Numeral & Character Recognition And Its Applicaiton

With the rapid development of image processing and pattern recognition, more and more attention has been paid on numeral & character recognition. Numeral & character of small language recognition have their characteristics: variety of classes; complex and variety of characters; research on late start; application of rising demand. To satisfy the demand of practical applications, the deeper research of numeral & single character recognition has academic and applicant value.Character recognition consists of image preprocessing, feature extraction and classifier. In the paper, the contributions of our work mainly focus on the following aspects:For image pre-processing, a shadow removal approach is based on the improved Bernsen algorithm combined with Gaussian filter. Shadow in the illumination image can be removed effectively.The feature extraction contributions of our work mainly focus on the following aspects:1. Feature extraction based on Kirsch mask. Kirsch mask can extract horizontal, left diagonal, vertical and right diagonal direction features, then these four feature maps and the original map are projected to a new feature vector by PCA.2. A feature extraction based on PCA reconstruction: All reconstruction images can be obtained by all reconstructed PCA models processed by a test image. Image principal component analysis is based on two dimentional principal component analysis. Image principal component analysis reduces the coefficients and achieves good performance.3. Improved linear discriminant analysis: Linear discriminant analysis can not promise the optimal subspace and it probably produced the overlap between the neighbor classes. To solve the problem, the proposed improved linear discriminant analysis can find the optimal subspace.The classifier contributions of our work mainly focus on the following aspects:1. A kernel combined with the threshold Bayesian discriminant classifier for numeral & character classification. The relation of kernel and Gaussian process is analized and the problem of matrix invertible is solved by threshold substitution. The algorithm foucuses on numeral and character recognition. 2. Support vector machine based on normalized polynomial kernel: the parameters in the kernel are hard to decided and influence on the classification effects. To obtain good performance, the proposed normalized polynomial kernel can easier to decide the parameter.In this research, Bangla handwritten numeral recognition, Japanese character recognition, Japanese address character string recognition, English character and Arab numeral recognition are the research objects. Bangla handwritten numeral recognition and Japanese license plate recognition are the applications of numeral & character recognition. The good practical performances of these recognition systems are achieved.