Gladkih V. The handwritten text extraction from postal order image for technological process optimization in post

The subject of this thesis work is the developing of handwritten text extraction methods and components of optical postal order image recognition system. Thesis work favors the providing of effective enterprise content management systems and technological process optimization in post. The method of postal order image classification by color content based on the analysis of the eigenvalues of covariance matrix have been proposed at first in this thesis work. Due to completed theoretic and experimental researches on relationship between the color content of postal order image and the eigenvectors of covariance matrix obtained for the whole image the coordinate space has been proposed. As the basis this coordinate space uses this eigenvectors sorted in eigenvalues descending order. For each of three classes of postal order images the segmentation method was developed. It was shown, that using only the image projection at the chromatic basis vector defined by the color content of postal order image is enough for handwritten text extraction. The modification of the entropy method of postal order blank image binarization, that allows determining the threshold value through several iteration have been proposed. The asymmetrical structure element of morphological operations that provides good filtering of almost all noise in the images of benchmark set without visible character distortion has also been proposed at first. The expert system based on mutual analysis of text character regions has been proposed. This expert system extracts the strings and words of text written in block letters. As a result the software application has been developed. This application implements the segmentation postal order blank image specifically the handwritten text extraction.