Я подозреваю, что самый прямой подход будет примерно таким:
for each frame in video
convert to HSL
select connected pixels which have luminance above a certain threshold
convert those areas that are "large" to regions
expand regions or soften the region (soft mask, Gaussian blur)
mask original image to that area
apply either brightness/contrast or histogram equalize on that area
overlay back onto original image