Image segmentation and matching using the binary object forest

摘要

This paper examines a robust method of image disparity analysis based on matching features derived from a graph theory technique of image segmentation. It is aimed particularly at tracking objects which exhibit large interframe motion. The approach proposed involves using multiple thresholds in an image to obtain a set of binary ‘slices’. These are processed to find all connected elementary regions in each slice. The topological relationships between all elementary regions (‘binary objects’) can be described by a set of trees. This set of trees constitute the binary object forest (BOF) for a given image. After applying this segmentation procedure to each image in turn, matching is performed between features (nodes) in the resulting BOFs to identify areas of change and then to extract moving objects.