Image matting is the process of estimating the foreground and background elements from a single image with limited user input. To solve this severely under-constrained problem, there exist various methods to construct color models for an image. Most previous color models can fail to estimate accurate mattes for complex images of nonlinear color distributions due to their simple color models. In this paper, we present a new dynamic curve color model for image matting that can handle nonlinear color distributions. We show that the colors of a local region can be fit to a curve when the local region includes three types of colors - foreground, background, and unknown mixed colors. Based on these colors in the local region, we adaptively construct a curve color model using a quadratic Bézier curve model. Our curve model allows the derivation of a new closed-form matting equation for estimating alpha values of colors forming a curve. We show that our method estimates alpha mattes more accurately than recent existing methods through visual and quantitative comparisons.
Bibliographical noteFunding Information:
This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (MEST) (NRF-2008-313-D00998).
All Science Journal Classification (ASJC) codes
- Signal Processing
- Computer Vision and Pattern Recognition
- Artificial Intelligence