Abstract. In the present paper, we introduced an arc recognition technique suitable for irregular isothetic object. It is based on the digital inter-pixel (DIP) circle model, a pixel-based representation of the Kovalevsky's circle. The adaptation to irregular image structurations allows us to apply DIP models for circle recognition in noisy digital contours. More precisely, the noise detector from Kerautret and Lachaud (2009) provides a multi-scale representation of the input contour with boxes of various size. We convert them into an irregular isothetic object and, thanks to the DIP model, reduce the recognition of arcs of circles in this object to a simple problem of point separation.