Descriptor for shapes and letters (feature extraction)

The descriptor represents a binary or gray-scale image by coordinates of hierarchical subdivisions.

The descriptor gets an image and computes the x-coordinate of the center of mass. Then, the descriptor subdivides the image into two images by the x-coordinate and calls itself (recursively) on the transpose of each of the two sub-images. The coordinate values are calculated relative to the complete image and returned. In order to balance the representation the descriptor is also computed on the transposed image and the two resulting vectors are concatenated.

This implementation used in the master’s thesis of Shahar Armon: Handwriting recognition and fast retrieval for Hebrew historical manuscripts. The Hebrew University of Jerusalem, December 2011
References:
[] A. Sexton, A. Todman, and K. Woodward. Font recognition using shape-based quad-tree and kd-tree decomposition. In Proceedings Of The Joint Conference On Information Sciences, volume 5, pages 212–215, 2000.
[] Jaehwa Park, V. Govindaraju, and S. N. Srihari. OCR in a hierarchical feature space. Pattern Analysis and Machine Intelligence, 22(4):400–407, April 2000.
[] Alan P. Sexton and Volker Sorge. Database-driven mathematical character recognition. Graphics Recognition. Ten Years Review and Future Perspectives, pages 218–230, August 2006.
[] Georgios Vamvakas, Basilis Gatos, and Stavros J. Perantonis. Handwritten character recognition through two-stage foreground sub-sampling. Pattern Recognition, 43(8):2807–2816, August 2010.