การออกแบบและพัฒนาโปรแกรมการแบ่งส่วนในหน้าเอกสารสำหรับการรู้จำอักษร / นิศารัฐ พรศิริประเสริฐ = Design and development of a page segmentation program for character recognition / Nisarath Pornsiriprasert
The objective of this thesis is to design and develop a program for page segmentation of document images to classify data to characters, tables, and pictures using a bottom-up method and statistical rules. The program is tested on 50 pages of document images. The program starts by reading a whole page of a document to calculate the nearest bounding box including all bounding boxes. This data is used for classification by statistical rules. From the experimental results, the program can classify data into 7,319 blocks of characters, 22 blocks of tables and 213 blocks of images. The error rates are 1.29%, 2%, 18.06% for characters, tables and pictures, respectively.
