Stability Of Moment Invariants And Their Applications In 3-D Object Recognition

In view of the imaging process and object recognition practical applications, an investigation was made systematically on various factors that may influence the stability of regular moment invariants. It includes: (1) Image blurring caused by the character of the optical system. (2) Image blurring caused by the variation of the distance between objects and imaging sensor while the focus of the optical system are unchangeable and the response elements number of the imaging sensor focal plane array is limited. (3) Whether the moment invariants of the object on a finite observing area will be different from that evaluated on infinite observing area, when the image is blurred. (4) Whether the moment invariants of the object image evaluated with or without background clutter will be different when the image is blurred. Our research has proved that blurring effect in some practical situation will significantly influence the stability of the regular moment invariants. In order to apply moment invariants to the object recognition correctly and effectively, novel multi-scale feature models for objects have been proposed, featuring quasi-invariance in a certain scale range and variability beyond this one.The variation of the gray-level of the image caused by illumination will affect the stability of the moment invariants in practical situations. We discussed and improved the method of normalization of moment invariants to increase the stability to illumination in this paper.In this paper a method based on affine moment invariants was presented for the identification of 3-dimensional polyhedron objects. By combining the method of affine moment invariants and the method of constructing the group figure of the polyhedron, the 3-dimensional objects can be recognized from a single observed image. Experiments of recognition of airports and buildings displayed the validity and stability of the method.