ABSTRACT
This work proposes a generic algorithm of shape detection in a grayscale image from its Max-tree. This structure organizes different connected components in all scales of the image in a hierarchical way offering more semantic information and less elements to process in relation to the traditional grid of pixels representation. Filtering methods and recognition of shapes, such as lines, straight lines, circles, and archs, are proposed through searching nodes of interest in the tree. The run time of this method is not necessarily increasing by the number of parameters of the geometric shape. Moreover other more complexer objects (non-parameterizable) can be detected by morphological analysis.
