Abstract: Focusing on the deficiencies of traditional data matrix (DM) code localization methods in speed and precision, we proposed a code localization method that structured Gaussian pyramid to extract Harris corner features. Corner density map was smoothed by radial basis function, and rough candidate regions were obtained by threshold and region growing. Next, we computed two stage minimum-area encasing rectangle to implement accurate localization. Finally, we chose the candidate region with maximum score based on three properties of DM code as the real DM region. The result shows that our method which used corner detection and regional growth is robust to the various interferences with high accuracy, such as metal texture, high reflection, scratch marks and occlusion on DM code. The localization time is 25ms, which is 30 times faster than the compared methods.

Key words: micro data matrix (DM) code; accurate localization; Harris corner; radial basis function; minimum-area encasing rectangle