varies from the source character by more than this
     * tolerance, will not be considered a candidate.
     */
    public static final float TOP_WHITE_SPACE_FRACTION_TOLERANCE = 0.3f;
    /**
     * This is the maximum variance between the source character's bottom white
     * space fraction and the training image's bottom white space fraction.  Any
     * training image which varies from the source character by more than this
     * tolerance, will not be considered a candidate.
     */
    public static final float BOTTOM_WHITE_SPACE_FRACTION_TOLERANCE = 0.3f;
    /**
     * Fraction of the row arrayHeight which is occupied by complete whitespace above the character.
     */
    public final float topWhiteSpaceFraction;
    /**
     * Fraction of the row arrayHeight which is occupied by complete whitespace below the character.
     */
    public final float bottomWhiteSpaceFraction;
    private int myMaxX;
    private int myMaxY;
/// private int xShift;
/// private int yShift;

    /**
     * Construct a new <code>TrainingImage</code> object from an array of
     * gray scale pixels.
     * @param pixels An array of pixels in the range 0-255.
     * @param width The arrayWidth of the image.
     * @param height The arrayHeight of the image.
     * @param topWhiteSpacePixels The number of scan lines at the top of this character cell which
     * are all white.  These are excluded from the arrayWidth and arrayHeight of the training image.
     * @param bottomWhiteSpacePixels The number of scan lines at the bottom of this character cell
     * which are all white.  These are excluded from the arrayWidth and arrayHeight of the training image.
     */
    public TrainingImage(
            int[] pixels,
            int width,
            int height,
            int topWhiteSpacePixels,
            int bottomWhiteSpacePixels)
    {

        super(pixels, width, height);
        int rowHeight =
                topWhiteSpacePixels + height + bottomWhiteSpacePixels;
        topWhiteSpaceFraction =
                (float) topWhiteSpacePixels / (float) rowHeight;
        bottomWhiteSpaceFraction =
                (float) bottomWhiteSpacePixels / (float) rowHeight;
        myMaxX = width - 1;
        myMaxY = height - 1;
/// xShift = arrayWidth/20;
/// yShift = arrayHeight/20;
    }

    /**
     * Calculate the error factor between a block of pixels and our image.
     * @param theirPixels An array of grayscale pixels which contains the block to be compared
     * This should be in binary format, with each pixel having a value of either <code>0</code>
     * or <code>255</code>.
     * @param w The arrayWidth of the pixel array.
     * @param h The arrayHeight of the pixel array.
     * @param x1 The position of the left border of the rectangle to be compared.
     * @param y1 The position of the top border of the rectangle to be compared.
     * @param x2 The position of the right border of the rectangle to be compared.
     * Note that pixels up to, but not including this position, will be compared.
     * @param y2 The position of the bottom border of the rectangle to be compared.
     * Note that pixels up to, but not including this position, will be compared.
     * @return A <code>double</code> representing the average per-pixel mean square error.
     * Lower numbers indicate a better match.
     */
    public double calcMSE(int[] theirPixels, int w, int h, int x1, int y1, int x2, int y2)
    {
        int theirXRange = Math.max((x2 - x1) - 1, 1);
        int theirYRange = Math.max((y2 - y1) - 1, 1);
        int theirNPix = (theirXRange + 1) * (theirYRange + 1);
        boolean firstCompare = true;
        int myX, myY;
        long thisError, totalError;
        long minError = Long.MAX_VALUE;
        int myLineIdx, theirIdx;
///for (int yo = -yShift; yo <= yShift; yo++) {
/// for (int xo = -xShift; xo <= xShift; xo++) {
        totalError = 0L;
        for (int theirY = y1, yScan = 0 /*yo */;
             theirY < y2; theirY++, yScan++)
        {
            theirIdx = (theirY * w) + x1;
            myY = ((yScan * myMaxY) / theirYRange);
            myLineIdx = myY * width;
            for (int theirX = x1, xScan = 0 /*xo */;
                 theirX < x2; theirX++, theirIdx++, xScan++)
            {
                myX = ((xScan * myMaxX) / theirXRange);
                if ((myX < 0) || (myX > myMaxX) || (myY < 0) || (myY > myMaxY))
                {
                    thisError = theirPixels[theirIdx] - 255;
                }
                else
                {
                    thisError = theirPixels[theirIdx] - pixels[myLineIdx + myX];
                }
                totalError += (thisError * thisError);
            }
        }
        if ((firstCompare) || (totalError < minError))
        {
            minError = totalError;
        }
        firstCompare = false;
/// }
///}
        return Math.sqrt((double) minError) / (double) theirNPix;
    }
    private static final Logger LOG = Logger.getLogger(TrainingImage.class.getName());
}