Generation of Slides from Hand-Drawn Sketches

As engineers, we often find ourselves spending a non-trivial amount of time converting sketched diagrams and figures into digital format. A lot of research has been done in conversion of hand-drawn sketches to computer drawings. However, most of it is domain-specific with training templates which must first demonstrate the shapes the user first intends to recognize as suggested by Kara and Stahovich [1] as well as Puzicha et al. [2]. Other algorithms are optimized for specific domains (e.g. for military drawings [3]). Due to the variability in drawing trends and variation in interpretation, a robust domainindependent algorithm is lacking. In this paper we describe a domain-independent algorithm for basic shape detection. The shape detection algorithm works by detecting corners using the Harris corner detector. The corners are then removed which allows for line detection. The lines are then combined into shapes. Since the images are captured using a hand-held device, perspective distortion is taken into account by applying homography. Slightly slanted lines are then converted into perfect horizontal/vertical lines due to unavoidable imperfections in human hand-drawn shapes.