| Popis: |
In this paper, a novel method to accelerate off-line Chinese writer identification by combining multi text-sensitive features and combining multi character level decisions is practiced. The used text-sensitive writer identification algorithm extracts Directional histogram feature, Moment feature and Wigner feature, reduces the dimensions using PCA and LDA, and adopts the simple Euclidean classifier. However the writer identification algorithms are text-sensitive, thus there are different mutual character combinations between writers. A method for retrieving in an amount of writers who has different handwriting script content is proposed in this paper, by combining and sorting the posterior probability measures of writing identification. An experiment, which is carried out in a handwriting script database, demonstrates the effectiveness of the proposed method. |
