Online time numeral recognition from monochrome video sequence

This paper presents a online multi-font numeral recognition method, whose main aim is to recognize overlaid time numeral from video. The portion of the video frame containing the time text is binarized and segmented. Minimum rectangular bounding box is inserted over the isolated numeral images. Euler number of numeral images is found out to initially differentiate into three groups. Then, the numerals are recognised by considering the individual distinct features of each numeral within the group. This recognition process is carried out for all numeral positions. Sometimes it is seen that either due to noise in video sequence or because of poor quality of recorded! online video, some of the numerals are not recognized. Therefore, image correlation is performed as the last step, only for the unrecognized numeral images. Experiments were conducted on various infrared and CCD camera video sequences as well as on commercially available CCTV CCD videos, by taking different sizes and types of fonts. The method provides accuracy of 98.67% and the per-frame computation time is 15.413 ms which is fast as per interlaced video.