In intelligent document processing system and geographical information systems (GIS), the image processing and recognition play an important role. This thesis deals with various problems in processing images in documents and GIS: image smoothing, filling, linearization and extraction of contour features, extraction of structural points, separation and recognition of spurious segments in handwritten digits, reconstruction and recognition of broken digits, and separation and recognition of colour document and GIS images. These approaches are also called Optical Character Recognition (OCR).
A new smoothing technique is developed to smooth follow contours of image. With the new smoothing algorithms, spurious pixels (points) of contours are removed based on smooth patterns, and smooth followed contours are found. Also, skeletons of image can be smoothed between neighboring �end� and �junction� points. Smooth following makes linearization of smoothed contours possible based on Freeman codes.
A new filling algorithm of contours, project filling, is described based on two kinds of structural patterns. By this method, any complicated contours of images can be filled correctly.
Different from other linearization methods, linearization and feature extraction of smoothed contours are based on difference chain codes. Curvature and bend angles of linearized are found. The convexity and concavity of linearized are described. In this way, a series of description features of contours is formed.
Structural points are new and useful features to describe morphological structures between neighboring linearized lines. Extraction of structural points is based on structural patterns which are determined by element chain codes. Also, extension Freeman codes are used in this thesis. Structural points make description and recognition of contours possible.
In order to recognize handwritten digits in document processing systems, separation of spurious segments, reconstruction of broken digits and recognition of handwritten digits are investigated. Experiments with large number of testing data set show satisfactory results for these algorithms.
Separation and recognition of colour document and GIS images are discussed. Object images of document and GIS images are extracted based on the description of shape structures, prior knowledge and color information, which are associated with each other. Color images can be described by a limited number of colors in color document and GIS images. Therefore, separation of color image is done by color reduction method, and recognition of object images is based on structure patterns, prior knowledge and colour information. It can be seen that specific information should be considered in many practical problems to achieve better processing results.
| Identifer | oai:union.ndltd.org:ADTP/216504 |
| Date | January 2005 |
| Creators | Yu, Donggang, dyu@venus.it.swin.edu.au |
| Publisher | Swinburne University of Technology. |
| Source Sets | Australiasian Digital Theses Program |
| Language | English |
| Detected Language | English |
| Rights | http://www.swin.edu.au/), Copyright Donggang Yu |
Page generated in 0.0014 seconds