In this demo, the geometrical 'shear' transform on XY-coordinates representing handwritten words is shown. Since slant is a free parameter to the writer (within a given range) it must be normalized before extracting the features from sampled handwriting. Here, you may see the effects of shearing on the handwritten trace, by interactively adjusting the slant angle. Within an automatic recognition system, the slant of the input sample must be estimated automatically. Subsequently, the shear transform is applied to obtain a normalized slant.
After this transform, the extracted features will be more stable, provided that the original slant estimate is reliable (Teulings et al., 1990). In short words or isolated letters, it is often difficult to obtain reliable estimates due to the small number of usable points. From a human reading experiment, it was determined that the angle of the trace at a point somewhere in the middle of down strokes coincides with the slant which human readers would indicate (Maarse & Thomassen (1983). Produced and perceived writing slant: difference between up and down strokes, Acta Psychologica, 54, 131-147.) A good choice would be to take the point of maximal velocity (minimum curvature) in down strokes, as is done in our handwriting recognizer. The histogram of slant angles per word is analyzed and outliers are removed to get a good estimate.
Please refer to our Publications when using anything from the shown material.
Notes:
to the Java-based Handwriting Demos page
to NICI's on-line handwriting recognizer description
Copyright Lambert Schomaker (April 1, 1996)
since 1/Mar/1996