A Projective Invariant Shape Matching Tool

• Project number: 202-08-07
• Students: Shira David
• Supervisor: Klara Kedem, Nati Hagby

In this work we developed a tool for the recognition of objects in images from different views based on algebraic projective invariants. Given an image of an object, we first perform edge detection to extract the objects contour. We then extract the points of bitangency from the contours, which are projective invariant measures. Based on the bitangent points we transform the curve into a canonical frame, in which any measurement performed is invariant. Finally, we use both the Hausdorff distance and the context shapes method to find the curve that matches the objects contour from the model library. The tool depicts the input and output of each step of the process. The tool can be used as a framework for researchers for developing new object recognition algorithms.