 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
| Future |
| One main point of development and enhancement of the current handwriting identification software and which has been a major point of difficulty in my work is the ability to successfully and accurately analyze natural freehand handwriting as apposed to the forced block writing found in the current implementation. An attempt of this can be seen in the first revision of the code (1.0.cs). The biggest difficulties I found in implementing a natural handwriting analyzer is in identifying which strokes or partial strokes belong to each letter. In the current implementation this is accomplished using a grid system where the strokes of each letter are in their own corresponding grid, but in natural handwriting this is very hard to distinguish. If we instead implement a system that employs a different basic data block other than characters, this could solve the current problem but also result in further problems. If we instead analyze on a stroke by stroke basis it could result in a decrease in accuracy as it could be too indistinct. However, if we instead analyze on a per word basis which is easier to distinguish as in freehand handwriting, the strokes of an entire word is often connected or in close vicinity, but this could result in an extremely large training set to capture the distinctions of each user. This is however a very interesting and complex problem. If you instead continue to use the current letter by letter analysis, there could be improvements, additions, and removals of identifiers to increase accuracy. Between the 4 versions of my analyzer, I have employed several identifiers, but such identifiers I see that may increase accuracy include the degree of curvature of strokes, better analysis of breezier curves, and analysis of the intersection between strokes in a letter. Future research can also use the handwriting data collected from writing numbers which was not addressed in this study. |
Zuye Zheng | Ananda Gunawardena |