Example usage for org.opencv.core Mat rowRange

List of usage examples for org.opencv.core Mat rowRange

Introduction

In this page you can find the example usage for org.opencv.core Mat rowRange.

Prototype

public Mat rowRange(int startrow, int endrow) 

Source Link

Usage

From source file:karthik.Barcode.CandidateMatrixBarcode.java

License:Open Source License

CandidateResult NormalizeCandidateRegion(double angle) {
    /* candidateRegion is the RotatedRect which contains a candidate region for the barcode
     // angle is the rotation angle or USE_ROTATED_RECT_ANGLE for this function to 
     // estimate rotation angle from the rect parameter
     // returns Mat containing cropped area(region of interest) with just the barcode 
     // The barcode region is from the *original* image, not the scaled image
     // the cropped area is also rotated as necessary to be horizontal or vertical rather than skewed        
     // Some parts of this function are from http://felix.abecassis.me/2011/10/opencv-rotation-deskewing/
     // and http://stackoverflow.com/questions/22041699/rotate-an-image-without-cropping-in-opencv-in-c
     *//*from w ww . j  av a2s .  c o m*/

    double rotation_angle;
    CandidateResult result = new CandidateResult();

    // scale candidate region back up to original size to return cropped part from *original* image 
    // need the 1.0 there to force floating-point arithmetic from int values
    double scale_factor = img_details.src_original.rows() / (1.0 * img_details.src_grayscale.rows());

    // expand the region found - this helps capture the entire code including the border zone
    candidateRegion.size.width += 2 * params.RECT_WIDTH;
    candidateRegion.size.height += 2 * params.RECT_HEIGHT;

    // calculate location of rectangle in original image and its corner points
    RotatedRect scaledRegion = new RotatedRect(candidateRegion.center, candidateRegion.size,
            candidateRegion.angle);
    scaledRegion.center.x = scaledRegion.center.x * scale_factor;
    scaledRegion.center.y = scaledRegion.center.y * scale_factor;
    scaledRegion.size.height *= scale_factor;
    scaledRegion.size.width *= scale_factor;

    scaledRegion.points(img_details.scaledCorners);
    // lets get the coordinates of the ROI in the original image and save it

    result.ROI_coords = Arrays.copyOf(img_details.scaledCorners, 4);

    // get the bounding rectangle of the ROI by sorting its corner points
    // we do it manually because RotatedRect can generate corner points outside the Mat area
    Arrays.sort(img_details.scaledCorners, CandidateBarcode.get_x_comparator());
    int leftCol = (int) img_details.scaledCorners[0].x;
    int rightCol = (int) img_details.scaledCorners[3].x;
    leftCol = (leftCol < 0) ? 0 : leftCol;
    rightCol = (rightCol > img_details.src_original.cols() - 1) ? img_details.src_original.cols() - 1
            : rightCol;

    Arrays.sort(img_details.scaledCorners, CandidateBarcode.get_y_comparator());
    int topRow = (int) img_details.scaledCorners[0].y;
    int bottomRow = (int) img_details.scaledCorners[3].y;
    topRow = (topRow < 0) ? 0 : topRow;
    bottomRow = (bottomRow > img_details.src_original.rows() - 1) ? img_details.src_original.rows() - 1
            : bottomRow;

    Mat ROI_region = img_details.src_original.submat(topRow, bottomRow, leftCol, rightCol);

    // create a container that is a square with side = diagonal of ROI.
    // this is large enough to accommodate the ROI region with rotation without cropping it

    int orig_rows = bottomRow - topRow;
    int orig_cols = rightCol - leftCol;
    int diagonal = (int) Math.sqrt(orig_rows * orig_rows + orig_cols * orig_cols);

    int newWidth = diagonal + 1;
    int newHeight = diagonal + 1;

    int offsetX = (newWidth - orig_cols) / 2;
    int offsetY = (newHeight - orig_rows) / 2;

    Mat enlarged_ROI_container = new Mat(newWidth, newHeight, img_details.src_original.type());
    enlarged_ROI_container.setTo(ZERO_SCALAR);

    // copy ROI to centre of container and rotate it
    ROI_region.copyTo(enlarged_ROI_container.rowRange(offsetY, offsetY + orig_rows).colRange(offsetX,
            offsetX + orig_cols));
    Point enlarged_ROI_container_centre = new Point(enlarged_ROI_container.rows() / 2.0,
            enlarged_ROI_container.cols() / 2.0);
    Mat rotated = Mat.zeros(enlarged_ROI_container.size(), enlarged_ROI_container.type());

    if (angle == Barcode.USE_ROTATED_RECT_ANGLE)
        rotation_angle = estimate_barcode_orientation();
    else
        rotation_angle = angle;

    // perform the affine transformation
    img_details.rotation_matrix = Imgproc.getRotationMatrix2D(enlarged_ROI_container_centre, rotation_angle,
            1.0);
    img_details.rotation_matrix.convertTo(img_details.rotation_matrix, CvType.CV_32F); // convert type so matrix multip. works properly

    img_details.newCornerCoord.setTo(ZERO_SCALAR);

