Analysis of texture and connected-component contours for the automatic identification of writers Marius Bulacu Lambert Schomaker AI Institute, Groningen University, The Netherlands We developed new techniques for offline writer identification that use probability dis- tribution functions (PDFs) extracted from scanned images of handwriting to characterize writer individuality. Our methods operate at two levels of analysis: the texture level and the character-shape (allograph) level. At the texture level, a generic descriptor that can be used to characterize individual handwriting style is the probability distribution of edge-angles p(φ). While this classical texture feature proves to be effective for writer identification, we obtained significant further improvements in performance by designing more complex features that use the edge orientation as a building block. These new features are bivariate edge-angle prob- ability distributions (p(φ1,φ2), p(φ1,φ3)) computed separately on the upper and lower halves of text lines and then adjoined (see fig. 1a). They encode, besides orientation, also curvature and location specific information, giving an intimate characterization of the individual handwriting style (see fig. 1b). 0.08 0.06 0.04 0.02 0 0.02 0.04 0.06 0.08 0.08 0.06 0.04 0.02 0 0.02 0.04 0.06 0.08 writer 1 - paragraph 1 writer 1 - paragraph 2 0.08 0.06 0.04 0.02 0 0.02 0.04 0.06 0.08 0.08 0.06 0.04 0.02 0 0.02 0.04 0.06 0.08 writer 2 - paragraph 1 writer 2 - paragraph 2 bottom halves bottom halves top halves top halves φ 2 φ 1 φ 3 INK BACKGROUND a) b) Figure 1: a) Extraction of the edge-based texture features on letter ”a”, b) Examples of lowercase handwriting from two different subjects and the corresponding polar diagrams of the ”split-line” edge-direction distribution p(φ). In our analysis at the allograph level, the writer is considered to be characterized by a stochastic pattern generator, producing a family of connected components. A codebook of connected-component-contours (COCOCOs or CO 3 s) is generated from an independent 0 This paper was published as: Marius Bulacu, Lambert Schomaker, Analysis of texture and connected-component contours for the automatic identification of writers, Proc. of 16th Belgium- Netherlands Conference on Artificial Intelligence (BNAIC 2004), 2004, pp. 371-372, 21-22 Octo- ber, Groningen, The Netherlands