List of usage examples for org.opencv.core MatOfPoint2f height
public int height()
From source file:ac.robinson.ticqr.TickBoxImageParserTask.java
License:Apache License
@Override protected ArrayList<PointF> doInBackground(Void... unused) { Log.d(TAG, "Searching for tick boxes of " + mBoxSize + " size"); // we look for *un-ticked* boxes, rather than ticked, as they are uniform in appearance (and hence easier to // detect) - they show up as a box within a box ArrayList<PointF> centrePoints = new ArrayList<>(); int minimumOuterBoxArea = (int) Math.round(Math.pow(mBoxSize, 2)); int maximumOuterBoxArea = (int) Math.round(Math.pow(mBoxSize * 1.35f, 2)); int minimumInnerBoxArea = (int) Math.round(Math.pow(mBoxSize * 0.5f, 2)); // image adjustment - blurSize, blurSTDev and adaptiveThresholdSize must not be even numbers int blurSize = 9; int blurSTDev = 3; int adaptiveThresholdSize = Math.round(mBoxSize * 3); // (oddness ensured below) int adaptiveThresholdC = 4; // value to add to the mean (can be negative or zero) adaptiveThresholdSize = adaptiveThresholdSize % 2 == 0 ? adaptiveThresholdSize + 1 : adaptiveThresholdSize; // how similar the recognised polygon must be to its actual contour - lower is more similar float outerPolygonSimilarity = 0.045f; float innerPolygonSimilarity = 0.075f; // don't require as much accuracy for the inner part of the tick box // how large the maximum internal angle can be (e.g., for checking square shape) float maxOuterAngleCos = 0.3f; float maxInnerAngleCos = 0.4f; // use OpenCV to recognise boxes that have a box inside them - i.e. an un-ticked tick box // see: http://stackoverflow.com/a/11427501 // Bitmap newBitmap = mBitmap.copy(Bitmap.Config.RGB_565, true); // not needed Mat bitMat = new Mat(); Utils.bitmapToMat(mBitmap, bitMat);// ww w . jav a2 s .c o m // blur and convert to grey // alternative (less flexible): Imgproc.medianBlur(bitMat, bitMat, blurSize); Imgproc.GaussianBlur(bitMat, bitMat, new Size(blurSize, blurSize), blurSTDev, blurSTDev); Imgproc.cvtColor(bitMat, bitMat, Imgproc.COLOR_RGB2GRAY); // need 8uC1 (1 channel, unsigned char) image type // perform adaptive thresholding to detect edges // alternative (slower): Imgproc.Canny(bitMat, bitMat, 10, 20, 3, false); Imgproc.adaptiveThreshold(bitMat, bitMat, 255, Imgproc.ADAPTIVE_THRESH_GAUSSIAN_C, Imgproc.THRESH_BINARY, adaptiveThresholdSize, adaptiveThresholdC); // get the contours in the image, and their hierarchy Mat hierarchyMat = new Mat(); List<MatOfPoint> contours = new ArrayList<>(); Imgproc.findContours(bitMat, contours, hierarchyMat, Imgproc.RETR_TREE, Imgproc.CHAIN_APPROX_SIMPLE); if (DEBUG) { Imgproc.drawContours(bitMat, contours, -1, new Scalar(30, 255, 255), 1); } // parse the contours and look for a box containing another box, with similar enough sizes int numContours = contours.size(); ArrayList<Integer> searchedContours = new ArrayList<>(); Log.d(TAG, "Found " + numContours + " possible tick box areas"); if (numContours > 0 && !hierarchyMat.empty()) { for (int i = 0; i < numContours; i++) { // the original detected contour MatOfPoint boxPoints = contours.get(i); // hierarchy key: 0 = next sibling num, 1 = previous sibling num, 2 = first child num, 3 = parent num int childBox = (int) hierarchyMat.get(0, i)[2]; // usually the largest child (as we're doing RETR_TREE) if (childBox == -1) { // we only want elements that have children continue; } else { if (searchedContours.contains(childBox)) { if (DEBUG) { Log.d(TAG, "Ignoring duplicate box at first stage: " + childBox); } continue; } else { searchedContours.add(childBox); } } // discard smaller (i.e. noise) outer box areas as soon as possible for speed // used to do Imgproc.isContourConvex(outerPoints) later, but the angle check covers this, so no need double originalArea = Math.abs(Imgproc.contourArea(boxPoints)); if (originalArea < minimumOuterBoxArea) { // if (DEBUG) { // drawPoints(bitMat, boxPoints, new Scalar(255, 255, 255), 1); // Log.d(TAG, "Outer box too small"); // } continue; } if (originalArea > maximumOuterBoxArea) { // if (DEBUG) { // drawPoints(bitMat, boxPoints, new Scalar(255, 255, 255), 1); // Log.d(TAG, "Outer box too big"); // } continue; } // simplify the contours of the outer box - we want to detect four-sided shapes only MatOfPoint2f boxPoints2f = new MatOfPoint2f(boxPoints.toArray()); // Point2f for approxPolyDP Imgproc.approxPolyDP(boxPoints2f, boxPoints2f, outerPolygonSimilarity * Imgproc.arcLength(boxPoints2f, true), true); // simplify the contour if (boxPoints2f.height() != 4) { // height is number of points if (DEBUG) { // drawPoints(bitMat, new MatOfPoint(boxPoints2f.toArray()), new Scalar(255, 255, 255), 1); Log.d(TAG, "Outer box not 4 points"); } continue; } // check that the simplified outer box is approximately a square, angle-wise org.opencv.core.Point[] boxPointsArray = boxPoints2f.toArray(); double