Using Learning Algorithms to Improve Corner Detection

This paper discusses some preliminary results obtained using learning algorithms to improve the detection ability of various corner detectors. Two main problems are considered. The first problem concerns the Harris detector, which is defined using a Gaussian weighting function. A genetic algorithm is described for modifying this function to optimise the corner detection performance. The second problem concerns methods for combining corner detectors. An attempt is made to find the best combination using supervised classification techniques.