Reading #22. Plushie: An Interactive Design system for Plush Toys (Mori, Igarashi)

COMMENTS:

SUMMARY:

Plushie is an interactive system that allows novice users to design plush toys by creating 2D patterns. Plushie has a pen-based interface that allows users to create a model of a 3D plush toy by sketching a silhouette. Users are also able to edit the model (i.e. cut and add parts to the model) using the pen.

DISCUSSION:

Like Teddy, Plushie is another excellent example of how pen-based interaction can facilitate a creative process. Here sketching is used to create a 2D pattern through which the system is able to create 3D model. The 3D model is a rapid prototype that can later be converted to an actual plush toy. Plushie is not only an innovative system, but it sounds like it would also be a lot of fun to use.

Reading #21. Teddy: A Sketching Interface for 3D Freeform Design (Igarashi)

COMMENTS:

SUMMARY:

Teddy is a sketching tool for designing 3D plush toys from 2D sketches.  Teddy is an example of a 3D modeling system with a pen-based interface.  The Teddy system allows the user to draw a 2D sketch and creates a 3D representation of the strokes.  The tool was meant for rapid prototyping and as such provides continuous design feedback through the creation of projected images.

DISCUSSION:

The Teddy application is a good example of how a pen-based interface can be used not only as a method for inputting data, but also as a means of being able to quickly create artistic models with a moderate amount of detail.  With the real-time rendering, I could definitely see an application for graphic development where sketched input could immediately be converted to animated models with fairly little detail.

Reading #20. MathPad: A System for the Creation and Exploration of Mathematical

COMMENTS:

SUMMARY:

Mathpad is a pen-based application created to draw mathematical equations and diagrams, and to help with problem-solving.  Mapthpad uses gesture recognition and parsing in order to recognize mathematical expressions and perform editing operations. 

The authors chose to use 3 recognition techniques to create their recognizer.  The first technique involves normalizing and filtering strokes.  The second technique seeks to reduce the number of possible recognized symbols using a combination of the calculated degree of difference between characters and the feature set classification.  The last technique is a fine classification o distinguish similar symbols.

DISCUSSION:

I think the MathPad application is a very useful tool that applies some of the functionalities of Matlab, only with a sketch-based interface.   The stroke grouping are done manually, wherein the user selects the set of strokes that make up an expression or a diagram.  I think it would be beneficial if MathPad could perform its own grouping as I can imagine that the bounding-box grouping feature provided in Mathpad could get a little tedious.  One feature I really like is that Mathpad allows users to make associations between expressions and diagrams in order to reflect changes from one associated element to the other.

Reading #18. Spatial Recognition and Grouping of Text and Graphics (Shilman)

COMMENTS:

SUMMARY:

The text vs graphics recognition method presented in this paper provides a possible classification to a grouping of strokes.  The system uses a neighborhood graph to show strokes that are in close proximity to each other and groups strokes whose vertices are connected in the neighborhood graph.  The authors use search-based optimization to determine the best grouping and labeling match.  The authors achieved a 97% percent accuracy rate for the for both grouping and recognition.

DISCUSSION:

The method in this paper has a similar goal as the Bishop paper.  However, instead of using the features of the strokes and spaces between them, this approach uses a neighborhood graph to determine the stroke groupings.  I thought it was interesting that the recognizer's thresholds are learned from a set of examples instead of from a list of predefined features.  This prevents the recognizer from the being constrained to recognize a specific set of symbols.

Did anyone else find a weird sentence in the 4th paragraph of the Previous Work section?

Reading #17. Distinguishing Text from Graphics in On-line Handwritten Ink(Bishop)

COMMENTS:

SUMMARY:

This paper presents a method for distinguishing strokes of digital ink as either text or graphics.  In the approach, the authors consider both the features of the strokes as well as the context.  Instead of trying to make hard classifications, the recognizer provides probabilities that the strokes belong to one group or the other.  This improves performance and provides information that can be passed on to more precise recognition processes.

The authors use 3 approaches:

·                     the features of each stroke are extracted and classified

·                     temporal information is used to find correlation between class labels

·                     information is extracted from the gaps between successive strokes

Testing showed that using the temporal context improved performance compared to the classification of individual strokes; the effects of using gap information was not clear.

DISCUSSION:

This is basically a paper on how to quickly distinguish text from graphic shapes.  The recognizer uses the features of strokes which is quite common, but the novelty of this approach is that it also uses the features of the gaps between the strokes.  This method is based on the assumption that strokes that belong to text and strokes that belong to graphics tend to follow other strokes of the same kind. 

Reading #16. An Efficient Graph-Based Symbol Recognizer(Lee)

COMMENTS:


SUMMARY:

The recognizer presented here is based on recognition of symbols in the form of attributed relational graphs representing the symbols' geometry and topology.  The nodes of the graphs (geometric primitives) and the edges (geometric relationships) are compared to determine if there's a match.  This approach is tolerant to large variations in size and rotation.

The authors compare four different graph matching techniques: stochastic matching, error-driven matching, greedy matching, and sort matching.  The first 3 methods achieved a recognition rate between 93.7% and 92.3%.  The Sort method had an accuracy rate of 78.5%, however it was significantly faster than any of the other methods and provided the correct match in the top 3 results 93% of the time.

DISCUSSION:

This paper gives a good overview of how symbols are represented and using attributed relational graphs.  It also gives a brief but clear explanation of the 4 graph matching techniques that were compared.  My only complaint is that I would have liked to see some of the ARGs of the symbols used in the user study.  I also wondered, since the Sort method assumed consistent orientation, how the results would have improved if that were in fact true.

Reading #15. An image-based, trainable symbol recognizer for hand-drawn sketches (Kara)

COMMENTS:

Paco

SUMMARY:

This paper discusses an image-based symbol recognizer that represents symbols as bitmap images.  The input symbols are cropped and turned into templates.  The template is then turned into polar coordinate representation for optimal orientation and compared to the definitions.  If there is a match then recognition switches to screen coordinates and definitions are analyzed using four classifiers based on template matching techniques.

The approach presented here does not rely on segmentation of strokes and therefore does not have difficulty recognizing symbols drawn with missing strokes, multiple strokes, and/or overtracing.  The accuracy for this recognizer was 95.7% in a user-dependent setting on a set of 20 symbols; the accuracy rate was 94.7% in a user-dependent setting.


DISCUSSION:

This paper presents a really interesting approach to the orientation challenge usually encountered when trying to do template matching.  Instead of rotating the template in small increments and doing a comparison to check for a match each time, the authors use the templates' polar coordinates to determine the translation between the patterns.  As the authors mentioned, I've seen this method used on rigid, well-defined objects, but not for sketched symbols.  This may have been my favorite part of the paper because it was explained very well...good reading choice.