ABSTRACT
I. INTRODUCTORY REMARKS Despite continuous deformation, visual information can be articulated into a set: an alphabet o f patterns, perhaps consisting of simple figures [5] or fractal patterns [15]. The articulation is the first step in image analysis for interactive design [11,17], cooperative decision [9,18], and propositional scene analysis [10,16]. Imagery consisting of simple artificial patterns can be articulated into a system of a priori alphabets via statistical and/or syntactic analysis [3,4]. For instance, 3D contours are matched for locating objects in a noisy image [1,7], and 2D lines are extracted for discriminating objects in the background [8]. However, these rigid alphabets are too susceptive to visual deformation in describing complicated natural im agery.