    // convert scaledCorners to contain locations of corners in enlarged_ROI_container Mat
    img_details.scaledCorners[0] = new Point(offsetX, offsetY);
    img_details.scaledCorners[1] = new Point(offsetX, offsetY + orig_rows);
    img_details.scaledCorners[2] = new Point(offsetX + orig_cols, offsetY);
    img_details.scaledCorners[3] = new Point(offsetX + orig_cols, offsetY + orig_rows);
    // calculate the new location for each corner point of the rectangle ROI after rotation
    for (int r = 0; r < 4; r++) {
        img_details.coord.put(0, 0, img_details.scaledCorners[r].x);
        img_details.coord.put(1, 0, img_details.scaledCorners[r].y);
        Core.gemm(img_details.rotation_matrix, img_details.coord, 1, img_details.delta, 0,
                img_details.newCornerCoord);
        updatePoint(img_details.newCornerPoints.get(r), img_details.newCornerCoord.get(0, 0)[0],
                img_details.newCornerCoord.get(1, 0)[0]);
    }
    rotated.setTo(ZERO_SCALAR);
    Imgproc.warpAffine(enlarged_ROI_container, rotated, img_details.rotation_matrix,
            enlarged_ROI_container.size(), Imgproc.INTER_CUBIC);
    // sort rectangles points in order by first sorting all 4 points based on x
    // we then sort the first two based on y and then the next two based on y
    // this leaves the array in order top-left, bottom-left, top-right, bottom-right
    Collections.sort(img_details.newCornerPoints, CandidateBarcode.get_x_comparator());
    Collections.sort(img_details.newCornerPoints.subList(0, 2), CandidateBarcode.get_y_comparator());
    Collections.sort(img_details.newCornerPoints.subList(2, 4), CandidateBarcode.get_y_comparator());

    // calc height and width of rectangular region

    double height = length(img_details.newCornerPoints.get(1), img_details.newCornerPoints.get(0));
    double width = length(img_details.newCornerPoints.get(2), img_details.newCornerPoints.get(0));

    // create destination points for warpPerspective to map to
    updatePoint(img_details.transformedPoints.get(0), 0, 0);
    updatePoint(img_details.transformedPoints.get(1), 0, height);
    updatePoint(img_details.transformedPoints.get(2), width, 0);
    updatePoint(img_details.transformedPoints.get(3), width, height);

    Mat perspectiveTransform = Imgproc.getPerspectiveTransform(
            Converters.vector_Point2f_to_Mat(img_details.newCornerPoints),
            Converters.vector_Point2f_to_Mat(img_details.transformedPoints));
    Mat perspectiveOut = Mat.zeros((int) height + 2, (int) width + 2, CvType.CV_32F);
    Imgproc.warpPerspective(rotated, perspectiveOut, perspectiveTransform, perspectiveOut.size(),
            Imgproc.INTER_CUBIC);

    result.ROI = perspectiveOut;
    return result;
}

From source file:qupath.opencv.classify.OpenCvClassifier.java

License:Open Source License

protected void createAndTrainClassifier() {

    // Create the required Mats
    int nMeasurements = measurements.size();

    Mat matTraining = new Mat(arrayTraining.length / nMeasurements, nMeasurements, CvType.CV_32FC1);
    matTraining.put(0, 0, arrayTraining);
    Mat matResponses = new Mat(arrayResponses.length, 1, CvType.CV_32SC1);
    matResponses.put(0, 0, arrayResponses);

    //      // Clear any existing classifier
    //      if (classifier != null)
    //         classifier.clear();

    logger.info("Training size: " + matTraining.size());
    logger.info("Responses size: " + matResponses.size());

    // Create & train the classifier
    try {//w  w w. j a  v a 2  s. com
        classifier = createClassifier();
        classifier.train(matTraining, Ml.ROW_SAMPLE, matResponses);
    } catch (CvException e) {
        // For reasons I haven't yet discerned, sometimes OpenCV throws an exception with the following message:
        // OpenCV Error: Assertion failed ((int)_sleft.size() < n && (int)_sright.size() < n) in calcDir, file /tmp/opencv320150620-1681-1u5iwhh/opencv-3.0.0/modules/ml/src/tree.cpp, line 1190
        // With one sample fewer, it can often recover... so attempt that, rather than failing miserably...
        //         logger.error("Classifier training error", e);
        logger.info("Will attempt retraining classifier with one sample fewer...");
        matTraining = matTraining.rowRange(0, matTraining.rows() - 1);
        matResponses = matResponses.rowRange(0, matResponses.rows() - 1);
        classifier = createClassifier();
        classifier.train(matTraining, Ml.ROW_SAMPLE, matResponses);
    }

    matTraining.release();
    matResponses.release();

    logger.info("Classifier trained with " + arrayResponses.length + " samples");
}