Analysis of “Recognizing Free-form Hand-sketched Constraint Network Diagrams by Combining Geometry and Context”
Comments Made Elsewhere:
Summary:
Seek to create an hand-sketch interface to constraint programming and optimization. Constraints satisfaction has two parts: modeling (define the variables, their possible values, and possible relationships between them) and solving (systematic determine the values of the variables given the constraints). Focuses on the former by creating a graphical vocab that can be read to define the problem. Effective modeling is hard and takes expertise (hence this paper). Sketching allows for more creative and interactive feedback. Use a geometric recognizer instead of gesture so not dependent on drawing style. Break into primitives and heuristically recognize.
Allows for some letter recognition. Uses context to resolve ambiguous shapes (little ‘v’, arrowhead, or less than sign?). Discussion of how recognition of nodes, undirected/directed links between nodes, variables, and operators occurs. Each stroke is counted as two (one in each direction) and these “partners” keep tract of each so as not to be counted twice.
Threshold determination is required for some recognition (e.g. ‘x’ and ‘y’ only differ by height). Instead of using set threshold (i.e. 10 pixels), determined that a coincidental point must be closer to an endpoint than the midpoint of the line (and vic versa for intersections). Still work to be done, but empirically works well.
Discussion:
Appears to be another great application of sketch recognition to an uncharted domain. The authors seek to take up the challenge of providing a smart interface to constraint satisfaction logic. Of note is the mixture of shapes and variables (i.e. letters) together. Here they describe the single letters geometrically since traditional handwriting recognition requires the context of other letters and words. They of course also will allow for implicit input via a modal switch and a keyboard.
Context is king for the success of this paper.