| Object recognition is the process of identifying objects in an image based on a database of object models that are known a priori. The state of our current technology is not yet capable of creating machines that can perform general object recognition. We know it is possible, however, because humans seem to do it with little effort.; We examine some of the complexities involved in building such a system, and propose that unifying local invariants and global models can increase the power of object recognition engines, thus allowing the system to recognize complex objects while making it tolerant to noise and occlusion, as well as insensitive to many environmental changes. In particular, we describe a framework for computing invariant features that are insensitive to rigid motion, affine transform, changes of parameterization and scene illumination, perspective transforms, and viewpoint change. The formulations are widely applicable to many popular basis representations, such as wavelets, short-time Fourier analysis, and splines. It enables a quasi-localized, hierarchical shape analysis that is rarely found in other known invariant techniques. Moreover, it does not require estimating high-order derivatives in computing invariants (unlike strict local invariants), whence is more robust.; To recognize multi-part, articulate objects, the scheme involves computing a local invariant signature for each segmented region in the image. Ambiguities between individual region matches are then resolved through relaxation labeling techniques. A final match is established when a collection of segmented regions in the image conform to an object model, both in terms of local shape description and global structural relation. The scheme thus allows for articulated movement of object parts within the scene. The procedure is easy to implement, yet shows promising results in its ability to isolate interesting regions in images and video, to account for structural and relational constraints among regions, and to integrate both local shape and global structural information for a detailed examination of the scene, in a way that is invariant to many visual variations. |
