a robust approach for local interest point detection in line-drawing images

21
A Robust Approach for Local Interest Point Detection in Line- Drawing Images 1 The Anh Pham, Mathieu Delalandre, Sabine Barrat and Jean-Yves Ramel RFAI group- Polytech’Tour, France. CIL Talk Wednesday 7 th March 2012 Athens, Greece

Upload: amandla

Post on 15-Feb-2016

24 views

Category:

Documents


0 download

DESCRIPTION

A Robust Approach for Local Interest Point Detection in Line-Drawing Images . The Anh Pham, Mathieu Delalandre , Sabine Barrat and Jean-Yves Ramel RFAI group- Polytech’Tour , France. CIL Talk Wednesday 7 th March 2012 Athens , Greece. Overview. Introduction - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

A Robust Approach for Local Interest Point Detection in Line-

Drawing Images

1

The Anh Pham, Mathieu Delalandre, Sabine Barrat and Jean-Yves RamelRFAI group- Polytech’Tour, France.

CIL TalkWednesday 7th March 2012

Athens, Greece

Page 2: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

2

Overview

Introduction Junction detection in line-drawing images Experiments and results Conclusion and future works

Page 3: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

Introduction (1)

This work is interested with graphic documents, especially the line drawings, some examples

3

Page 4: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

Introduction (2)

Interest points are a kind of local features (i.e. an image pattern which differs from its immediate neighborhood).

Popular interest points include edges, blobs, regions, salient points, etc. In graphics documents, interest points are end-points, corners and junctions:

4

Local interest points

Approach Corner Junction RobustnessHigh curvature detection

[The-Chin89] ++++

Intensity-based methods [Harris89] ++

Model-based methods[Chul05] +++

Segmentation-based methods [Burge98] ++

Contour matching methods [Ramel00] ++

Tracking methods [Song02] ++

Comparison of the approaches for corner and junction detection

Page 5: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

Introduction (3)5

Key idea of the work is to drive high curvature detection methods to achieve junction detection.Two problems:

(1) How to extract the curves(2) How to merge the multiple detections

High curvature detection is the task of segmenting a curve at distinguished points of high local curvature (e.g. corners, bends, joints).

High curvature detection methods often includes include polygonal and B-splines approximation, wavelet analysis, etc.

Page 6: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

6

Overview

Introduction Junction detection in line-drawing images Experiments and results Conclusion and future works

Page 7: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

7

Junction detection in line-drawing images (1)

Flow-work of our approach

Page 8: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

8

Junction detection in line-drawing images (2)

Skeletonization, branch linking

High curvature detection

Path extraction

1D signals

Skeleton graph

Path representation

2D paths

Refining & Correcting

Candidates

(1) Skeletonization based on Di Baja (3,4)-chamfer distance [DiBaja94]

(2) Branch linking and Skeleton Connective Graph Construction (SCG) based on [Popel02]

Skeleton Connective Graph (SCG):

node: ended and crossing points

edge: skeleton branch

Page 9: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

9

Junction detection in line-drawing images (3)

Path definition: a sequence of edges of SCG that describes a complete stroke or a circuit.Three types of paths: Stroke path, Circuit path and Hybrid path.

Paths are extracted using anticlockwise direction between the nodes of graph SCG:

A skeleton graph

A stroke path A circuit path

Skeletonization, branch linking

High curvature detection

Path extraction

1D signals

Skeleton graph

Path representation

2D paths

Refining & Correcting

Candidatesd0

are branch pixelsare branch extremitiesis a crossing pixel

d0 is the extremity-crossing direction

Page 10: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

Junction detection in line-drawing images (4)

Skeletonization, branch linking

High curvature detection

Path extraction

1D signals

Skeleton graph

Path representation

2D paths

Refining & Correcting

Candidates

A 2D path P consists in N points: (x1y1), (x2y2),…,(xNyN) To represent a 2D path in 1D signal, we selected the Rosenfeld-Johnston method:

10

pi-q

pi pi+q

tqa

tqb

tqtq

tqtq

ba

baar

cos

,, qttqtttq yyxxa

,, qttqtttq yyxxb

tqtq

tqtq

ba

batf

)cos()(

pI-q pI pI+qtqa

tqb

f(t)straight-line -1high curvature

/2 0 pI-q pI

pI+q

tqa

tqb

Page 11: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

Junction detection in line-drawing images (5)

Skeletonization, branch linking

High curvature detection

Path extraction

1D signals

Skeleton graph

Path representation

2D paths

Refining & Correcting

Candidates

Due to the q parameter, we must make the method shift invariant.To do so, we select starting point of lowest curvature i.e. f(t)-

1

11

A good starting point here (shift-invariant).

Not goodstarting point.

Page 12: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

12

Skeletonization, branch linking

High curvature detection

Path extraction

1D signals

Skeleton graph

Path representation

2D paths

Refining & Correcting

Candidates

Using multi-resolution wavelet analysis because of its robustness and scale invariance (i.e. multi-resolution)[Gao06].

Junction detection in line-drawing images (6)

1D representationImage (I) 2D curcuit path

Multi-resolution wavelet analysis

Page 13: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

13

(1) Single path level: Remove the “unreliable” segments (i.e. length less than line thickness) and Connect the “reliable” segments togethers.

(2) Inter-path level (using voting scheme): merging close junctions together based on line thickness.

Junction detection in line-drawing images (7)

Skeletonization, branch linking

High curvature detection

Path extraction

Skeleton graph

Path representation

2D paths

Refining & Correcting

Candidates

1D signals

a path with high curvature points

a SCG with high curvature points

result after removing short segments

Page 14: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

14

Overview

Introduction Junction detection in line-drawing images Experiments and results Conclusion and future works

Page 15: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

15

Experiments and Results (1)

Evaluation protocol:Evaluation Criteria is the repeatability score [Schmid00]

2

2

p is a model pointq is a detected point

pq

Detection of p ispositive if d(p,q)<with d(p,q) the Euclidean distance

Page 16: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

16

Experiments and Results (2)

Datasets:

Logos-UMD ISRC2011Models 106 150Degradation

Rotation + Scaling + Kanungo noise

Test images 1272 3600

Page 17: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

17

Experiments and Results (3)

Some results

+ Liu99: “Identification of Fork point on the Skeletons of Handwritten Chinese Characters”, PAMI (1999).+ Haris detector: “A combined corner and edge detector”. Alvey Vision Conference, (1988).

Page 18: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

18

Experiments and Results (4)

Some visual results

Page 19: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

19

Overview

Introduction Junction detection in line-drawing images Experiments and results Conclusion and future works

Page 20: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

Conclusions and future works

Conclusions: A junction detector is proposed for line-drawing images The obtained results are rather promising

Future works The method is threshold dependent, we are looking for threshold adaptation

(e.g. region of support Improve the robustness of the merging step using topological analysis

(e.g. line bending energy minimization) More experiments with more interest points detector and datasets Applications of recognition of spotting (logos, symbols) and image indexing

20

Page 21: A Robust Approach for Local Interest Point Detection in Line-Drawing Images

Thank you for your attention!