Image segmentation by tree pruning

The image foresting transform (IFT) has been proposed for the design of image operators based on connectivity. The IFT reduces image processing problems into a minimum-cost path forest problem in a graph derived from the image. It has been successfully used for image filtering, segmentation, and analysis. In this work, we propose a novel image operator which solves segmentation by pruning trees of the forest. First, an IFT is applied to create an optimum-path forest whose roots are pixels selected inside a desired object. In this forest, the background consists of a few subtrees rooted at pixels on the object's boundary. These boundary pixels are identified and their subtrees are eliminated, such that the remaining forest defines the object. The tree pruning is an effective alternative to situations where image segmentation methods based on competing seeds fail. We present an interactive implementation of the tree-pruning technique, show several examples and discuss some experiments toward fully automatic segmentation.