Marginal Noise Removing and Text Extraction from Overlaped Background Image
| Autor: | Lai, C.R., 賴燦然 |
|---|---|
| Rok vydání: | 1997 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 85 When scanning a thick book or when the document is skewly placed, it will arise a condition of poor and non-uniform illumination in the inflectional portion of books. The non-uniform illumination will generate noises on the marginal region of the document image.This kind of noise is called "Marginal Noise". "Marginal noise" makes segmentation and the following processes of Document Analysis are performed incorrectly. To quickly detect the position of marginal noise regions,we reduce the image into smaller one and filter out the text portion. According to the shape, length, and location of splitted blocks,we can distinguish marginal noise from non-marginal noise blocks and find out the corresponding regions, which were classified as marginal noise in the source image. After detecting the marginal noise regions, we can directly remove the marginal noise regions in the binary source image, and re-threshold the regions to preserve the characters that are covered by shadow in the gray scale image. In addition to marginal noise detection and removing, we also discuss another important document analysis topic in this thesis, that is text extraction from symbolic background. Liang et al. proposed a morphological method to extract text string from text/background image.In this thesis, we propose an improved method to quickly estimate the parameters for extracting background symbols. Then define a structure element according to the estimated width of background symbols. While compensating the gaps of characters, it will produce unwanted stripes in the characters at the same time. To resolve this problem, the width of characters is estimated to filter out the stripes. Experimenting with a wide variety of test samples reveal the feasibility and effectiveness of our proposed methods in resolving these two important document analysis topics. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|