Identification of Distortions, Erroneous Inclusions, and Omittance in Shape Sketches

Sketches are common in fields such as mechanical engineering and electrical engineering for drafting early designs. The ability for a computer to interpret a sketch in a domain of interest is the basis for sketch-based user interfaces. In order to analyze a sketch composition, the sketch first needs to be segmented into groups of strokes which are subsequently classified as a specific shape or component. Incorrectly grouped strokes can lead to errors during classification.;We present a novel approach to identify erroneous inclusions and omittances of strokes that can occur from the segmentation step. Once the strokes are identified as missing or extra, a refined grouping can be constructed. Our approach utilizes a novel shape representation in order to facilitate identifying erroneous inclusions and omittances while ignoring minor distortions. We represent each shape by a list of vectors called Length Cross Section (LCS) vectors, which measure relative lengths of the cross sections of a shape. Erroneous inclusions and omittances become apparent with the aligning of LCS vectors of a query shape to the LCS vectors of a template. Our approach is rotational, translational, and affine invariant.;We evaluate our method on the domain of digital logic circuits. We find that our method is best at discovering erroneous inclusions and performs moderately well at recognizing omittances. The LCS model can also be adapted to classify shapes and results are presented against the Image Deformation Model by Ouyang and Davis.