algorithm - Suggests or methods of tv logo auto finding/locating/detection -
usually logo detection means find logo , recognize logo. common works 2 steps using sift/surf matching method, detailed in
but, if logo tiny , blur, result poor, , kind of time consuming; want split 2 steps, firstly finding logo in video; recognize logo using template matching or other method, like:
my problem focused on finding logo automatically in video. tried 2 methods:
- brightness method. logo on tv screen there when show goes on, select list of frames randomly , difference between frames, logo area tend 0; statistics of 0 brightness threshold determine whether pix logo or not. method failed while show has static background.
- edge method. likely, if logo there, border tends obvious. statistic work brightness method, edge unstable,such bright background.
are there suggestions or state of art methods auto finding logo areas , other logo recognition method except sift or template matching ?
let's assume list of logos known before hand and have access examples (video streams/frames) of logos.
the 2017 answer question train logo classifier, , deep neural network.
with sufficient training data, if identifiable tv viewers able detect it. able handle local blurring , intensity changes (which may thwart "classic" image processing methods of brightness , edges).
opencv can load , run network models multiple frameworks caffe, torch , tensorflow, can use 1 of pre-trainined models or train 1 yourself.
Comments
Post a Comment