Object Identification and Recognition Using Multiple Contours Based Moment Invariants

A particular kind of invariants known as moment invariants have been used for object recognition and identification due to its invariabilities under translation, rotation and similitude transformation and approximate invariability even under small-scale perspective transformation, while contours representing shapes and structures play key role in people´s tasks of reorganization and identification. A combining approach incorporating contours into moment invariant is proposed in this paper such that advantages of both are employed. Redefined on contours of both the whole objects and their inners, not just are moment invariants related to geometric information but they are less sensitive to non-essential changes and less complex for computations. Vectors formed by invariants act as features of objects and are classified by Nearest Neighborhood. If common features of kinds of objects are defined in advance, this algorithm can be used for objects belonging to those kinds. Time and space complex are rather low and experimental results show desirable accuracy.