negative images in haarcascade - training

asked 2013-11-12 16:21:24 -0500

ioanna gravatar image

updated 2013-11-12 16:53:29 -0500

I would like to detect specific characteristic from laparoscopy surgery. The item that I want to detect appears here https://www.dropbox.com/s/n0yxrt7o4llv6l8/im1_55.bmp?m The negative images that I use are similar with that https://www.dropbox.com/s/lse4qzi0drn0m1q/im_14.bmp But when I take the xml file and use it in my program can not recognize any object. Do anyone know what is the problem?

edit retag flag offensive close merge delete

Comments

your 2nd link requires login.

berak gravatar imageberak ( 2013-11-12 16:40:32 -0500 )edit

sorry I fix it

ioanna gravatar imageioanna ( 2013-11-12 16:54:18 -0500 )edit

honestly , if you don't get any detections, it's probably, because your positive images don't have any clear shape in it.

what's the thing, you're actually trying to find ? the tip of the instrument ? some kind of 'skin state' ?

berak gravatar imageberak ( 2013-11-12 16:59:39 -0500 )edit

the bubbles that appear when doctor burn the tissue.

ioanna gravatar imageioanna ( 2013-11-12 17:03:53 -0500 )edit

lol, ok. i see ;)

still, haar classifiers try to approx the scene with combinations of small boxes, hog classifiers with 'gradient' lines.

you need proper contours for that, not a mass of foam ( whoever was responsible for that .. )

so, it's the instrument, you're trying to track ?

berak gravatar imageberak ( 2013-11-12 17:08:42 -0500 )edit

instrument when burn something cause I want the effect of the operation

ioanna gravatar imageioanna ( 2013-11-12 17:21:14 -0500 )edit