Changes between Version 15 and Version 16 of GSoC/2016/Additional_segmentation_algorithms/mean_shift

Timestamp:

Sep 13, 2016, 9:05:35 AM (8 years ago)

Author:

hao2309

Comment:

--

GSoC/2016/Additional_segmentation_algorithms/mean_shift

@@ -167 +167 @@
 
 [[Image(New_MS_Fig1.jpg)]]
+
 Compare with the original image, there are many small segmentations showing because minsize has been set to 1.
 Then zoom into a smaller region and compare with the original image.
+
 [[Image(New_MS_Fig2.jpg​)]]
 
 Compare with the original image, most of the features in original image has been segmented out. the noise can be significantly reduced by increasing the threshold of minsize​.
+
 [[Image(New_MS_Fig5.jpg​)]]
 
+== Segmentation using different range bandwidth and spatial bandwidth ==
+This example uses fixed range bandwidth(hr) value, larger hr value indicates pixel value in a larger range take part in the filtering calculation. 
+[[Image(New_MS_Fig3.jpg​)]]
+
+keeping all else constant, changed the value of spatial bandwidth. A larger spatial bandwidth hardly results in larger segments.
+
+[[Image(New_MS_Fig4.jpg​)]]
+
 == Segmentation using adaptive bandwidth and progressive bandwidth ==
-
-
- 
+This example uses -a and -p flag to add adaptive bandwidth and progressive bandwidth respectively 
+
+[[Image(New_MS_Fig6.jpg​)]]
+
+In the zoomed-in image, adaptive bandwidth enhanced the result significantly. Because range bandwidth is the most important parameter for mean shift.
+
+= REFERENCES =
 [1] http://stackoverflow.com/questions/4831813/image-segmentation-using-mean-shift-explained [[BR]]
 [2] Adams, Rolf, and Leanne Bischof. "Seeded region growing." IEEE Transactions on pattern analysis and machine intelligence 16.6 (1994): 641-647[[BR]]