Presentation is loading. Please wait.

Presentation is loading. Please wait.

A split-and-merge framework for 2D shape summarization D. Gerogiannis, C. Nikou and A. Likas Department of Computer Science, University of Ioannina, Greece.

Similar presentations


Presentation on theme: "A split-and-merge framework for 2D shape summarization D. Gerogiannis, C. Nikou and A. Likas Department of Computer Science, University of Ioannina, Greece."— Presentation transcript:

1 A split-and-merge framework for 2D shape summarization D. Gerogiannis, C. Nikou and A. Likas Department of Computer Science, University of Ioannina, Greece

2 Presentation Outline Problem definition ▫Multiple line fitting on 2D unordered set of points Description of the proposed methodology ▫Line segments are represented by major axis of ellipses ▫An iterative split and merge algorithm Experimental results Conclusion and Future Work

3 Problem Definition Input: 2D unordered set of points describing a contour Output: A set of lines that describe the contour Point set may contain joints and inner structures or scattered data

4 Let be the points of the unordered set Let be the set of line segments of the model Then we can define the distortion Δ as: is the distance of from is 1 if corresponds to Iterative split and merge algorithm (I)

5 Iterative split and merge algorithm (II)

6 Iterative split and merge algorithm (III) The problem is augmented to a multiple lines fitting problem ▫Minimize distortion Δ Each line segment is modeled with an ellipse ▫Mean and covariance computed from the corresponding points A two step optimization method ▫Step1: Split (iterative) ▫Step2: Merge (iterative)

7 Split process (I) Split Step (time t) ▫The ellipse modeling the line segment (blue line) in time t ▫Split (split criterion)? ▫If yes then compute the new centers else keep the current ellipse After splitting (time t+1) ▫Update centers ▫The resulting ellipse is more eccentric (elongated)

8 Split process (II) Split Step (time t) After splitting (time t+1) Update centers ▫Select a direction - eigenvectors ▫Move by a step - eigenvalue ▫Nearest neighbor classification ▫Final centers and covariances of new clusters

9 When to split: Linearity Linearity: estimated by the minimum eigenvalue of the corresponding covariance matrix (Threshold T1) Trivial calculation

10 When to split: Connectivity Connectivity: the maximum distance (gap) between two successive points (Threshold T2) Demands ordering of points

11 Connectivity Computation Determines the split direction ▫one of the axis of the ellipse Defined for each axis and keep the maximum Computation ▫Project on the corresponding axis (scalar values) ▫Sort the projections ▫Find the maximum difference between two successive projections

12 When to split: The split criterion A set of points correspond to a line segment: ▫ if the corresponding ellipse is eccentric (linearity) ▫if it is tightly connected (connectivity) ▫Thresholds T1 (for linearity), T2 (for connectivity) Black lines: tightly connected points

13 Merge process Iterative process Merge collinear neighboring ellipses. Aims to reduce the complexity of the model. The split criterion is employeed. Merge two ellipses if the resulting ellipse DOES NOT satisfy the split criterion.

14 Demonstration of the method steps

15 Code available at: www.cs.uoi.gr/~dgerogia The proposed algorithm

16 Thresholds computation Computed from the data Experiments show that the final result is not strongly dependant on the value of α.

17 Experimental Results(I) Comparison with Gaussian Mixture Models and Hough Transform Silhouettes of objects (MPEG7) and fishes (GatorBait 100) Edge points extracted by Canny Edge Detector α = 0.6

18 Experimental Results(II) Our method GMM

19 Experimental Results(III) Hough Transform Our method

20 Conclusions A low level methodology. ▫Extracting shape features (oriented line segments) Does not necessitate ordering of points. It handles joints and multiple structures.

21 Future Work Improve the split criterion Improve the definition of the thresholds Extend to 3D sets of points

22 A split-and-merge framework for 2D shape summarization THANK YOU


Download ppt "A split-and-merge framework for 2D shape summarization D. Gerogiannis, C. Nikou and A. Likas Department of Computer Science, University of Ioannina, Greece."

Similar presentations


Ads by Google