Presentation is loading. Please wait.

Presentation is loading. Please wait.

GP-based Image Segmentation (GPIS) with Applications to Biomedical Image Segmentation (c) Louis Charbonneau and Nawwaf Kharma, 2009.

Similar presentations


Presentation on theme: "GP-based Image Segmentation (GPIS) with Applications to Biomedical Image Segmentation (c) Louis Charbonneau and Nawwaf Kharma, 2009."— Presentation transcript:

1 GP-based Image Segmentation (GPIS) with Applications to Biomedical Image Segmentation (c) Louis Charbonneau and Nawwaf Kharma, 2009

2 Problem statement (c) Louis Charbonneau and Nawwaf Kharma, 2009 How do we select an optimal sequence of low-level image operators (& parameters) to get the segmented image?

3 Segmentation example: cell nuclei (c) Louis Charbonneau and Nawwaf Kharma, 2009

4

5

6

7

8

9 Model description We use Cartesian GP: – Primitive operators are clearly defined, their right combination is the problem – CGP allows for an easy interpretation of the resulting sequence – Segmentation is a class of problems without one perfect solution; CGP can handle this (c) Louis Charbonneau and Nawwaf Kharma, 2009

10 System objectives Effectiveness: segmentation should be correct Efficiency: The smallest number of operations Transparency: operation sequences should be easy to understand (c) Louis Charbonneau and Nawwaf Kharma, 2009

11 System objectives (cont.) Segmentation should be doable without a priori information (except for training ground truths) Generality: effective on wide classes of images Ease of Use: Minimal human intervention (c) Louis Charbonneau and Nawwaf Kharma, 2009

12

13 Fitness criterion (c) Louis Charbonneau and Nawwaf Kharma, 2009

14 Fitness criterion (c) Louis Charbonneau and Nawwaf Kharma, 2009

15 Fitness criterion (c) Louis Charbonneau and Nawwaf Kharma, 2009

16 Crossover (c) Louis Charbonneau and Nawwaf Kharma, 2009

17 Mutations (I) (c) Louis Charbonneau and Nawwaf Kharma, 2009

18 Mutations (II) (c) Louis Charbonneau and Nawwaf Kharma, 2009

19

20 Data (c) Louis Charbonneau and Nawwaf Kharma, 2009 1026 images, 512 x 384 pixels 120 images, 340 x 780 pixels

21 System settings, database 1 (c) Louis Charbonneau and Nawwaf Kharma, 2009

22 Pixel segmentation accuracy, database 1 (c) Louis Charbonneau and Nawwaf Kharma, 2009

23 Cell segmentation accuracy, database 1 (c) Louis Charbonneau and Nawwaf Kharma, 2009

24 Statistical results, database 1 (c) Louis Charbonneau and Nawwaf Kharma, 2009

25 Example of evolved program, database 1 (c) Louis Charbonneau and Nawwaf Kharma, 2009

26 Example of evolved program, database 1 (c) Louis Charbonneau and Nawwaf Kharma, 2009

27 System settings, database 2 (c) Louis Charbonneau and Nawwaf Kharma, 2009

28 Pixel segmentation accuracy, database 2 (c) Louis Charbonneau and Nawwaf Kharma, 2009

29 Cell segmentation accuracy, database 2 (c) Louis Charbonneau and Nawwaf Kharma, 2009

30 Statistical results, database 2 (c) Louis Charbonneau and Nawwaf Kharma, 2009

31 Example of evolved program, database 2 (c) Louis Charbonneau and Nawwaf Kharma, 2009

32 Intermediate steps of evolved program, database 2 (c) Louis Charbonneau and Nawwaf Kharma, 2009

33 Intermediate steps of evolved program, database 2 (c) Louis Charbonneau and Nawwaf Kharma, 2009

34 Intermediate steps of evolved program, database 2 (c) Louis Charbonneau and Nawwaf Kharma, 2009

35 Superimposed input + evolved program (c) Louis Charbonneau and Nawwaf Kharma, 2009

36 GPIS on other types of images (c) Louis Charbonneau and Nawwaf Kharma, 2009 Lane detection tree detection

37 GPIS on other types of images (c) Louis Charbonneau and Nawwaf Kharma, 2009 Intra-cellular content of Wright-stained white blood cell images

38 Conclusion CGP was able to adapt to the complexity of input images: – A short program was evolved to solve the easy problem – a longer program was evolved to solve the harder problem Operator pool can be extended with specialized operators Injection was a reliable means of maintaining population diversity (c) Louis Charbonneau and Nawwaf Kharma, 2009

39 Conclusion A training window approach is very effective for operator refinement A small but accurate set of ground truths is enough to evolve segmentation algorithms without a priori information on the images (c) Louis Charbonneau and Nawwaf Kharma, 2009


Download ppt "GP-based Image Segmentation (GPIS) with Applications to Biomedical Image Segmentation (c) Louis Charbonneau and Nawwaf Kharma, 2009."

Similar presentations


Ads by Google