An image encoded with character information can be created by binarizing an input image followed by connected component labeling, and then repeating the binarization and connected component labeling on an inverted version of the input image. This results in identification of connected components. Related connected components are arranged in a family tree in which successive generations of the connected components alternate between two tree layer classifications. One of the tree layer classifications is selected based on whether certain connected components define characters. A label image is created which includes labels for the connected components except for the connected components in the selected tree layer classification.