Bibliography:
Bo Yu and Shijie Cai. A Domain-Independent System for Sketch Recognition. Proceedings of the 1st international conference on Computer graphics and interactive techniques in Australasia and South East Asia. pp. 141-146. 2003.
Summary:
This paper talks about a domain independent low-level sketch recognition system, whose output is organized in a hierarchical structure, with semantic and syntactic information, along with the raw data. The system that can handle curves and hybrid shapes, as well as it handles polylines. The author aimed to solved two issues in existing systems,
1) Ability for user to sketch naturally. 2) Decompose hybrid and
smooth curves into primitive shapes.
The system records coordinates of the strokes, along with time. The process is divided into 2 stages, imprecise stroke approximation and post-process. The stroke approximation is done immediately (to either lines, curves, circles, ellipses or helixes), giving the user rapid feedback. Once the sketch is done, postprocessing is done and gives a set of hierarchical outputs, lowest level being raw-data and highest level being semantic relations table.
The imprecise approximation is done by looking at the direction and curvature graphs of the strokes. Feature-area is used as a guide during this phase. One main feature of this paper is that, it does not identify vertices in a seperate stage, and is combined with the approximation phase, allowing more information to be used and taken. If a given stroke, cannot be approximated to a primitive shape, it is recursively broken down at the highest curvature curve point and recursively checked. (This process is a little different for self-intersection strokes). The system also provides some basic object recognition, that are common accross all domains.
The system was evaluated by ten randomly chosen students, who all thought it was very easy to use. The system achieved 98% in correctly identifying primitive shapes, about 70% for hybrid shapes with smooth curves and 90% for other hybrid shapes.
Discussion:
The idea of recursively breaking down shapes and trying to create primitive approximations for them is very interesting. It's nice to see that curvature and direction data can get us a long way in primitive shape identification. One thing that was not clear to me was the rapid feedback. Won't the rapid feedback affect the flow of the user while sketching? If the system got the recognition wrong, its better to leave the original stroke as is (like LADDER), than recognizing it incorrectly.
Bo Yu and Shijie Cai. A Domain-Independent System for Sketch Recognition. Proceedings of the 1st international conference on Computer graphics and interactive techniques in Australasia and South East Asia. pp. 141-146. 2003.
Summary:
This paper talks about a domain independent low-level sketch recognition system, whose output is organized in a hierarchical structure, with semantic and syntactic information, along with the raw data. The system that can handle curves and hybrid shapes, as well as it handles polylines. The author aimed to solved two issues in existing systems,
1) Ability for user to sketch naturally. 2) Decompose hybrid and
smooth curves into primitive shapes.
The system records coordinates of the strokes, along with time. The process is divided into 2 stages, imprecise stroke approximation and post-process. The stroke approximation is done immediately (to either lines, curves, circles, ellipses or helixes), giving the user rapid feedback. Once the sketch is done, postprocessing is done and gives a set of hierarchical outputs, lowest level being raw-data and highest level being semantic relations table.
The imprecise approximation is done by looking at the direction and curvature graphs of the strokes. Feature-area is used as a guide during this phase. One main feature of this paper is that, it does not identify vertices in a seperate stage, and is combined with the approximation phase, allowing more information to be used and taken. If a given stroke, cannot be approximated to a primitive shape, it is recursively broken down at the highest curvature curve point and recursively checked. (This process is a little different for self-intersection strokes). The system also provides some basic object recognition, that are common accross all domains.
The system was evaluated by ten randomly chosen students, who all thought it was very easy to use. The system achieved 98% in correctly identifying primitive shapes, about 70% for hybrid shapes with smooth curves and 90% for other hybrid shapes.
Discussion:
The idea of recursively breaking down shapes and trying to create primitive approximations for them is very interesting. It's nice to see that curvature and direction data can get us a long way in primitive shape identification. One thing that was not clear to me was the rapid feedback. Won't the rapid feedback affect the flow of the user while sketching? If the system got the recognition wrong, its better to leave the original stroke as is (like LADDER), than recognizing it incorrectly.
No comments:
Post a Comment