- Training: There are three different classes of object shapes: triangular shapes (../images/img1.jpg), quadrangular shapes (../images/img2.jpg) and pentagon (../images/img2.jpg) shapes. Each of the three shape classes has 10 samples.
- Testing: (../images/img4.jpg) 15 objects from all of the above mentioned three classes
- Thresholding and pre-processing: generate binary images
- Connected component labeling and fragment removal
- Boundary tracing
- Computing Fourier Descriptors
- Alternative: 6 different general shape features: compactness, rectangularity, circularity, eccentricity, area/perimiter, etc. (General shape features are not robust a lot but it is not bad for classifying simple shapes such as triangular, pentagons, etc.)
- Nearest-mean classification