Image Segmentation - Berkeley Dataset

Question

I've been given an image from Berkeley dataset to segmentate. I am an undergrad student and so far I threw everything i know to this thing. But the colors are extremely close. I tried;

Clustering
Edge Detection
Histogram Thresholding

I need suggestions. Which path should I take? I need to segmentate the animals from this picture but I was only able to remove the green leafs.

enter image description here

You should consider reading some kind of survey of methods. That is what I do when I start working on something new — Andrey Rubshtein, Dec 23 '14 at 10:57

score 0 · Answer 1 · answered Dec 23 '14 at 09:21

0

You can try grabcut algorithm. Check this link: Interactive Foreground Extraction using GrabCut Algorithm

answered Dec 23 '14 at 09:21

ekarem

11
2

Thanks for your answer. This is also an algorithm i would like to try and Berkeley's website states graphcut algorithms work well with this image. But I applied SLIC + DBScan so that's why I accepted Ameya005's answer. – Mert Çelikok Dec 25 '14 at 02:04

score 0 · Accepted Answer · answered Dec 23 '14 at 13:39

0

Take a look at SLICO : http://infoscience.epfl.ch/record/177415 .

Superpixels along with graphcut/grabcut methods are quite good at similar segmentation problems. I have had really good results with SLICO and MRFs/Simulated Annealing approaches.

answered Dec 23 '14 at 13:39

Ameya005

16
1

SLIC was quite useful thank you. I used Peter Kovesi's article and applied SLIC, afterwards I applied DBSCAN clustering for merging superpixels together. – Mert Çelikok Dec 26 '14 at 22:39

Image Segmentation - Berkeley Dataset

2 Answers2