Automatic ROI detection

asked 2016-12-26 23:17:29 -0600

AJ94 gravatar image

Hi All,

I am working on foreground extraction using Grabcut algorithm which will extract the object located inside the region of interest. However, instead of specifying the ROI manually, is there any way we can detect the object automatically without explicitly specifying the ROI? I am using python.

Thanks in advance

edit retag flag offensive close merge delete

Comments

You try SIFT algorithm, https://robwhess.github.io/opensift/

greenworld gravatar imagegreenworld ( 2016-12-27 02:07:17 -0600 )edit

@greenworld Sorry I don't understand. How can you use sift to initialize grabcut?

@AJ94 you can read this paper to help you :

Automatic GrabCut color Image Segmentation Based on EM Algorithm

LBerger gravatar imageLBerger ( 2016-12-27 05:00:09 -0600 )edit

@greenworld i don't think so sift can initialize grabcut. @LBerger Is any code reference available?

AJ94 gravatar imageAJ94 ( 2016-12-27 05:27:31 -0600 )edit

What kind of object you want to detect? There is a lot of detection techniques in opencv.

pi-null-mezon gravatar imagepi-null-mezon ( 2016-12-28 06:45:35 -0600 )edit

@pi-null-mezon The object can be anything ranging from car,tree to human being. Is it possible to have any algorithm which will automatically detect object from unknown images??

AJ94 gravatar imageAJ94 ( 2016-12-29 00:34:21 -0600 )edit

if you want to detect text you can use er filter text detection present in text module of open cv contrib otherwise you can use the function find contours (incase there is no specific object )and then apply grabcut on the corresponding rectangle or you can try using mser

shreya gravatar imageshreya ( 2016-12-29 13:13:23 -0600 )edit