Reading #7 Early Processing of Sketch

Reading #7  Early Processing of Sketch  

Comments on:
Yue Li

Summary:

This paper aims to find ways to process the freehand sketches in the early phase. This include three parts: stroke approximation, beautification and basic recognition

In the stroke approximation steps, this paper deals differently with lines and curves.

The stroke was process in order to find the vertexes of the lines   (which are also treated as corners). In order to achieve this, the curvature graph and speed graph are made. Intuitively, the maxima of curvature and minima of speed have a great possibility of being a corner. In addition, the authors use the mean of each date as threshold to filter the candidate corners.  The authors also combine the curvature and speed graph to increase the accuracy of finding the correct corners. This is done by generating the hybrid fits and selecting the fit that has error below a threshold and with the fewest vertices.

For curves, firstly, the authors use the ratio between the actual lengths of the stroke segment with the Euclidean distance between its endpoints. (Ratio significantly higher than 1 is treated as a signal of being a curve). In order to efficiently find the control points of the curve, this paper choose to recursively divided the curve in the middle and compute their control points until the error is below the threshold.

In the beautification steps, the author chooses to rotate the lines around their middle point to make the lines within the same cluster to have the same slope.

This paper also implements some basic recognition algorithm for some basic shapes like rectangle and oval.

Discussion:

This paper offers a good way to find the corners of a stroke. The authors beautify the lines based on the slopes. However, this paper doesn’t say much about how to beautify a curve. And, the basic object recognition is too simple. And I hope to know how this early process would facilitate the later works of sketch recognition.