Direct Estimation of Appearance Models for Segmentation
This work addresses a bottleneck in image segmentation for computer vision applications, but it appears incremental as it builds on existing algebraic and statistical approaches without a major paradigm shift.
The authors tackled the problem of estimating appearance models for image segmentation without needing pixel-level region assignments, by developing algebraic expressions linking local image statistics to region appearances. They introduced two algorithms—a least squares solver and a spectral method—and showed through experiments that these methods work well in practice for effective segmentation.
Image segmentation algorithms often depend on appearance models that characterize the distribution of pixel values in different image regions. We describe a new approach for estimating appearance models directly from an image, without explicit consideration of the pixels that make up each region. Our approach is based on novel algebraic expressions that relate local image statistics to the appearance of spatially coherent regions. We describe two algorithms that can use the aforementioned algebraic expressions to estimate appearance models directly from an image. The first algorithm solves a system of linear and quadratic equations using a least squares formulation. The second algorithm is a spectral method based on an eigenvector computation. We present experimental results that demonstrate the proposed methods work well in practice and lead to effective image segmentation algorithms.