SPECTRAL TUTORIAL ================= Getting Started --------------- Start up Mosaic, put it in the background hys298 >> Mosaic& Follow the links to the Spectral Tutorial. That should bring you to here. Move to the directory where you want to work. There should be a directory that resembles your name. This holds your images. As an aside, when your following along in this tutorial, you can use the mouse to cut any of the commands shown here, and paste them in your window. This will save time and reduce typing errors. hys298 >> pwd /proj/hys298 hys298 >> ls andrew/ dem.ipw ipwxv maddogg/ prhdr ann/ docs/ isaac/ make_rgb* quinn/ app-defaults/ eric/ jeff/ make_rgb~* rgb.ipw bill/ george/ joann/ martha/ scott/ bob/ gf.ipw julie/ miguel/ ustin/ cat/ glen/ karen/ mux wanli/ claudia/ hoosse/ katie/ num.ipw william/ cynn/ index.html ken/ paul/ yululong/ david/ index.html~ lincom peter/ hys298 >> cd quinn hys298 >> pwd /proj/hys298/quinn hys298 >> ls blu.hst bw.pcx grn.pcx red.pcx rgb_lin.pcx blu.pcx grn.hst red.hst rgb_hist.pcx users.etx Start another terminal window. This is the window you will use to do your spectral work. hys298 >> dxterm& The software (ipw) that you will be using is only available on one machine in the lab. That machine is vaca. You need to login to this machine to use the ipw software. So in your new window: hys298 >> xrlogin vaca You're now logged onto vaca. Cd to your directory. hys298 >> cd quinn hys298 >> pwd /proj/hys298/quinn hys298 >> ls blu.hst bw.pcx grn.pcx red.pcx rgb_lin.pcx blu.pcx grn.hst red.hst rgb_hist.pcx users.etx Viewing your data ----------------- We're going to be using a free software package known as xv to do our image viewing. Try this out be looking at some of our images. Type: hys298 >> xv& A window will appear on your terminal. Move yor mouse into this window and click your right mouse button. The XV control panel will appear on your terminal. Click the 'Visual Schnauzer' button in this panel. Because of the display aspects of your terminal, the Visual Schnauzer window will look strange until you move your mouse pointer into it. When you do, the rest of the terminal will appear strangely colored. This is just a method that is used by the program to allow the terminal to display more colors than are normally allowed. This will bring up a graphical version of your directory. Click the 'Update' button in the Visual Schnauzer window and thumbnail versions of your images will be created. You can double click on any of the images to load it and display it. Double click on grn.pcx right now. This is the green filtered image. Note that using the buttons along the bottom of the xv control panel, you can change the size, rotate, mirror, crop, etc. your image. Unless you save the new image, if won't affect your file at all. Play around with the controls for a little bit. Probably the best tool that any image processing software can have is a way to manipulate the intensities and color shades of a digital image. Xv has very nice capabilities for this kind of thing. Click on the ColEdit button in the Xv control panel, and investigate some of these. Basically, the Color editing panel is divided into three areas. The left most allows manipulation of individual pixels, the middle allows HSV Modification, and the right most one allows rgb Manipulation. You really need to play with this some to get a feeling of how this works. Since this image is a grey scale image, the tools aren't quite as interesting as they can be. However, the Intensity window, can result in more balanced contrast, or thresholds, etc. Simple Classification --------------------- Double click on one of the color images and reexamine the Color Edit functions. One thing you may notice is that you can now imagine ways that you might classify your image. For example, you could change all the greenish pixel colors to bright green. You now have a classified image of green regions. Pick a greenish pixel, use the RGB color wheels to change the color to all green, no red, no blue. Do the same for a number of the greenish pixels. Now, use the Intensity window to only display values of 255. The resultant image is a classification of the image, displaying areas in the image that are green. Save this to a file, Click on the 'Save' button in the XV control panel. Choose an X11 bitmap format, and call the file gv1.bit. You can now use your schnauzer to load this image, and verify. You might think that this an overly simple classifiaction, but really this is an example of a Supervised Maximum likelyhood classification scheme, one of the most popular classification methods used. Three Color Image ----------------- The color images that are in your directory did not come from the digital camera. The camera output was simply three grey scale images that had different color filters. Some image processing software was used to put these images together. Lets redo this process. The steps involved are : **1. Reformat the image file to a format ipw can use.** Here we use a combination of two free software packages, and some unix commands to accomplish this task. hys298 >> pcxtoppm red.pcx | ppmtopgm | dd bs=15 skip=1 | \ > mkbih -l 360 -s 496 > red.ipw Don't be concerned if you get a few error messeges, the files should be ok anyways. These are 4 commands strung together. **pcxtoppm** converts the pcx file to a portable color file. **ppmtopgm** converts this to a normal grey scale image, that is just the raw bytes plus a small header. **dd** removes the header. **mkbih** is an ipw command that creates an ipw format file from a raw file. You must specify the number of rows and columns in your image (you cn get these using xv). The resultant file is stored in red.ipw. You can look at the header information of this file using another ipw command. hys298 >> prhdr red.ipw !

basic_image_i -1 $Revision: 1.10 $ byteorder = 0123 nlines = 360 nsamps = 496 nbands = 1 !

