邊緣檢測算法的基本步驟
（1）濾波。邊緣檢測主要基于導數計算，但受噪聲影響。但濾波器在降低噪聲的同時也導緻邊緣強度的損失。 （2）增強。增強算法将鄰域中灰階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門檻值判定。
（4）定位。精确确定邊緣的位置。
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:用雙門檻值算法檢測和連接配接邊緣。
效果圖如下：
代碼如下：
package tools; import java.awt.*; import java.awt.image.*; publicclass EdgeDetector extends Component { public EdgeDetector() { threshold1 = 50; threshold2 = 230; setThreshold(128); setWidGaussianKernel(15); } publicvoid process() throws EdgeDetectorException { if (threshold < 0 || threshold > 255) thrownew EdgeDetectorException("The value of the threshold is out of its valid range."); if (widGaussianKernel < 3 || widGaussianKernel > 40) thrownew EdgeDetectorException("The value of the widGaussianKernel is out of its valid range."); width = sourceImage.getWidth(this); height = sourceImage.getHeight(this); picsize = width * height; data = newint[picsize]; magnitude = newint[picsize]; orientation = newint[picsize]; float f = 1.0F; canny_core(f, widGaussianKernel); thresholding_tracker(threshold1, threshold2); for (int i = 0; i < picsize; i++) if (data[i] > threshold) data[i] = 0xff000000; else data[i] = -1; edgeImage = pixels2image(data); data = null; magnitude = null; orientation = null; } privatevoid canny_core(float f, int i) { boolean flag = false; boolean flag1 = false; derivative_mag = newint[picsize]; float af4[] = newfloat[i]; float af5[] = newfloat[i]; float af6[] = newfloat[i]; data = image2pixels(sourceImage); int k4 = 0; do{ if (k4 >= i) break; float f1 = gaussian(k4, f); if (f1 <= 0.005F && k4 >= 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); af4[k4] = (f1 + f2 + f3) / 3F / (6.283185F * f * f); af5[k4] = f3 - f2; af6[k4] = 1.6F * f4 - f1; k4++; }while (true); int j = k4; float af[] = newfloat[picsize]; float af1[] = newfloat[picsize]; 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) data[k1] * af4[0]; float f10 = f8; int l6 = 1; int k7 = k1 - width; for (int i8 = k1 + width; l6 < j; i8 += width) { f8 += af4[l6] * (float) (data[k7] + data[i8]); f10 += af4[l6] * (float) (data[k1 - l6] + data[k1 + l6]); l6++; k7 -= width; } af[k1] = f8; af1[k1] = f10; } } float af2[] = newfloat[picsize]; 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 += af5[i7] * (af[l1 - i7] - af[l1 + i7]); af2[l1] = f9; } } af = null; float af3[] = newfloat[picsize]; 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 += af5[j7] * (af1[i2 - l7] - af1[i2 + l7]); j7++; } af3[i2] = 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 < i1; k6 += width) { int j2 = k5 + k6; int k2 = j2 - width; int l2 = j2 + width; int i3 = j2 - 1; int j3 = j2 + 1; int k3 = k2 - 1; int l3 = k2 + 1; int i4 = l2 - 1; int j4 = l2 + 1; float f6 = af2[j2]; float f7 = af3[j2]; float f12 = hypotenuse(f6, f7); int k = (int) ((double) f12 * 20D); derivative_mag[j2] = k >= 256 ? 255 : k; float f13 = hypotenuse(af2[k2], af3[k2]); float f14 = hypotenuse(af2[l2], af3[l2]); float f15 = hypotenuse(af2[i3], af3[i3]); float f16 = hypotenuse(af2[j3], af3[j3]); float f18 = hypotenuse(af2[l3], af3[l3]); float f20 = hypotenuse(af2[j4], af3[j4]); float f19 = hypotenuse(af2[i4], af3[i4]); float f17 = hypotenuse(af2[k3], af3[k3]); float f5; if (f6 * f7 <= (float) 0 ? Math.abs(f6) >= 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)) { magnitude[j2] = derivative_mag[j2]; orientation[j2] = (int) (Math.atan2(f7, f6) * (double) 40F); } } } derivative_mag = null; af2 = null; af3 = null; } privatefloat hypotenuse(float f, float f1) { if (f == 0.0F && f1 == 0.0F) return 0.0F; else return (float) Math.sqrt(f * f + f1 * f1); } privatefloat gaussian(float f, float f1) { return (float) Math.exp((-f * f) / ((float) 2 * f1 * f1)); } privatevoid thresholding_tracker(int i, int j) { for (int k = 0; k < picsize; k++) data[k] = 0; for (int l = 0; l < width; l++) { for (int i1 = 0; i1 < height; i1++) if (magnitude[l + width * i1] >= i) follow(l, i1, j); } } privateboolean 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 < 0) i2 = 0; if (j1 >= width) j1 = width - 1; if (k1 < 0) k1 = 0; if (data[j2] == 0) { data[j2] = magnitude[j2]; boolean flag = false; int l = k1; do{ if (l > j1) break; int i1 = i2; do{ if (i1 > l1) break; int k2 = l + i1 * width; if ((i1 != j || l != i) && magnitude[k2] >= k && follow(l, i1, k)) { flag = true; break; } i1++; }while (true); if (!flag) break; l++; } while (true); returntrue; }else{ returnfalse; } } private Image pixels2image(int ai[]) { MemoryImageSource memoryimagesource = new MemoryImageSource( width, height, ColorModel.getRGBdefault(), ai, 0, width); return Toolkit.getDefaultToolkit().createImage(memoryimagesource); } privateint[] image2pixels(Image image) { int ai[] = newint[picsize]; 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 = (ai[k1] & 0xff0000) >> 16; int k = (ai[k1] & 0xff00) >> 8; int i1 = ai[k1] & 0xff; if (i != k || k != i1) { flag = true; break; } k1++; }while (true); if (flag) { for (int l1 = 0; l1 < picsize; l1++) { int j = (ai[l1] & 0xff0000) >> 16; int l = (ai[l1] & 0xff00) >> 8; int j1 = ai[l1] & 0xff; ai[l1] = (int) (0.29799999999999999D * (double) j + 0.58599999999999997D * (double) l + 0.113D * (double) j1); } }else{ for (int i2 = 0; i2 < picsize; i2++) ai[i2] = ai[i2] & 0xff; } return ai; } publicvoid setSourceImage(Image image) { sourceImage = image; } public Image getEdgeImage() { return edgeImage; } publicvoid setThreshold(int i) { threshold = i; } publicvoid setWidGaussianKernel(int i) { widGaussianKernel = i; } finalfloat ORIENT_SCALE = 40F; privateint height; privateint width; privateint picsize; privateint data[]; privateint derivative_mag[]; privateint magnitude[]; privateint orientation[]; private Image sourceImage; private Image edgeImage; privateint threshold1; privateint threshold2; privateint threshold; privateint widGaussianKernel; } //second file package tools; publicclass EdgeDetectorException extends Exception { public EdgeDetectorException() { //do something? } public EdgeDetectorException(String s) { super(s); } } //使用示例
EdgeDetector edgeDetector=new EdgeDetector(); edgeDetector.setSourceImage(sourceImage); edgeDetector.setThreshold(128); edgeDetector.setWidGaussianKernel(5); try{ edgeDetector.process(); } catch(EdgeDetectorException e) { System.out.println(e.getMessage()); } Image edgeImage=edgeDetector.getEdgeImage(); yourPanel.show(edgeImage); 轉自：http://blog.csdn.net/Haohappy2004/article/details/476820 ![第六章 層的濾鏡[圖]](data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=)