A method aimed at the formation of intermediate and high level primitives in angiographies, i.e. vessel segments and branches, is reported. The pairing of vessel contours and centrelines is performed using attributed string matching techniques. After a brief review of string-matching technique, some key elements involved in these algorithms for imaging applications: chain encoding, cost function, normalized edit distance are proposed and analyzed. Their performances are compared on simulated data by means of subjective and objective criteria and further evaluated on coronarographic images.