Morphological approach of handwritten word skew correction MARISA E. MORITA -FL ´ AVIO BORTOLOZZI -JACQUES FACON -ROBERT SABOURIN CEFETPR–Centro Federal de Educac ¸˜ ao Tecnol´ ogica do Paran´ a Avenida Sete de setembro 3165, 80230-901 Curitiba-Pr, Brazil marisa@bsi.com.br PUCPR–Pontif´ ıcia Universidade Cat´ olica do Paran´ a Rua Imaculada Conceic ¸˜ ao 1155, Prado Velho, 80215-901 Curitiba-Pr, Brazil fborto@rla01.pucpr.br PUCPR–Pontif´ ıcia Universidade Cat´ olica do Paran´ a Rua Imaculada Conceic ¸˜ ao 1155, Prado Velho, 80215-901 Curitiba-Pr, Brazil facon@rla01.pucpr.br Laboratoire d’Imagerie, de Vision et d’Intelligence Artificielle (LIVIA) Ecole de technologie sup´ erieure, D´ epartement de g´ enie de la production automatis´ ee 1100, rue Notre Dame Ouest, Montr´ eal, Qu´ ebec, H3C IK3, Canada sabourin@exa.gpa.etsmtl.ca Abstract. The correction of handwritten word skew is an arduous task that must be independent of due to style and writing conditions variations. We propose here a morphology-based method to detect and correct handwritten word skew in the treatment of dates written on bank checks. Our aim is to limit the number of parameters and heuristic features necessary for a good skew correction. Our approach is based on the morphological pseudo-convex hull. We will illustrate the accuracy of this new method with real examples of dates handwritten on bank checks. Keywords: Mathematical morphology, convex hull, handwriting. 1 Introduction Word treatment and automatic recognition poses a diffi- cult problem. By nature, handwriting is very unsteady in shape and quality of tracing. The fact that research in this domain has been going on for about thirty years evidences that an all-encompassing solution is still to be found. Handwriting location and recognition stages depend to a large extent on its disposition and more especially on its skew. The extraction of reference lines from a word (superior line, baseline, lower-case letter body line) is paramount for efficient recognition. A lot of words present an unknown arbitrary skew that can generate mis- takes in the extraction of these lines and failure of the recognition process. Handwriting skew correction be- comes necessary. It is used both in handwritten and prin- ted contexts. [El Yacoubi (1996)] uses it for the determi- nation of lower-case letter bodies whereas [Trupin (1993)] uses it to localize the different lines of a paragraph. Some authors [Cˆ ot´ e (1997)], [Madhvanath (1996)] extract these reference lines without previous skew cor- rection. These approaches are generally applied when words don’t suffer any distortion other than rotation and translation. If some letters of a word are not aligned, that means that the skew is not constant; the application of such methods doesn’t allow accurate definition of refe- rence lines. In the particular case of the handwriting skew cor- rection, most works require the use of heuristics-based parameters, which render the generalization of such meth- ods difficult. We will show that it is possible to reduce the number of empirical parameters to allow the definition of an automatic approach that can be independent from handwriting type. Our approach leans on the concept of pseudo-convex hull extracted from morphological tools. In section 2, we will try to demonstrate that cur- rent approaches to word correction are built using heuris- tic factors that impair the generalization of the correc- tion process for any kind of handwritten word. In sec- tion 3, we will show how the morphological pseudo- convex hull can solve our problem. The necessary mor- phological tools to obtain the pseudo-convex hull will be presented. In section 4, we will demonstrate how the pseudo-convex hull approach can serve to correct the handwritten word skew. We will illustrate the degree of application with real examples of handwritten dates af- fixed on bank checks, (section 5). 2 State of art of the skew correction of handwritten words Many works in the literature describe heuristic parameter techniques. For instance, [El Yacoubi (1996)] after hav- Anais do XI SIBGRAPI (1998) 1–?