A new method of character line extraction from mixed-unformatted document image for Japanese mail address recognition

Presents a new method of horizontal and vertical character line extraction in mixed (handwritten/printed) unformatted document images, in various character sizes, gaps and orientations nested among advertisement characters, drawings and photographs. We use the inherent features of a character line, such as the number and size of the characters it contains and the angular spectrum of the characters. When an area has characters along both horizontal and vertical lines, then competitive judgment is applied. Using multi-set thresholds in a bottom-up methodology, we can successfully extract Japanese mail address character lines. 957 address character lines, taken from 252 pieces of mail, were tested, and a 95.9% correct extraction rate was achieved.