|
In
this talk we present a multi-frame based enhancement scheme to
improve the OCR of artificial text (movie credit and news caption)
and the readability of scene text (car license plate) in digital
video. From the video text creation process we model the enhancement
of artificial text as a denoising problem and the enhancement
of scene text as a deblurring problem. Our enhancement algorithm
for artificial text is based on the image registration and fusing,
and we propose a projection onto convex sets (POCS) based method
to enhance scene text. We will show experimental results and discuss
the conditions under which the algorithm performs best.
|
