Separation of Touching Convex Objects Using Proximity of Corner Points

Separation of touching objects is a generic problem frequently encountered in various machine vision inspection applications. A popular approach for separation of touching convex objects involves determination of boundary corner points, pairing of corner points using suitable criteria and interconnection of the paired corner points. The present paper describes a new criterion for pairing of corner points based on checking the proximity of corner points within the cluster of touching objects with the help of medial axis of the cluster. The location of each corner point is indicated by two location points, namely closest and opposite points, on the medial axis. The closest point is determined as a medial axis pixel closest to the corner point and the opposite point is determined as a closest median axis pixel lying along the corner direction. The proximity of corner points is judged by distances between their closest and opposite points. The proximity distance between two corner points is determined as the minimum among the distances between their closest points and between their opposite points. The pairs of corner points are sequenced in ascending order of proximity distance and paired using supplementary pairing criteria. The proposed proximity criterion gives better separation performance than conventional minimum distance criterion. The proposed criterion is suitable for separation of touching long convex objects. The performance of proposed approach is demonstrated on images of touching rice grains.