Smoothing Algorithms

Preprocessing is one of the most important stages in pattern recongition. The idea behind preprocessing is to eliminate the artifacts that arise in input patterns due to certain factors like image acquisition noise, digitization errors, and lighting conditions when images of patterns are captured. Such artifacts are not intrinsic properties of the input patterns and thus should not be used in the classification process. That is, in the feature extraction and pattern classification stages that follow preprocessing, we are concerned with analyzing the real input pattern and not that pattern cluttered with all kinds of noise from various sources. Preprocessing can be achieved with a wide variety of techniques depending on the particular application. Examples of popular preprocessing techniques include smoothing, sharpening, and contour tracing.

Polygonal smoothing is an example of a preprocessing technique usually used in character recognition applications. It is often required to eliminate the noise from the input characters. Such characters may be distorted due to various noise processes which include digitization noise. The following figure shows a distorted character and what we desire out of the preprocessing stage of a character recognition system.

Some methods for smoothing and approximating the boundaries of shapes represented as polygons are: Hysteresis smoothing, Iterative end-points fit, and Mid-point smoothing.