Abstract:
The direct use of the hand as an input device is an attractive method for providing natural human–computer interaction. Computer Vision community tried to solve the hand gesture recognition problem by following the paradigm: hand localization, hand segmentation, feature extraction and next step hand classification, without any abstract representation in between the last two steps. In this work is proposed to use compositional techniques in order to recognize the hand gestures. Compositional representations split complex objects into simpler parts, which are easier to recognize and using the relationships between them, the complex object is recognized. The main advantage of the compositional techniques is their generality; these techniques are more independent of application. Using these techniques we address also to the semantic gap that exists between the low level features and high level representations.
This work is an attempt to extend the types of problems solved based on the new, compositional approach. The hand posture representation is based on compositions of parts: descriptors are grouped according to the perceptual laws of grouping obtain a set of possible candidate compositions. These groups are a sparse representation of the hand posture based on overlapping subregions.
The power of compositional techniques for hand gesture recognition is proved by the results that have been obtained.