I am currently trying to distinguish between an image's foreground and its background using texture of sub-image regions. I have use Law's texture energy calculation but i am not satisfied am not satisfied with the result, I need better method for this purpose.
Have you tried Speeded Up Robust Features (SURF)? In some of my works, this methods provided the best result for texture pattern recognition.
We would really need to see a sample image as noted above.
It might also help if you can briefly summarize why you are not satisfied with what you are now getting for segmentation -- lost objects in foreground? too much background kept as foreground? ruined boundaries of foreground? foreground broken up throughout?...
Hi Abass,
You need to explicitely state what is the problem you are facing and the imaging set-up you are dealing with. For a general answer I would refer you to a thread in which we discuss a related topic on texture extraction (see the link below). It might hint you with some ideas.
As a side note, teasing apart what is background and what is foreground brings us, quite often I would say, to the topic of stereo-imaging where a given scene is projected from two different angled cameras. Therein, a disparity map is calculated to form a response to what is considered a background or foreground.
https://www.researchgate.net/post/Can_texture_features_find_differences_among_the_same_visually_homogeneous_texture?tpr_view=1801b8d4-4bea-489a-bc4d-98bee2b83a7d_3
A sample image is shown in Figure w of the paper below;
https://www.researchgate.net/publication/272492822_UNSUPERVISED_REGION_OF_INTEREST_DETECTION_USING_FAST_AND_SURF
Conference Paper Unsupervised Region of Intrest Detection Using Fast and Surf
Is it possible to use Markov Rondom Fields as texture features extration. It is very good.
- Badges
- Science topic