Progress in the Automated Recognition of Sheep Pelt Branding

Small coded hole patterns can be branded, for identification purposes, on to sheep pelts. This paper describes our approach to the problem of automatically finding the pattern in a given back-lit wholepelt image and then recognising the orientation and other major components of that pattern despite noise and pattern warping. The pattern-finding stage principally uses binary morphology combined with simple cluster analysis whilst the remaining stages notably use cost function methodology. In the latter, as applied here, all the low level constraints of the pattern are listed and corresponding 2D geometric equations and inequalities formulated. Each point or group of points is then scored against these criteria with the leading contenders used for further scoring against higher level constraints. Results are encouraging.