Accurate Image Cutout Using Automatically Generated Trimaps

Image cutout has been widely used in various image processing applications. Image cutout techniques can be divided into semi-automatic and automatic methods. Semi-automatic methods require user interaction for precise cutout of target objects. Automatic methods adopt machine-learning approaches which do not require any user interaction. In this paper, we proposed a new cutout method which combines the two types of methods for obtaining precise cutout results without any user interaction. The proposed method, first, roughly extracts an object region using a machine-learning approach. Next, using the extracted result instead of user input for a semi-automatic method, the proposed method obtains a precise cutout result. From experiments using COCO dataset and AIM-500 dataset, we confirmed that the proposed method can precisely extract object regions without any user interaction.