matrix factorizations appeared in 1974 [3]. However, until recently there was no fast and stable updating algorithm During the past few years several interesting applications of eigenspace representation of images have been proposed. These for the SVD .
include face recognition, video coding, and pose estimation.
In the context of image analysis in eigenspace, this paper However, the vision research community has largely overlooked makes the following contributions: parallel developments in signal processing and numerical linear algebra concerning efficient eigenspace updating algorithms.
โข We provide a comparison of some of the popular tech-These new developments are significant for two reasons: Adoptniques existing in the vision literature for SVD/KLT coming them will make some of the current vision algorithms more putations and point out the problems associated with robust and efficient. More important is the fact that incremental those techniques.
updating of eigenspace representations will open up new and
โข We outline a low-rank SVD update algorithm which interesting research applications in vision such as active recogis efficient and numerically stable. Using this we suggest nition and learning. The main objective of this paper is to put these in perspective and discuss a new updating scheme for a technique for adaptively modifying the number of basis low numerical rank matrices that can be shown to be numerivectors and provide an error analysis.
cally stable and fast. A comparison with a nonadaptive SVD โข We provide preliminary experimental results for the scheme shows that our algorithm achieves similar accuracy case of 3D object representation using image projections. levels for image reconstruction and recognition at a significantly Other interesting applications in vision are identified.
lower computational cost. We also illustrate applications to adaptive view selection for 3D object representation from
Although SVD updating techniques have been used by projections. ยฉ 1997 Academic Press several researchers in the past, to the best of our knowledge this is the first time that a scheme is suggested for adaptively modifying the number of basis vectors.