Revision history [back]

You shouldn't use contours to teach a machine learning engine, but you should use its properties. From a contour you can directly derive properties such as centroid, area, bounding rectangle, minimum enclosing rectangle and perimeter. Furthermore, you can compute extra contour properties, such as solidity, form factor, euler number and many others. So, what you want is to store this kind o features that describe the contour in a feature vector, rather than using the contour itself