Home
People
Research
Publications
Courses
Seminars
Links

1

Software

NOTE: The following software is provided for research purposes only. You may not distribute this software in any form without the prior approval of the University of California. 

Copyright (C) 1999, The Regents of the University of California, Samsung Electronics Corporation.

All rights reserved. 6/19/99 Author: Yining Deng and B.S. Manjunath

THE AUTHOR DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE, INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS, IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER NOTORIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.

Download (Click right mouse button and choose save link as) :

SUN Solaris

SGI IRIX 6.3

Windows 95/98/NT

Source code: 

The following source code jseg.zip (182 Kb),  with supporting JPEG free source code jpeg-6b.zip (743Kb), is provided for research purposes only. You may not distribute this software in any form without the prior approval of the University of California.

PC Windows: 

The programs are built to "Console Application" using visual C. You can use dos prompt to run the program and specify the command line arguments.

To run the program, type ProgramName to show a list of command line arguments:

Usage: segdist {arguments} {options}  

arguments must be provided: 

-i file  input media filename  

-t type  input media type:     

      type 1,2,3 must provide image size

 1:  image yuv        

5:  image ppm 

 2:  image raw rgb 

6:  image jpg   

 3:  image raw gray

9:  image gif  

 4:  image pgm 

Options: 

-o file factor 

output image (region boundary superimposed} filename

output image format same as input image

factor: dim original image to show boundaries, 0-1.0 

-rn file 

output region map filename, n type

   3: image raw gray 

   9: image gif  

-s height width

image height and width

-q thresh

color quantization threshold, 0-600, default automatic

-m thresh

region merge threshold, 0-1.0, default 0.4

-l scale 

number of scales, default automatic

 

Example:  

segdist -i test.rgb -t 2 -o test.seg.rgb 0.9 -s 128 192 -r9 test.map.gif -v test -q 255  

The region map file is a gray-scale image. It labels the image pixels. If pixel (0,0) belongs to region 1, its value is 1. The label starts at 1 and ends at the total number of regions. Your probably will only see black when you try to view the map image. That's because most values are too small. If you equalize the image, you'll see patches of regions. The equalization changes the original pixel values though.

The region map is very useful for many applications, for example, if you want to process all the pixels in a particular region of interest. On the other hand, if you want to get the region boundaries, you need to do a little job by yourself. For example, you can check if each pixel is of same label as its neighboring pixels. If no, it indicates that pixel is at the boundary.

Program limitation: The total number of regions in the image must be less than 256 before the region merging process. This works for most images smaller than 512x512.

Processing time will be about 10 seconds for an 192x128 image and 60 seconds for a 352x240 image. It will take several minutes for a 512x512 image. Minimum image size is 64x64.

 

Optional Parameters

color quantization threshold - specify  values 0-600, leave blank for automatic determination. The higher the value, the less number of quantized colors in the image. For color images, try 250. If you are unsatisfied with the result because two neighboring regions with similar colors are not getting separated, please try a smaller value say 150.

number of scales - The algorithm automatically determines the starting scale based on the image size and reduces the scale to refine the segmentation results. If you want to segment a small object in a large-sized image, use more number of scales. If you want to have a coarse segmentation, use 1 scale only.

region merge threshold specify values 0.0-0.7, leave blank for default value 0.4. If there are two neighboring regions having identical color, try smaller values to avoid the merging.

Home | People | Research | Publications | Courses | Seminars | Links