Options
2012
Conference Paper
Titel
Automatic segmentation of wood logs by combining detection and segmentation
Abstract
The segmentation of cut surfaces from a stack of wood logs is a challenging task and leads to many problems. Wood logs theoretically have a certain shape and color, which is the main reason to apply object detection methods. But in real world images there are many disturbing factors, such as defects, dirt or non-elliptical logs. In this paper we mainly address the problem of wood and wood log segmentation by combining object detection with a graph-cut segmentation. We introduce an iterative segmentation procedure, which detects the stack of wood, segments foreground and background, and separates the logs. Our novel approach works fully automatically and has no restrictions on the image acquisition other than well visible log cut surfaces. All three steps of our approach are novel and could be applied on similar problems. We implemented and evaluated different methods and show that of these approaches, our methods leads to the best results.