Robust interactive segmentation via coloring

Ozan Sener, Kemal Ugur, A. Aydin Alatan

Abstract: User centered and high performance interactive image segmentation is important for ground truth annotation tools as well as many mobile computer graphics and vision applications which requires user to select an object from non-trivial background. On the other hand, These methods should also tolerate interaction errors due to the small sized screens of the mobile devices. We namely propose a new interaction mechanism called coloring. Coloring is completely dynamic interaction methodology and specifically designed for touchscreen devices and mobile applications. Moreover, in order to compensate interaction errors, we have proposed a mechanism to handle errors. Proposed interaction methodology is also tested subjectively. And, superior performance of both interaction methodology and error-correction mechanism is also presented.