Signage and Text Localization
### (Under Construction)
Signage and text localization is to extract the regions containing signage symbols and text strings from images/videos.
Signage symbols and text strings provide unambiguous and concise information about their surrounding environments, so they always play a signficiant role in many computer vision applications, such as scene understanding, geo-coding, content-based image retrieval, assistive navigation, and SLAM.
In image/video examples, signage and text is always embedded into clutterred background, and there is no fixed patterns of signage symbols and text strings.
Most existing Optical Character Recognition (OCR) systems are applied only to signage/text regions, where the background outliers from the images/videos have been eliminated. Therefore, it requires effective and efficient solutions of extracting image regions containing text instances from clutterred image/video frames.
!(example_textdetections.png "Figure 1. Example results of text localization in the ICDAR 2003 Robust Reading Dataset, where the text regions are marked by cyan boxes.")
Figure 1. Example results of text localization in the ICDAR 2003 Robust Reading Dataset, where the text regions are marked by cyan boxes.
Scene text localization and recognition can be integrated into wearable devices and help blind or visually-impaired people to distinguish hand-held objects through the text characters and strings on their surfaces.
!(flowchart_handheld_system.png "Figure 2. Scene text localization and recognition are reading text from hand-held objects for blind users.")
Figure 2. Scene text localization and recognition are reading text from hand-held objects for blind users.
Scene text instance is also important to transit navigation, especially for self-driving cars. Text could provide unambiguous information about the surrounding environments, and guide humans/robots to make correct decisions about their way-findings.