maxCosine = 0; for (int j = 0; j < 4; j++) { org.opencv.core.Point pL = boxPointsArray[j]; org.opencv.core.Point pIntersect = boxPointsArray[(j + 1) % 4]; org.opencv.core.Point pR = boxPointsArray[(j + 2) % 4]; getLineAngle(pL, pIntersect, pR); maxCosine = Math.max(maxCosine, getLineAngle(pL, pIntersect, pR)); } if (maxCosine > maxOuterAngleCos) { if (DEBUG) { // drawPoints(bitMat, new MatOfPoint(boxPoints2f.toArray()), new Scalar(255, 255, 255), 1); Log.d(TAG, "Outer angles not square enough"); } continue; } // check that the simplified outer box is approximately a square, line length-wise double minLine = Double.MAX_VALUE; double maxLine = 0; for (int p = 1; p < 4; p++) { org.opencv.core.Point p1 = boxPointsArray[p - 1]; org.opencv.core.Point p2 = boxPointsArray[p]; double xd = p1.x - p2.x; double yd = p1.y - p2.y; double lineLength = Math.sqrt((xd * xd) + (yd * yd)); minLine = Math.min(minLine, lineLength); maxLine = Math.max(maxLine, lineLength); } if (maxLine - minLine > minLine) { if (DEBUG) { // drawPoints(bitMat, new MatOfPoint(boxPoints2f.toArray()), new Scalar(255, 255, 255), 1); Log.d(TAG, "Outer lines not square enough"); } continue; } // draw the outer box if debugging if (DEBUG) { MatOfPoint debugBoxPoints = new MatOfPoint(boxPointsArray); Log.d(TAG, "Potential tick box: " + boxPoints2f.size() + ", " + "area: " + Math.abs(Imgproc.contourArea(debugBoxPoints)) + " (min:" + minimumOuterBoxArea + ", max:" + maximumOuterBoxArea + ")"); drawPoints(bitMat, debugBoxPoints, new Scalar(50, 255, 255), 2); } // loop through the children - they should be in descending size order, but sometimes this is wrong boolean wrongBox = false; while (true) { if (DEBUG) { Log.d(TAG, "Looping with box: " + childBox); } // we've previously tried a child - try the next one // key: 0 = next sibling num, 1 = previous sibling num, 2 = first child num, 3 = parent num if (wrongBox) { childBox = (int) hierarchyMat.get(0, childBox)[0]; if (childBox == -1) { break; } if (searchedContours.contains(childBox)) { if (DEBUG) { Log.d(TAG, "Ignoring duplicate box at loop stage: " + childBox); } break; } else { searchedContours.add(childBox); } //noinspection UnusedAssignment wrongBox = false; } // perhaps this is the outer box - check its child has no children itself // (removed so tiny children (i.e. noise) don't mean we mis-detect an un-ticked box as ticked) // if (hierarchyMat.get(0, childBox)[2] != -1) { // continue; // } // check the size of the child box is large enough boxPoints = contours.get(childBox); originalArea = Math.abs(Imgproc.contourArea(boxPoints)); if (originalArea < minimumInnerBoxArea) { if (DEBUG) { // drawPoints(bitMat, boxPoints, new Scalar(255, 255, 255), 1); Log.d(TAG, "Inner box too small"); } wrongBox = true; continue; } // simplify the contours of the inner box - again, we want four-sided shapes only boxPoints2f = new MatOfPoint2f(boxPoints.toArray()); Imgproc.approxPolyDP(boxPoints2f, boxPoints2f, innerPolygonSimilarity * Imgproc.arcLength(boxPoints2f, true), true); if (boxPoints2f.height() != 4) { // height is number of points // if (DEBUG) { // drawPoints(bitMat, boxPoints, new Scalar(255, 255, 255), 1); // } Log.d(TAG, "Inner box fewer than 4 points"); // TODO: allow > 4 for low quality images? wrongBox = true; continue; } // check that the simplified inner box is approximately a square, angle-wise // higher tolerance because noise means if we get several inners, the box may not be quite square boxPointsArray = boxPoints2f.toArray(); maxCosine = 0; for (int j = 0; j < 4; j++) { org.opencv.core.Point pL = boxPointsArray[j]; org.opencv.core.Point pIntersect = boxPointsArray[(j + 1) % 4]; org.opencv.core.Point pR = boxPointsArray[(j + 2) % 4]; getLineAngle(pL, pIntersect, pR); maxCosine = Math.max(maxCosine, getLineAngle(pL, pIntersect, pR)); } if (maxCosine > maxInnerAngleCos) { Log.d(TAG, "Inner angles not square enough"); wrongBox = true; continue; } // this is probably an inner box - log if debugging if (DEBUG) { Log.d(TAG, "Un-ticked inner box: " + boxPoints2f.size() + ", " + "area: " + Math.abs(Imgproc.contourArea(new MatOfPoint2f(boxPointsArray))) + " (min: " + minimumInnerBoxArea + ")"); } // find the inner box centre double centreX = (boxPointsArray[0].x + boxPointsArray[1].x + boxPointsArray[2].x + boxPointsArray[3].x) / 4f; double centreY = (boxPointsArray[0].y + boxPointsArray[1].y + boxPointsArray[2].y + boxPointsArray[3].y) / 4f; // draw the inner box if debugging if (DEBUG) { drawPoints(bitMat, new MatOfPoint(boxPointsArray), new Scalar(255, 255, 255), 1); Core.circle(bitMat, new org.opencv.core.Point(centreX, centreY), 3, new Scalar(255, 255, 255)); } // add to the list of boxes to check centrePoints.add(new PointF((float) centreX, (float) centreY)); break; } } } Log.d(TAG, "Found " + centrePoints.size() + " un-ticked boxes"); return centrePoints; }