Technical Report Number
This paper describes a novel method of character recognition targeted for extracting complex annotations found in engineering documents. The results of this work will make it possible to capture the information contained in documents used to support facilities management and manufacturing. The recognition problem is made difficult in part because characters and text may be expressed in arbitrary fonts and orientations.
Our approach includes a novel incremental strategy based on the multi-scale representation of wavelet decompositions. Our approach is motivated by biological mechanisms of the human visual systems. Using wavelets as a set of basis functions, we may decompose an image into multiresolution hierarchy of localized information at different spatial frequencies. Wavelet bases are more attractive than traditional hierarchical bases because they are orthonormal, linear, continuous, and continuously invertible. The multi-scale representation of wavelet transforms provides a mathematically coherent basis for multi-grid techniques. In contrast to previous ad-hoc approaches, our method promises a practical solution embedded in a unified mathematical theory.
A feasibility study is described in which several hundred characters extracted from engineering drawings were recognized without error by a neural network trained using multi-scale representations from a class of 36 distinct alphanumeric patterns. We observed a 16-fold reduction in the amount of information needed to represent each character for recognition. These results suggest that high reliability is possible at a reduced cost of representation.
Laine, Andrew Francis; Ball, William; and Kumar, Arun, "A Multi-Scale Approach for Recognizing Complex Annotations in Engineering Documents." Report Number: WUCS-90-23 (1990). All Computer Science and Engineering Research.