Natural image matting based on surrogate model

Image matting is an important process in digital image processing, with pixel pair optimizationbased methods having distinct advantages in parallelization and handling mislabeled trimaps or spatially disconnected foregrounds. Nevertheless, such methods cannot provide high-quality alpha mattes under a limited computing time, limiting their computing time-sensitive applications. Thus, this paper presents a natural image matting method based on surrogate models to address this problem. Specifically, the surrogate models for pixel pair optimization is established to approximate a pixel pair evaluation function, and its optimal solution obtained efficiently is used as the approximate optimal solution of the pixel pair optimization problem, saving much computing time. Experimental results demonstrate that the image matting based on surrogate models provides high-quality matting mattes with a little computing time and a competitive image matting performance compared to state-of-the-art pixel pair optimization-based methods that impose an excessive computing time.& COPY; 2023 Elsevier B.V. All rights reserved.