Recognizing arbitrarily connected and superimposed handwritten numerals in intangible writing interfaces. (January 2017)

• Record Type: Journal Article
• Authors: Chiang, Cheng-Chin; Wang, Ren-Hong; Chen, Bo-Ruei
• Is Part Of: Pattern recognition. Volume 61(2017:Jan.)
• Page Start: 15
• Page End: 28
• Publication Date: 2017-01
• Subjects: Intangible writing interface -- Finger gesture recognition -- Dynamic time warping -- Graph path finding -- Key numeral spotting; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2016.07.018
• Languages: English
• ISSNs: 0031-3203

Abstract

Abstract: An intangible writing interface is a new input mechanism that allows users to write in the air with hands or fingers. Since all written characters are connected in one writing trajectory, each writing trajectory is referred to as a one-stroke finger gesture . Because of no visual aid to indicate the effective writing area in the air, users often write conservatively the texts in a restrictive area to avoid invalid motion sensing. Consequently, each handwritten character may not only arbitrarily connect to others, but also highly superimpose over others. Aiming at addressing the challenging issue to segment and recognize the arbitrarily connected and superimposed characters in one-stroke finger gestures, this paper proposes a solution comprising (1) a simple but effective method to spot any numeral of interest, called a key numeral, from arbitrary one-stroke finger gestures and (2) a novel method to simultaneously accomplish the segmentation and recognition of all individual numerals in one-stroke finger gestures by a graph-based path finding algorithm. The evaluated performance shows that the proposed solution achieves satisfactory recognition accuracy even though no sophisticated matching features and context-dependent information are employed. Abstract : Highlights: An intangible writing interface recognizes mid-air gesture trajectories of numerals. A key numeral spotting method locates any specific numeral within a gesture trajectory. A recognition by concatenation method recognizes connected numerals without using lexical dependency. The method achieves good performance without extracting sophisticated character features.