Detecting Difficult Japanese Characters using Eye-tracking from Multi-row Long Text

Abstract

This study aimed to develop a method for detecting difficult-to-read characters by tracking the gazes of personal computer (PC) users. A method for detecting English words that are difficult to read was proposed previously. However, this method required a dedicated user interface because the accuracy of the eye tracker is low (eye-tracking errors). In actual PC work, the range of use of a dedicated user interface is limited, and efficient work may be hindered. In this study, we propose a character detection method with eye-tracking error correction using projective transformation. The usefulness of the proposed method was examined based on the detection accuracy. The results showed that the proposed method was effective in detecting difficult-to-read characters in F-measure > 0.58 in the first row of the text. Moreover, the usefulness of the proposed method was demonstrated for nine rows or shorter texts compared with no-correction eye-tracking data.