Outline • Perceptual organization, grouping, and segmentation – Active contours and deformable templates File: week14-m.ppt Introduction • Segmentation – Roughly speaking, segmentation is to partition the images into meaningful parts that are relatively homogenous in certain sense 5/29/2016 Visual Perception Modeling 2 Introduction – cont. • Computational models/implementations – There are generally two kinds of computational models/implementations for segmentation • Based on homogeneity measure to group pixels with similar attributes together – Region growing/split-and-merge • Based on discontinuity of attributes to detect boundaries/contours of regions – Active contours 5/29/2016 Visual Perception Modeling 3 Introduction – cont. • Problems with edge detection – Edge detectors do not give rise to meaningful contours of objects due to their intrinsically local nature 5/29/2016 Visual Perception Modeling 4 Active Contours 5/29/2016 Visual Perception Modeling 5 SNAKE • A snake is an active contour defined by ( s ) ( x( s ), y ( s )) – Was introduced first by Kass, Witkin, and Terzopoulos – An energy is associate with each contour 1 Esnake Eint ((s)) Eimage((s)) Econ ((s))ds 0 5/29/2016 Visual Perception Modeling 6 SNAKE – cont. • The internal energy term provides a smoothness constraint on contours d ( s ) d( s ) Eint ( s ) ( s ) 2 ds ds 2 2 2 – What kind of contours is preferred 5/29/2016 Visual Perception Modeling 7 SNAKE – cont. • The image energy term provides an image related measure Eimage | I I | – Should have a large gradient along the contour, i.e., the contour should consist of edge points 5/29/2016 Visual Perception Modeling 8 SNAKE – cont. • The constraint term can be used to impose additional constraints – For example, some control points are available and they should be very close to the contour • Called spring – Some information may indicate the contours should be as far as possible from some points • Called volcano 5/29/2016 Visual Perception Modeling 9 SNAKE – cont. • Minimization of the energy – This is a standard variational problem • In order to apply calculus of variations, one needs to use a smooth representation of contours – Minimization by steepest descent • Requires the functional derivatives of the energy 5/29/2016 Visual Perception Modeling 10 SNAKE – cont. 5/29/2016 Visual Perception Modeling 11 SNAKE – cont. • Problems with the original SNAKE model – A good initial result must be available • If the initial conditional is too far from the correct solution, the snake might be trapped to a meaningless local minimum • Balloon model – Introduce an additional term which pushes the contour out or in along its normal 5/29/2016 Visual Perception Modeling 12 Deformable Templates • Active contours are closely related to deformable templates – Objects are often represented by their contours – Object recognition is then to deform the standard contour such that the energy is minimum 5/29/2016 Visual Perception Modeling 13 Deformable Templates – cont. 5/29/2016 Visual Perception Modeling 14 Deformable Templates – cont. 5/29/2016 Visual Perception Modeling 15 Deformable Templates – cont. 5/29/2016 Visual Perception Modeling 16 Deformable Templates – cont. 5/29/2016 Visual Perception Modeling 17 Tracking • Active contours can be used very effectively for tracking 5/29/2016 Visual Perception Modeling 18 Tracking – cont. 5/29/2016 Visual Perception Modeling 19 Lip-Reading 5/29/2016 Visual Perception Modeling 20 Actor-Driven Facial Animation 5/29/2016 Visual Perception Modeling 21 Human-computer Interaction 5/29/2016 Visual Perception Modeling 22 Traffic Monitoring 5/29/2016 Visual Perception Modeling 23 Medical Image Analysis 5/29/2016 Visual Perception Modeling 24 Medical Image Analysis 5/29/2016 Visual Perception Modeling 25