A computationally efficient algorithm for the detection of objects in complex scenes will be described. Spatial arrangements of local features, defined in terms of star type graphs (spiders), are used to represent object classes. Detection then involves an efficient implementation of a generalized Hough transform. The algorithm will be discussed in the context of various deformable template models (snakes, 2d image matching), in particular in its role as a simplification of the graphs underlying these models, and hence as an initialization tool for the associated algorithms. Multiple examples will be presented such as face detection and registration, symbol detection, 3d object detection, anatomy detection in medical images. Parallel implementations of the detection algorithm and connections to biological visual systems will also be discussed.

References:
Deformable Templates for Object Detection. Notes for a tutorial presented at ICIP 1998.