Author: Christopher F. Herot
Summary:
HUNCH - multi level sketch interpreting system. Sketch interpreted in one level acts as the input for the next level and so on. Constant sample rate input device is used for sketching. Based on STRAIT which assumed that speed of stroke can be linked to intent with which it was drawn. Fast strokes were less intended than the slow ones. Speed of the stroke was less around the corners.Model worked better for some people. STRAIN , modification of the old system uses speed and bentness to find the corners.
Latching - STRAIT used latching to join endpoints which was lying within a radius. This simple technique produced lot of errors.Features that can be considered to determine latching - change in size/ angle, closure, number of lines, speed and line length, user profiles ( training to specific user).
Overtracing - Reducing the amount of data given by the user. line.The amount and style of overtracing may serve as a measure of a particular user's attitude toward the design, with heavy overtracing indicating emphasis or reinforcement of selected areas.
Other levels of inferences - Program uses simple rules of projective geometry to map two-dimensional networks of lines into three-dimensional data structures. this has also helped in finding the interdependencies of latching and finding the third dimension.
Program maps the sketches to 2 dimensional grids (similar to bitmap). lines passing through point is '1' and others have value '0'. Used in room finding algorithm - works for simple sketches.
Recognising sketch is done based on context. HUNCH system generates a context free data structure which follows a top down approach with general nodes on the top and specific details in the bottom. Matching is done in a top down fashion. First, match is found for the top most node then moved down.
Discussion:
This system doesnot use the domain based information instead allows users to manipulate the results and trains based on them. Some interesting mappings are given in this paper - speed of stroke to intent of user , speed/bentness - corner, overtracing - attitude of the user . It has also thrown light on the problems of latching and overtracing which i never knew before reading this paper.
Subscribe to:
Post Comments (Atom)
2 comments:
Yes this paper introduces some concepts which I didn't knew earlier. It also very successfully explains the complexities involved in the problem of latching and overtracing.
This idea of mapping aspects of the stroke to user intention and expressiveness is very interesting. I'm curious about ways this can affect recognition as well as user interaction; what things the system can do with this information.
Post a Comment