boofcv.alg.feature.orientation.impl.ImplOrientationSlidingWindowIntegral Maven / Gradle / Ivy

Go to download
Show more of this group Show more artifacts with this name
Show all versions of boofcv-feature Show documentation
BoofCV is an open source Java library for real-time computer vision and robotics applications.
The newest version!
/*
 * Copyright (c) 2021, Peter Abeles. All Rights Reserved.
 *
 * This file is part of BoofCV (http://boofcv.org).
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *   http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

package boofcv.alg.feature.orientation.impl;

import boofcv.abst.feature.orientation.RegionOrientation;
import boofcv.alg.feature.orientation.OrientationIntegralBase;
import boofcv.struct.image.ImageGray;
import boofcv.struct.sparse.GradientValue;
import georegression.metric.UtilAngle;
import org.ddogleg.sorting.QuickSort_F64;

/**
 * 
 * Implementation of {@link boofcv.alg.feature.orientation.OrientationSlidingWindow} for integral images.
 * TODO comment how this implementation works
 * 
 *
 * @author Peter Abeles
 */
public class ImplOrientationSlidingWindowIntegral, G extends GradientValue>
		extends OrientationIntegralBase {
	// where the output from the derivative is stored
	double[] derivX;
	double[] derivY;

	// the size of the angle window it will consider in radians
	protected double windowSize;
	// the angle each pixel is pointing
	protected double[] angles;

	// clockwise ordering of angles
	protected int[] order;

	int total = 0;

	QuickSort_F64 sorter = new QuickSort_F64();

	/**
	 * Specifies configuration parameters and initializes data structures
	 *
	 * @param samplePeriod How often the image is sampled. This number is scaled. Typically 1.
	 * @param windowSize Angular window that is slide across
	 * @param sampleRadius Radius of the region being considered in terms of samples. Typically 6.
	 * @param weightSigma Sigma for weighting distribution. Zero for unweighted.
	 * @param sampleKernelWidth Size of kernel doing the sampling. Typically 4.
	 * @param integralType Type of integral image being processed.
	 */
	public ImplOrientationSlidingWindowIntegral( double radiusToScale, double samplePeriod, double windowSize,
												 int sampleRadius, double weightSigma, int sampleKernelWidth,
												 Class integralType ) {
		super(radiusToScale, sampleRadius, samplePeriod, sampleKernelWidth, weightSigma, true, integralType);
		this.windowSize = windowSize;

		derivX = new double[sampleWidth*sampleWidth];
		derivY = new double[sampleWidth*sampleWidth];

		angles = new double[sampleWidth*sampleWidth];
		order = new int[angles.length];
	}

	@Override
	public double compute( double c_x, double c_y ) {

		double period = scale*this.period;
		// top left corner of the region being sampled
		double tl_x = c_x - sampleRadius*period;
		double tl_y = c_y - sampleRadius*period;

		computeGradient(tl_x, tl_y, period);

		// apply weight to each gradient dependent on its position
		if (weights != null) {
			for (int i = 0; i < total; i++) {
				double w = weights.data[i];
				derivX[i] *= w;
				derivY[i] *= w;
			}
		}

		for (int i = 0; i < total; i++) {
			angles[i] = Math.atan2(derivY[i], derivX[i]);
		}

		// order points from lowest to highest
		sorter.sort(angles, 0, angles.length, order);

		return estimateAngle();
	}

	private void computeGradient( double tl_x, double tl_y, double samplePeriod ) {
		// add 0.5 to c_x and c_y to have it round when converted to an integer pixel
		// this is faster than the straight forward method
		tl_x += 0.5;
		tl_y += 0.5;

		total = 0;
		for (int y = 0; y < sampleWidth; y++) {

			for (int x = 0; x < sampleWidth; x++, total++) {

				int xx = (int)(tl_x + x*samplePeriod);
				int yy = (int)(tl_y + y*samplePeriod);

				if (g.isInBounds(xx, yy)) {
					GradientValue deriv = g.compute(xx, yy);
					double dx = deriv.getX();
					double dy = deriv.getY();

					derivX[total] = dx;
					derivY[total] = dy;
				} else {
					derivX[total] = 0;
					derivY[total] = 0;
				}
			}
		}
	}

	private double estimateAngle() {
		int start = 0;
		int end = 1;

		int startIndex = order[start];
		int endIndex = order[end];
		double sumX = derivX[startIndex];
		double sumY = derivY[startIndex];
		double best = sumX*sumX + sumY*sumY;
		double bestX = sumX;
		double bestY = sumY;

		double endAngle = angles[endIndex];

		while (start != total) {
			startIndex = order[start];
			double startAngle = angles[startIndex];

			// only compute the average if the angles are close to each other
			while (UtilAngle.dist(startAngle, endAngle) <= windowSize) {
				sumX += derivX[endIndex];
				sumY += derivY[endIndex];

				// see if the magnitude of the gradient inside this bound is greater
				// than the previous best
				double mag = sumX*sumX + sumY*sumY;
				if (mag > best) {
					best = mag;
					bestX = sumX;
					bestY = sumY;
				}
				end++;
				if (end >= total)
					end = 0;
				endIndex = order[end];
				endAngle = angles[endIndex];

				// if it cycled all the way around stop
				if (endIndex == startIndex)
					break;
			}

			// remove the first element from the list
			sumX -= derivX[startIndex];
			sumY -= derivY[startIndex];
			start++;
		}

		return Math.atan2(bestY, bestX);
	}

	@Override
	public RegionOrientation copy() {
		return new ImplOrientationSlidingWindowIntegral(
				objectRadiusToScale, period, windowSize, sampleRadius, weightSigma, kernelWidth, getImageType());
	}
}