Ensemble symbol recognition with Hough forest

We present an ensemble recognition method for graphic symbols that could be interfered by intersecting objects from the context. The symbol is first represented as a set of shape points, each of which is described by a shape context pyramid capturing the local shape characteristics of multi-scale regions surrounding the shape point. A Hough forest ensemble classifier is then employed to learn the mapping between the statistical shape feature of individual parts and the category of the whole symbol. For an unknown symbol image, the probabilistic votes on the potential symbol by each of its parts are aggregated by a generalized Hough transform to form the final recognition output for the symbol. The experimental results demonstrate the effectiveness of the proposed method, especially in handling non-segmented intersecting symbols.