Fast and Robust Image Cutout Using Bilateral Grid and Confidence Based Color Model

We present a novel method to image cutout based on both bilateral grid and confidence-based color model, which fast and accurately extract the foreground object from complex natural images with less user interaction. First, we resample each pixel of an input image into a regularly sampled bilateral grid, which accomplishes an efficient mapping from image to bilateral space. The resolution of the bilateral grid is significantly lower than the input image, such that the amount of the image data to be processed is greatly reduced. Secondly, we design a graph-cut based energy on the vertices of the bilateral grid. The key is to model the confidence-based color distributions according to the predefined discrimination criteria that are reliable against ambiguous colors. Finally, we use a standard max flow/min cut algorithm to solve the global optimization problem, achieving a foreground extraction with high quality. The experimental results show that our method deals with high resolution images in 1 second, which runs at real-time frame-rates and cuts out meaningful objects, and realize the robust image cutout for complex scenes.