Abstract
This paper presents consistent transformation processing from pixel coordinates to dimensional coordinates, which plays an important role in the final stage of the whole recognition method for mechanical drawings. The output data of the both bottom-up and top-down processing are represented in pixel coordinates, which have inevitable quantization errors. So, they should be transformed into coordinates in terms of dimensions that give consistent geometric constraints to all drawing elements. The authors have realized to solve this problem by developing a unique method called 'dimension-matching, ' making use of the structural model and automatically generating simultaneous linear equations with respect to internal dimensions for outlines. The way of implementing this processing is described and some examples are finally shown.
