Engineering Drawings Russian Recognition Of Handwritten Character Figures

In this thesis we mainly discuss methods for handprinted character recognition, and the way to apply them to solve the problem of automatically recognizing handprinted Russian alphanumerics in engineering drawings. Main works in this thesis include:(1) After a comprehensive analysis of existing normalization methods, we decide to correct the skew distortions of character image by taking a combination of moment-based normalization method with gravity-geometry center normalization method. There are some limitations in moment-based normalization method, and after pointing out them, we propose several ways for compensation. We also propose two highly efficient implementation methods for the normalization algorithm we decide to use.(2) By combining two classifiers, one based on features extracted from character contour, the other based on features so called Characteristic Loci which we have modified, the recognition system outputs several candidates sorted by their distances to the unknown. And following that, some procedures are employed to discriminate similiar characters. In this thesis, we also make a discussion on the problem of combining several classifiers.(3) Based on recognition methods mentioned above, We design and implement a complete recognition module (in DLL form) for handprinted Russian alphanumerics, and conduct an experiment on the recognition of handprinted Chinese digits and unit characters (totally 21 classes) which are used in bank checks, tax forms etc. Details for the experiment results can be found in Chapter 5.