Quadrilateral Detection Using Genetic Algorithms

An approach based on the use of genetic algorithms to detect quadrilateral shapes in images is presented in this paper. The proposed approach finds the best sets of four edge points that are the vertices of quadrilateral shapes in the image. The proposed method uses the evidence provided by the image resulting of the application of an edge detection operator to the input image. Individuals having the best fitness scores are those that are supported by the edge evidence as being the vertices of a quadrilateral present in the input image. We use a sharing operator to avoid detecting similar quadrilaterals. This procedure is used to detect multiple quadrilaterals in a single run of our algorithm. Our method can handle perspective distortion and Gaussian noise corruption on the quadrilaterals to be detected. We have fulfilled tests to validate our approach on synthetic, noise-corrupted and real world images. Tests are both quantitative and qualitative. The proposed approach has shown also to be fast for real-time quadrilateral detection.