Visually-Salient Region Detection in the Pixelized Image using Context-Based Model

Abstract

To overcome the resolution limitation in the visual prosthesis, various image processing methods have been proposed. These methods are limited for close-up images of faces or objects. However, images in real environment contain not only the predefined objects such as faces and letters but also unexpected visually-salient objects which give important information on the images. In this paper, we propose the region-of-interest detection method which is appropriate for real situations using a context-based model, and demonstrate that the detected region can guide attention. The gazes are estimated while subjects are watching two kinds of pixelized images obtained by a conventional method and the proposed method, and are compared with experimentally detected conspicuous region. The results show that the context-based model detecting visually-attractive region is useful to solve the resolution limitation.