Embedded Computer Vision Project

Key Points
• Individual junior project within image processing course
• Geometric shape detection using binary image shape features and fuzzy classifier
• Application differentiates between three shapes (circle, cone, rectangle) captured via CMOS camera

Processing and Analysis Steps
• Data Reduction, Level I
• Image Restoration
• Data Reduction, Level II
• Segmentation using spatial information
• Feature Extraction
• Feature Analysis and Pattern Classification

Image Restoration

           Camera Noise Model                                                                    Low Pass/Arithmetic Mean Filter (AMF)                                Red-Channel Images
                                                                                                                            (1/N2) * Sd(r,c) (r,c)eW                                            Before and After
                                                                                                                                                                                                              AMF Application
                                                                                                                       

                                                                                                                       Minimal Convolution Kernel

                                Segmentation
Red threshold AND 4-neighbor connectivity

Feature Extraction/Vector Generation

Binary Shape Features
• Horizontal & Vertical Projections
• Thinness Ratio
• C(x) = 4 * p * Area / Perimeter2

Principal Projections
• Reduce feature vector size
• Retain basic shape signature

A few membership functions used with the fuzzy classifier

All content © 2009 DigiPen (USA) Corporation, all rights reserved.