(Canny+算法)的Java實現(xiàn).doc

邊緣檢測算法的基本步驟 （1）濾波。邊緣檢測主要基于導數(shù)計算，但受噪聲影響。但濾波器在降低噪聲的同時也導致邊緣強度的損失。 （2）增強。增強算法將鄰域中灰度有顯著變化的點突出顯示。一般通過計算梯度幅值完成。 （3）檢測。但在有些圖象中梯度幅值較大的并不是邊緣點。最簡單的邊緣檢測是梯度幅值閾值判定。 （4）定位。精確確定邊緣的位置。Canny邊緣檢測算法 step1:用高斯濾波器平滑圖象； step2:用一階偏導的有限差分來計算梯度的幅值和方向； step3:對梯度幅值進行非極大值抑制； step4:用雙閾值算法檢測和連接邊緣。效果圖如下：代碼如下：package tools;import java.awt.*;import java.awt.image.*;public class EdgeDetector extends Component public EdgeDetector() threshold1 = 50;threshold2 = 230;setThreshold(128);setWidGaussianKernel(15);public void process() throws EdgeDetectorException if (threshold 255)throw new EdgeDetectorException(The value of the threshold is out of its valid range.);if (widGaussianKernel 40)throw new EdgeDetectorException(The value of the widGaussianKernel is out of its valid range.);width = sourceImage.getWidth(this);height = sourceImage.getHeight(this);picsize = width * height;data = new intpicsize;magnitude = new intpicsize;orientation = new intpicsize;float f = 1.0F;canny_core(f, widGaussianKernel);thresholding_tracker(threshold1, threshold2);for (int i = 0; i threshold)datai = 0xff000000;elsedatai = -1;edgeImage = pixels2image(data);data = null;magnitude = null;orientation = null;private void canny_core(float f, int i) boolean flag = false;boolean flag1 = false;derivative_mag = new intpicsize;float af4 = new floati;float af5 = new floati;float af6 = new floati;data = image2pixels(sourceImage);int k4 = 0;do if (k4 = i)break;float f1 = gaussian(k4, f);if (f1 = 2)break;float f2 = gaussian(float) k4 - 0.5F, f);float f3 = gaussian(float) k4 + 0.5F, f);float f4 = gaussian(k4, f * 0.5F);af4k4 = (f1 + f2 + f3) / 3F / (6.283185F * f * f);af5k4 = f3 - f2;af6k4 = 1.6F * f4 - f1;k4+; while (true);int j = k4;float af = new floatpicsize;float af1 = new floatpicsize;int j1 = width - (j - 1);int l = width * (j - 1);int i1 = width * (height - (j - 1);for (int l4 = j - 1; l4 j1; l4+) for (int l5 = l; l5 i1; l5 += width) int k1 = l4 + l5;float f8 = (float) datak1 * af40;float f10 = f8;int l6 = 1;int k7 = k1 - width;for (int i8 = k1 + width; l6 j; i8 += width) f8 += af4l6 * (float) (datak7 + datai8);f10 += af4l6 * (float) (datak1 - l6 + datak1 + l6);l6+;k7 -= width;afk1 = f8;af1k1 = f10;float af2 = new floatpicsize;for (int i5 = j - 1; i5 j1; i5+) for (int i6 = l; i6 i1; i6 += width) float f9 = 0.0F;int l1 = i5 + i6;for (int i7 = 1; i7 j; i7+)f9 += af5i7 * (afl1 - i7 - afl1 + i7);af2l1 = f9;af = null;float af3 = new floatpicsize;for (int j5 = k4; j5 width - k4; j5+) for (int j6 = l; j6 i1; j6 += width) float f11 = 0.0F;int i2 = j5 + j6;int j7 = 1;for (int l7 = width; j7 j; l7 += width) f11 += af5j7 * (af1i2 - l7 - af1i2 + l7);j7+;af3i2 = f11;af1 = null;j1 = width - j;l = width * j;i1 = width * (height - j);for (int k5 = j; k5 j1; k5+) for (int k6 = l; k6 = 256 ? 