Symbol recognition using directional and spatial features

This paper is interested in shape representation and recognition with a particular target to technical and line-drawing symbols. Specifically, two sorts of directional and spatial features are explored to construct a new descriptor for symbol matching and recognition. These features are rotation-, translation- and scale-invariant and can be extracted with a low cost of computation. The descriptor is constructed by vertical and horizontal binning of these features. The proposed approach works well for both types of object representation (i.e., contour and skeleton). Experimental results show the robustness of the proposed method on various datasets (e.g., technical symbols and logos) compared to other baseline systems in the literature.