basic_image 0 $Revision: 1.10 $ bytes = 1 bits = 8 !

image -1 $Revision: 1.5 $ This shows the size of the file, and the number of bands, etc. If you would ever like to see a listing of what an ipw command does you can get a description using the command ipwman. For example, hys298 >> ipwman prhdr Repeat this process for the blu.pcx and grn.pcx. You should then have three ipw files. **2. Put the three bands together.** Use the ipw command: hys298 >> mux red.ipw grn.ipw blu.ipw > rgb.ipw hys298 >> prhdr rgb.ipw !

basic_image_i -1 $Revision: 1.10 $ byteorder = 0123 nlines = 360 nsamps = 496 nbands = 3 !

basic_image 0 $Revision: 1.10 $ bytes = 1 bits = 8 !

basic_image 1 $Revision: 1.10 $ bytes = 1 bits = 8 !

basic_image 2 $Revision: 1.10 $ bytes = 1 bits = 8 !

image -1 $Revision: 1.5 $ Note that now we have a file with three image bands. 3. Change the three band image back into a file format that xv can use. hys298 >> rmhdr rgb.ipw | rawtoppm -rgb -interpixel 496 360 > rgb.ppm Now look at this file using Xv. Compared to the color images you have, this one looks pretty washed out. That's because the other images were stretched to give them more contrast, and this one wasn't. You might also notice that this image is in 24-bit mode, where the previous images you were looking at were not. You could use xv to perform constrast streching on your image, (look at the .hst files) to obtain less muddy images. But we want to perform the some more analysis on this image, so let's keep it the way it is. Green Vegetation Indicator -------------------------- Before, we made a classified image of the green veg areas by classifing certain pixels in the 3 color image. We couldn't use this technique on the green image alone, because there are lots of areas that are high in green but also high in red and blue too. What we'd like to come up with is an indicator that somehow took this into account. We want an index that is high whenever green is bright, in comparison to the red and blue values. If we did something like : Greenness = green-blue + green-red We'd get positive values when green was brighter than blue and red, and negative value otherwise. However, in the above scale, Dark green pixels would have a lower greenness factor than bright green pixels. Since, were not sure that's accurate, lets normalize the function to the overall brightness of the pixel. Green Factor = ((green-blue)+(green-blue))/(green+blue+red) We can accomplish this in ipw with the commands lincom, mux and mult. hys298 >> lincom -c -1,2,-1 rgb.ipw > num.ipw hys298 >> lincom rgb.ipw > dem.ipw hys298 >> mux num.ipw dem.ipw | mult -r 1 > gf.ipw hys298 >> rm num.ipw dem.ipw hys298 >> rmhdr rgb.ipw | rawtopgm 496 360 > gf.pgm Part of the problem with the setup that we have is that we have to constantly make more than one version of the files, one for ipw to read and one that xv can read. It would be much better if we could avoid that step. Well, we can. Since xv is a public domain software, the developers of xv modified it to read ipw files. YOu can use this version by using 'ipwxv&' on the machine vaca, instead of 'xv&'. This is an older version of xv however, and you don't get all functionality, like the Visual Schnauzer. Now look at this image in xv and you can see that the green areas show up brighter in this image. You can again use the Intensity window in the Color Edit to threshold this image. Save the thresholded image, and compare it to your first classification scheme. The above procedure is very similar to using NDVI to determine vegetation locations in satalitte imagery. In NDVI, though the green band is not used, but Near IR bands instead. Why? Unsupervised Classification --------------------------- Now, we will perform an unsupervised classification on our image. We will use clustering. In clustering, signatures in the image are generated automatically. They are chosen so as to be unique and to span the range of signatures in the image. Since the images are classified automatically, we don't really know what the resultant classes are, and normally another step is taken to assign the classes to some real land cover. Create some clustered classifications with the commands hys298 >> ustats -c 10 -r 10 rgb.ipw > class10.stats hys298 >> ustats -c 10 -r 2 rgb.ipw > class2.stats hys298 >> bayes -s class10.stats rgb.ipw > class10.ipw hys298 >> bayes -s class2.stats rgb.ipw > class2.ipw These images look black when viewed in ipwxv. This is because the image has values only between 0 and 10. These all appear black with the default color table. Go into the Color Edit window of ipwxv and change the color table to Random. Do any of the classes seems to correspond with the green veg areas? How about other classes? Why do you think they are different? What can you tell about the illumination from this classified image? Supervised Classification ------------------------- Now, let's revisit the Supervised Classification. Reload a color image, and try and use xv to pick out a small number of pixel colors that you can use for a number of different classes in your scene. Try and get about 4-5 different classes. You want your xv image to have only 5 different colors, and the rest of the image black. You can use any color image so maybe the rgb_lin.pcx is the best choice. Save that image and convert to ipw (from above you should know how to do this) with the name sup.ipw. Now, you want to compute the stats for the different classes that you identified in this file, and do another classification. hys298 >> mstats -m sup.ipw rgb.ipw > sup.stats hys298 >> bayes -s sup.stats rgb.ipw > supclass.ipw hys298 >> bayes -s sup.stats -t 2 rgb.ipw > supclass2.ipw Look at these images and decide how well they classified your image. Also, what do you think the -t 2 in the second command means?