CK 1 Intelligent Surface Modeler By Yu Wing

CK 1 Intelligent Surface Modeler By Yu Wing TAI Kam Lun TANG Advised by Prof. Chi Keung TANG 2002 -2003 FYP Presentataion

Overview of presentation Motivation • theory and applications Tensor Voting Algorithm • tensor and voting Implementation • data representation and communication • bunny Results • dragon • etc Conclusion Presented by: Yu Wing TAI 2002 -2003 FYP Presentataion Kam Lun TANG

Motivation Theoretical interest – Emulate human visual perception Applications – 3 D modeling Presented by: Yu Wing TAI 2002 -2003 FYP Presentataion Kam Lun TANG

Overview of presentation Motivation Tensor Voting Algorithm Implementation Results Conclusion Presented by: Yu Wing TAI 2002 -2003 FYP Presentataion Kam Lun TANG

What is Tensor Voting? Representation Constraint propagation Data communication TENSOR VOTING FIELDS VOTING ALGORITHM Presented by: Yu Wing TAI 2002 -2003 FYP Presentataion Kam Lun TANG

Tensor = Ellipse SMOOTH CURVE + = POINT JUNCTION ELLIPSE (TENSOR) Presented by: Yu Wing TAI 2002 -2003 FYP Presentataion Kam Lun TANG

Tensor = Ellipse - 2 D tensor ball tensor stick tensor l Ball Tensor - 100% uncertainty in all directions l Stick Tensor - 100% certainty in normal directions Presented by: Yu Wing TAI 2002 -2003 FYP Presentataion Kam Lun TANG

2 D Stick Voting Field Encode smoothness ? Presented by: Yu Wing TAI 2002 -2003 FYP Presentataion Kam Lun TANG

2 D Ball Voting Field – Derived from 2 D stick voting field Rotation and integration Presented by: Yu Wing TAI 2002 -2003 FYP Presentataion Kam Lun TANG

Voting Algorithm Each input site propagates its information in a neighborhood voting = summation of tensor votes accumulated in a neighborhood + = + = Presented by: Yu Wing TAI 2002 -2003 FYP Presentataion Kam Lun TANG

3 D Tensor Voting points curvels surfels balls plates sticks Encode Sparse Tensor Voting ball voting field plate voting field stick voting field tensor tokens Dense Tensor Voting dense tensor map Decompose surface saliency map Feature Extraction Presented by: surface 2002 -2003 FYP Presentataion Yu Wing TAI Kam Lun TANG

Results Noisy data Sparse data Large scale reconstruction Efficient neighborhood searching in 3 D space Code Optimization Qualitative and quantitative analysis Presented by: Yu Wing TAI 2002 -2003 FYP Presentataion Kam Lun TANG

Result: Robustness 1000% 1500% 2000% 3000% 5000% 100% noise 300% 500% noise 2002 -2003 FYP Presentataion

Large scale reconstruction 1, 153, 856 triangles 35974 points Presented by: Yu Wing TAI 2002 -2003 FYP Presentataion Kam Lun TANG

Conclusion Intelligent surface modeler – 3 D surface description Tensor voting Results – Robustness – Large scale reconstruction Future work – Multiscale feature segmentation and extraction Presented by: Yu Wing TAI 2002 -2003 FYP Presentataion Kam Lun TANG

Thank you Q&A 2002 -2003 FYP Presentataion