Font Size: a A A

Handwriting Identification Based On Improved Multi-channel Gabor Wavelet

Posted on:2003-09-21Degree:MasterType:Thesis
Country:ChinaCandidate:C ShenFull Text:PDF
GTID:2168360062986206Subject:Pattern Recognition and Intelligent Systems
Abstract/Summary:PDF Full Text Request
This thesis is based on the Writer Identification (WI) issue. The improved multi-channel Gabor wavelet technique is solved in the handwriting identification issue. Based on this method, the system is also set up to fulfill writer identification. The main contents of this paper include: get the handwriting images, preprocess the handwriting images, extract the textural characters from the handwriting images, classify the handwriting images, construct and realize the writer identification system based on the improved multi-channel Gabor wavelet technique. The achievements of this thesis are as follows.(1) The advanced salt and pepper filter is used in the thesis. The original method just can be used in removing the noise from binary images. The gray images contain much information about writer habit. So the advanced salt and pepper filter is applied in the gray handwriting images. The best result is achieved in these handwriting images, which have the broad and pockety noise.(2) The shading of paper has a lot of grids or lines. These are not writers' handwriting styles and even influence the veracity of texture extraction. Therefore, the important preprocessing step is to separate the background from handwriting images. The experiments prove this method can remove the background effectively such as the grids of French folio paper and the lines of letter paper.(3) Due to the handwriting image may contain lines of different height and the different spacing between words and lines. These factors will influence the veracity of texture extraction. So I advance a method to normalize the handwriting image, which combine the Horizonal Projection Profile method, Vertical Projection Profile method and padding method. This method completes the normalized processing once and guarantee the consistence of contain sequence and the concordance of paragraph.(4) The paper advances Gabor wavelet technique and proposes the textural arithmetic to adapt to hanwriting. This arithmetic is a text independent method, so we need not segmente the handwriting text. The arithmetic records the mean and standard deviation of each channel. These are the important information of textural characters. We choose 80 vectors of 40 channels and save them in handwriting database.(5) We use a weighted Euclidean Distance (WED) classifier and k-nearest neighbor (k-NN) classifier to fulfill the identification task. The eigenvectors are recorded in different handwriting database with text format.The system is programmed with Microsoft Visual C++ 6.0 under Windows 98 and adopts multithreading mode. Experiments are made using Chinese handwriting from SO different people and very promising results (97.6%) were achieved. Otherwise, the thesis addresses the concept of multi-biometric personal identification system, hi addition, the promising results (99%) were also achieved in print font recognition. So the off-line Writer Identification system has definite practicability.
Keywords/Search Tags:writer identification, text independent, texture analysis, multi-channel Gabor wavelet
PDF Full Text Request
Related items