Detector-based visual identification of known objects

Regarding input image as a function defined on sensor frame, the structural identification of a known object is considered as extraction of domains of sensor frame responsible for all object details. The low-level processing tools are formed by a set of detectors. The detector is introduced as a generalization of any feature detection destined to recognize a certain class of patterns. As an operation of intermediate-level processing, the detector filling is introduced. It cuts a domain of sensor frame in response to input parameters of the detector. Then, a new object representation (TSG-model) is introduced. It consists of three components: the one responsible for the object structure, the one for construction of a detail domain by domains of its sub-details, and the one reducing identification of a detail if other details are already identified. The image processing for structural identification of an object is reduced to a typical AI search. It executes operations of detector filling, interacting with the TSG-model of the object. The paper outlines a new technology that can be applied in many areas. The presented approach is flexible with respect to the kind of input sensor, say, usual gray-scale camera, stereo image system, range image system etc.