Presentation is loading. Please wait.

Presentation is loading. Please wait.

2010/05/061 Author: Monga, O.; Deriche, R.; Malandain, G.; Cocquerez, J.P. Source: Pattern Recognition, 1990. Proceedings., 10th International Conference,

Similar presentations


Presentation on theme: "2010/05/061 Author: Monga, O.; Deriche, R.; Malandain, G.; Cocquerez, J.P. Source: Pattern Recognition, 1990. Proceedings., 10th International Conference,"— Presentation transcript:

1 2010/05/061 Author: Monga, O.; Deriche, R.; Malandain, G.; Cocquerez, J.P. Source: Pattern Recognition, 1990. Proceedings., 10th International Conference, Page(s): 652 - 654 Student: Jia – Hong Chen Advisor: Ku – Yaw Chang

2  Introduction  A 3D edge detection scheme  Closing 3D edges  Conclusion 2010/05/062

3  3D edge detectors are issued  a generalization in 3D of 2D edge detectors  we propose an unified formalism for 3D edge detection  recursive filters  tracking/closing algorithm 2010/05/063

4  Introduction  A 3D edge detection scheme  Closing 3D edges  Conclusion 2010/05/064

5  Canny 演算法  Step1: smoothing, 降低雜訊  Step2: 找邊緣, 求梯度的強度  Step3: hysteresis 2010/05/065

6  Step1: smoothing  使用 Gaussian filter, 在 Gaussian filter 函式中使用到 variance 來調整 Gaussian filter 的值 2010/05/066

7 7

8 8

9 9

10  Step2: 找邊緣, 求梯度的強度  使用 Sobel method 找出邊緣, 他會先計算出每一個像素 的值再利用 求出梯度  另外利用 求出梯度方向 2010/05/0610

11  1st Derivative (Gradient) 2010/05/0611

12 2010/05/0612

13 2010/05/0613

14 2010/05/0614

15 2010/05/0615

16  Step3: hysteresis  設定 2 個門檻值 T1 和 T2, 由 G 得到的結果, 任何像素如果 大於 T1 則被認定為邊緣像素, 而在這個邊緣像素鄰近的 點如果有大於 T2 也會被認定為邊緣像素. 2010/05/0616

17  Introduction  A 3D edge detection scheme  Closing 3D edges  Conclusion 2010/05/0617

18  It is often very difficult  select adequate thresholds for the thresholding stage  High thresholds allows to remove noisy edge points but also true edge points  low thresholds allow to obtain all true edge points but also noisy edge points 2010/05/0618

19  Generally it is easier  extend uncompleted contours than validate true contours in a noisy edge image  choose high thresholds  remove false edge points  use a tracking/closing algorithm 2010/05/0619

20  Deriche and Cocquerez  2D edge closing method proposed  supposes that it is possible to recognize endpoints of contours  Examination of a neighbourhood 3 x 3 of an edge point  The implementation of this algorithm is easy  if an edge point is identified as an extremity the algorithm is applied recursively to the involved extremity until a stop condition is verified 2010/05/0620

21  The 3D extension of this algorithm  applying it on each plane XY, YZ, ZX  adding the three edge images obtained  This can be justified by the assumption  the intersection of a 3D surface by at least one plane among XY,YZ and ZX 2010/05/0621

22  it remains some localized information lacks  cause holes of width 1 along X,Y or Z  To solve this problem  select the codes corresponding to holes of width 1  2D implementation consists in scanning the image  fill up each identified hole 2010/05/0622

23  Introduction  A 3D edge detection scheme  Closing 3D edges  Conclusion 2010/05/0623

24  We have proposed a 3D edge detection scheme  saving in computational effort is of great interest  Currently we investigate true 3D closing edge methods 2010/05/0624

25 2010/05/0625


Download ppt "2010/05/061 Author: Monga, O.; Deriche, R.; Malandain, G.; Cocquerez, J.P. Source: Pattern Recognition, 1990. Proceedings., 10th International Conference,"

Similar presentations


Ads by Google