Abstract:
|
Collection of a group of images is becoming more and more common. In this talk, we propose an efficient dimension reduction approach for a group of two-dimensional images. Each image is modeled by a product of three terms, two common group-level components and a subject-level one, and an additive noise term. The components are estimated via a two-step singular value decomposition (SVD) approach. The first step SVDs are applied on each image and the second one on the aggregated data. We demonstrate the superior performance of this approach through simulations and a real data example.
|