Automatic Thai Character Extraction from Digital Video

K. Porkaew and C. Sibunruang (Thailand)


Multimedia Database Systems / Video Processing / Video Indexing / Text Extraction


Most texts appearing in a video often summarize the video content. Thus, it is reasonable to use these texts to build keywords for video retrieval. Unfortunately, current Thai Optical Character Recognition (OCR) technology recognize texts in digital video not well since they emerge from the complex background, and their locations are unpredictable. This work consists of two phases: Text Detection referring to locating texts in digital video by using Wavelet Transform, and Text Segmentation referring to separating texts from the complex background by using color and intensity analysis. The experimental results show that the proposed approaches can detect text location with the accuracy of 89% for graphic text. The result of text segmentation from complex background is 93.2% for less complex background and 83.8% for more complex background.

Important Links:

Go Back