Research On The Feature Extraction For Off-line Handwriting Identification

With the development of science, writer identification (WI) based on handwriting has become an important technology of biometric personal identification. This technology has been widely used in the public security, administration of justice, archaeology, finance and electronic business areas. In recent years, social backgrounds urge more achievements in computer (WI). Features extraction is an important part of handwriting recognition, and has a direct impact on the recognition results. So researching on feature extraction technology has important theoretical significance and application value.After detailed researching on the related technology of handwriting feature extraction, this thesis proposes to segment handwriting image to sub-image and extract the center of gravity features, shape features of each sub-image, then extract texture features for the whole image to realize handwriting identification, and at last to synthesize the results of each way to get the final result. In these ways the central of gravity features reflect handwriting gravity, it is the performance of writing habits and it is also a kind of text independent features. Shape features are proposed after analyzing the physical properties of geometric moments. These features are extracted from handwriting blocks, and it is also a text independent feature. Texture features reflect the texture characteristics of handwriting. These features are obtained by filtering image used Gabor filter. It is a kind of text independent features absolutely and it is used widely.Finally in this thesis, I design a simple handwriting identification system for testing effectiveness of each feature The system includes pre-processing module, the central of gravity feature extraction and classification modules, shape features extraction and classification modules, texture features extraction and classification modules, also includes an integrated module for each recognition result. Each proposed method has been tested through these modules. Results showed that: the handwriting identification based on these three kinds' features achieves good results, and integrated result of each way can get better results.