Text Detection from Scene Videos having Blurriness and Text of Different Sizes

摘要

Text in videos needs to be detected correctly as it contains important information related to natural scene in videos. Text detection is useful for automatic number plate recognition, street boards reading, Optical Character Recognition (OCR), automatic document scanning and to help blind or visually impaired people. The existing scene text detection techniques are unable to detect text accurately from the videos having low contrast, complex background or excessively small fonts. Hence, we propose a new text detection technique that enhances accuracy of detecting text and reduceses average processing time. Our text detection technique incorporates Edge-enhanced Maximally Stable Extremal Regions (eMSER) method to preserve shape of characters and modified fuzzy C-means clustering to converge faster. In this paper, we provide an experimental analysis of our improved text detection technique. To show the effectiveness of our proposed text detection technique, we have performed experiments considering videos having text of different sizes as well as considering blur videos. The experimental results show that an improved text detection technique outperforms an eMSER based text detection technique.