255 : k;float f13 = hypotenuse(af2k2, af3k2);float f14 = hypotenuse(af2l2, af3l2);float f15 = hypotenuse(af2i3, af3i3);float f16 = hypotenuse(af2j3, af3j3);float f18 = hypotenuse(af2l3, af3l3);float f20 = hypotenuse(af2j4, af3j4);float f19 = hypotenuse(af2i4, af3i4);float f17 = hypotenuse(af2k3, af3k3);float f5;if (f6 * f7 = Math.abs(f7)? (f5 = Math.abs(f6 * f12)= Math.abs(f7 * f18 - (f6 + f7) * f16)& f5 Math.abs(f7 * f19 - (f6 + f7) * f15) : (f5 = Math.abs(f7 * f12)= Math.abs(f6 * f18 - (f7 + f6) * f13)& f5 Math.abs(f6 * f19 - (f7 + f6) * f14) : Math.abs(f6)= Math.abs(f7)? (f5 = Math.abs(f6 * f12)= Math.abs(f7 * f20 + (f6 - f7) * f16)& f5 Math.abs(f7 * f17 + (f6 - f7) * f15) : (f5 = Math.abs(f7 * f12)= Math.abs(f6 * f20 + (f7 - f6) * f14)& f5 Math.abs(f6 * f17 + (f7 - f6) * f13) magnitudej2 = derivative_magj2;orientationj2 = (int) (Math.atan2(f7, f6) * (double) 40F);derivative_mag = null;af2 = null;af3 = null;private float hypotenuse(float f, float f1) if (f = 0.0F & f1 = 0.0F)return 0.0F;elsereturn (float) Math.sqrt(f * f + f1 * f1);private float gaussian(float f, float f1) return (float) Math.exp(-f * f) / (float) 2 * f1 * f1);private void thresholding_tracker(int i, int j) for (int k = 0; k picsize; k+)datak = 0;for (int l = 0; l width; l+) for (int i1 = 0; i1 = i)follow(l, i1, j);private boolean follow(int i, int j, int k) int j1 = i + 1;int k1 = i - 1;int l1 = j + 1;int i2 = j - 1;int j2 = i + j * width;if (l1 = height)l1 = height - 1;if (i2 = width)j1 = width - 1;if (k1 j1)break;int i1 = i2;do if (i1 l1)break;int k2 = l + i1 * width;if (i1 != j | l != i)& magnitudek2 = k& follow(l, i1, k) flag = true;break;i1+; while (true);if (!flag)break;l+;while (true);return true; else return false;private Image pixels2image(int ai) MemoryImageSource memoryimagesource =new MemoryImageSource(width,height,ColorModel.getRGBdefault(),ai,0,width);return Toolkit.getDefaultToolkit().createImage(memoryimagesource);private int image2pixels(Image image) int ai = new intpicsize;PixelGrabber pixelgrabber =new PixelGrabber(image, 0, 0, width, height, ai, 0, width);try pixelgrabber.grabPixels(); catch (InterruptedException interruptedexception) interruptedexception.printStackTrace();boolean flag = false;int k1 = 0;do if (k1 = 16)break;int i = (aik1 & 0xff0000) 16;int k = (aik1 & 0xff00) 8;int i1 = aik1 & 0xff;if (i != k | k != i1) flag = true;break;k1+; while (true);if (flag) for (int l1 = 0; l1 16;int l = (ail1 & 0xff00) 8;int j1 = ail1 & 0xff;ail1 =(int) (0.29799999999999999D * (double) j+ 0.58599999999999997D * (double) l+ 0.113D * (double) j1); else for (int i2 = 0; i2 picsize; i2+)aii2 = aii2 & 0xff;return ai;public void setSourceImage(Image image) sourceImage = image;public Image getEdgeImage() return edgeImage;public void setThreshold(int i) threshold = i;public void setWidGaussianKernel(int i) widGaussianKernel = i;final float ORIENT_SCALE = 40F;private int height;private int width;private int picsize;private int data;private int derivative_mag;private int magnitude;private int orientation;private Image sourceImage;private Image edgeImage;private int t