1. Efficient Color Boundary Detection with Color-opponent Mechanisms CVPR2013 Posters

2. Outline Introduction Approach Experiments Conclusions

3. Introduction

4. Propose a new framework for boundary detection in complex natural scenes based on the color-opponent mechanisms of the visual system. Image source: http://en.wikipedia.org/wiki/Opponent_process

5. Introduction One of the key limitations of opponent-based approaches is that they are blind to the luminance-defined boundaries. In order to obtain the complete contours of objects, these methods had to spend extra computational cost to combine more cues to detect luminance boundaries [3]. [3] D. R. Martin, C. C. Fowlkes, and J. Malik, "Learning to detect natural image boundaries using local brightness, color, and texture cues," IEEE Trans. on PAMI, vol. 26, pp. 530-549, 2004.

6. Introduction Simulate the biological mechanisms of color information processing along the Retina-LGN-Cortex visual pathway Image source: http://en.wikipedia.org/wiki/Opponent_process

7. Introduction Image source: [20] S. G. Solomon and P. Lennie, "The machinery of colour vision," Nature Reviews Neuroscience, vol. 8, pp. 276-286, 2007.

8. Introduction Color Mechanisms in the Visual System. Properties : 1. Trichromacy. 2. Two opponent channels. 3. Color opponency.

9. Approach Boundary Detection System : 1.Cone Layer 2.Ganglion/LGN Layer 3.Cortex Layer

10. A feedforward hierarchical system

11. 1.Cone Layer Type II cells in the ganglion/LGN layer is mainly for the perception of color region. Four channels: red (R), green (G), blue (B) and yellow (Y) components, where Y = (R+G)/2. Gaussian filters are used to simulate the receptive field of the cones in the retina. Outputs:

12. Approach Boundary Detection System : 1.Cone Layer 2.Ganglion/LGN Layer 3.Cortex Layer

13. 2.Ganglion/LGN Layer Single-opponent cells in ganglion/LGN layer are important for separating color and achromatic information,which is clearly shown by Equation 1. w1 > 0 and w2 < 0 response : R-on/G-off cells w1 0 response : R-off/G-on cells

14. Approach Boundary Detection System : 1.Cone Layer 2.Ganglion/LGN Layer 3.Cortex Layer

15. In the cortex layer of V1, the receptive fields of most color- and color-luminance-sensitive neurons are both chromatically and spatially opponent.

16. 3.Cortex Layer

17. The boundary responses at each orientation is given by (6)

18. 3.Cortex Layer The boundaries are detected in four channels (i.e., R+ wG, wR+ G, B+ wY and wB+Y ) with Equations 1-8.

19. Experiments

26. Conclusions 1. Presented a novel biologically plausible computational model for contour detection of color images. 2. Our model exhibits excellent capability of detecting both color and luminance boundaries synchronously in a time-saving manner.