Evaluation of Different User Input Types in Interactive Segmentation

Abstract

Interactive image segmentation is the process of extracting the desired object(s) with the help of the human operator. It is a looping process whereby users are required to provide input until the desired object is segmented. The placement accuracy of the user input will help to reduce the number of loops required. Normally, users are required to place strokes on the foreground and background of the desired objects. However, segmentation algorithms tend to misinterpret the intention of the users although the foreground and background strokes were placed. Therefore, user input constitutes a very important step in the success of interactive segmentation algorithm. Currently, to the knowledge of the authors, there is no research on the use of these different types of input on the accuracy of the segmentation results. Therefore, this paper intends to fill this gap. In this paper, we present a brief review on the different input types that had been used as the initial input and in the refinement process. Next, a series of experiments had been carried out on the use of these different types of user input to study the effect on these common users input types on the segmentation results. The experiments will look into the location, number and length of these different input types using Berkeley image database with the nonparametric higher-order learning. It was noticed that, the location, number and length of the different user input types will affect the segmentation results on complex images while remain consistent for simple images.