Color Segmentation High Resolution Aerial Images
As an experiment in using color segmentations for building
extraction we tried several steps. Images are from Purdue.
30 micron scanning. Giving a 7692X7692 image.
- Basic color segmentation algorithm:
Ohlander, R., Price, K.E., and Reddy, R.,
Picture Segmentation by a Recursive Region Splitting
Method, CGIP(8), 1978, pp. 313-333.
- Extract a subwindow of the image.
- Apply Segmentation using only RGB -- results were poor, no
examples are shown.
- Use the color transforms from Ohta
Density: (R+G+B)/3. I1: 3R/2 - 3B/2. I2: 3G - 3(R + B)/2.
The last two are shifted by 128 and forced to fit in the 0-255
Apply the segmentation on this triple of images.
See: Ohta, Y.[YuIchi], Kanade, T.[Takeo], and Sakai, T.[Toshiyuki],
Color Information for Region Segmentation,
CGIP(13), No. 3, July 1980, pp. 222-241.
- Given the resolution, First reduce the image and apply the
segmentation of this reduced image.
- For display simplicity, test images are around 900X900.
To limit the size of this file -- go to the
Next page to see the real
results. Essentially, 50-150 regions in the 900X900 images,
execution times around 15-20 seconds.