LAB 10: IMAGE PROCESSING

Transcription

1 NAME: LAB 10: IMAGE PROCESSING This laboratory exercise will involve you writing scripts to analyze several Earth science-related images. With the rise of satellite data availability and microscope imagery, quantitative image processing techniques are a valuable tool for the Earth scientist. MATLAB has several built-in functions that make it an effective tool for performing quantitative calculations on raster images. Your final code will be handed in via a single zip file following the instructions at the end of the lab. Make sure to comment your code and include the proper MATLAB header information in your script. Some of the SEM description were taken from the Science Education Resource Center s website. Part I: Making Your First Image: 4 Pixels of Fun Before we embark on something scientifically useful, let s start simple by making two 4 pixel images by hand. Your task is simple. Write a script called makeimage.m That reproduces the figure below (Figure 1, two plots in the same figure window). Note that the first image must be made using indexed color (colormapped) and the second using true color (RGB). Other than that, the two images are identical. Note that the pixel colors are pure yellow, green, blue, and red. Make your plots using image. Figure 1: Plot of two 4 pixel images that will be reproduced with makeimage.m Page 1 of 6

2 Part II: Automated Point Counting in a Scanning Electron Microscope Image The scanning electron microscope (SEM) uses a focused beam of high-energy electrons to generate a variety of signals at the surface of solid specimens. The signals that derive from electron-sample interactions reveal information about the sample including external morphology (texture), chemical composition, and crystalline structure and orientation of materials making up the sample. In short, there is arguably no other instrument with the breadth of applications in the study of solid materials. For this reason, the college of Arts and Sciences at Appalachian State University has a fully-functional SEM that faculty and students from several departments use for their research and classes. A common (and very tedious task) is called point counting. This involves randomly selecting a point on an SEM image and identifying the mineral. This is repeated tens to hundreds of times, and then the percentage of each mineral can be estimated. Your task is to write a script that reads in a grayscale SEM image, identifies the composition of each pixel, and then calculates the percentages of each mineral. Figure 2: SEM image of a rock sample from the Dolomite Mountains, Italy (image courtesy of Dr. Carmichael). The composition consistently varies with grayscale value. See the table below for the grayscale ranges in this image, but note that different SEM images will have different grayscale ranges for a given mineral. Pixel Composition Color Name Grayscale Range Iron Oxides Light Gray to White > 175 Calcite Medium Gray Dolomite Dark Gray Porosity Very Dark Gray to Black < 61 1) Make a new script called processsem.m 2) Go to the course website and download and extract the zip file for this lab. It contains and SEM image named L-8F.jpg. This is the SEM image that you will use here. Because there are four possible compositions for each pixel, you will need to pre-allocate a matrix completely filled with white pixels for each composition. Name these matrices IronOxide, Calcite, Dolomite, and Pores. Because this rock unit is a hydrocarbon reservoir, determining the porosity is especially important. 3) Load in the image as a matrix and loop through the entire matrix to identify which of the 4 different compositions applies to each pixel. When a given pixel is identified, make the corresponding pixel in the appropriate matrix (i.e. IronOxide, Calcite, Dolomite, or Porosity) black. 4) To visualize your results and make sure that the pixels you selected look reasonable, you should make a plot of each image matrix in a single figure. Please use the gray colormap, but make sure that everything is properly scaled. All of the figures should have the titles shown below. Note that all Page 2 of 6

3 of the images will consist of only either black or white pixels, except for the raw image, which is grayscale. The black pixels just show the selected pixels. If we just grabbed the actual grayscale value, many of the calcite and iron oxide pixels would be very close to white, and thus not visible. So, by forcing the selected pixels to be black, you will be able to clearly see which pixels were selected in each case. RAW Image Pore Space Dolomite Figure 3. The layout required for the 5 plots of the RAW SEM image and the four possible pixel compositions. Calcite Iron Oxides 5) Your script then calculate the percentages of each type of pixel for the entire image and print the results to the screen stating the abundances to two decimal places. Your printout should look like the example below (but with actual numbers), with all of the % values lined up neatly. The print command should refer to variables, so if a different SEM image were loaded, the command window output would automatically update. Pixel Abundances Porosity: xx.xx % Dolomite: xx.xx % Calcite: xx.xx % Iron Oxides: x.xx % Total: xxx.xx % Page 3 of 6

4 Part III: Detection of Snow and Ice in a Landsat Image Climate change is a hot topic right now (pun intended!) wait, didn t I make that same joke last lab? Anyway, one of the key pieces of evidence for recent warming is the decrease in glacier ice mass in numerous locations throughout the world. One such retreating glacial body is the Quelccaya Ice Cap in Peru (Figure 3). Figure 4. The Quelccaya Ice Cap, Peru as viewed by the Landsat satellite on September 16, Note the distinctive color of the snow and ice covered regions. Your script should detect both the snow and ice covered regions relatively well, but not the water bodies or other dry land. The Quelccaya Ice Cap in Peru is situated at 13.5 degrees South latitude, which puts it in the tropics between the Tropic of Capricorn and the Equator. The Quelccaya Ice Cap contains collection of glaciers that grow on a high-altitude plateau in the Andes Mountains, between the Amazon jungle to the east and tropical, eastern Pacific waters to the west. Your task is to write a script that reads in two Landsat images of the Quelccaya Ice Cap from two different time periods (22 years apart) and to compare the total ice cap surface areas. If the ice cap is retreating, you should see a decrease in total snow/ice surface area with time. I have provided the two images (Quelccaya_ jpg and Quelccaya_ jpg) in a zip file on the course website. The number string after the filename gives the year, month, and day of each image. For more information about these images and the region in general, go to 1) Make a new script called compareicecap.m. 2) Your script should read in both images into MATLAB and store each in a separate matrix. Store the first image in imat1988 and the second image in imat2010. Once this is done, like the previous section, you should pre-allocate two new matrices that are the same dimensions as each image, but Page 4 of 6

5 full of white pixels only. Call these matrices imat1988b and imat2010b. We will use these matrices to later store only the selected snow/ice pixels. 3) You should write a nested for loop for each image (i.e. two nested loops) to search though all of the pixels and determine which ones are covered in snow/ice. To do this, I will save you a ton of time by giving you the rules to determine if a pixel is covered in snow/ice. Any pixel that meets the following conditions can be considered to be covered in snow/ice and thus part of the Quelccaya Ice Cap. Red value < 100 Green value > 130 Blue value > 155 When your code identifies a snow/ice pixel, you should store the color values for that pixel in your imat1988b or imat2010b matrix in the corresponding location. This way, your imat1988b and imat2010b matrices will plot as images that only show the colors of the pixels that are snow/ice with everything else being left white. This will allow you to visually verify that your code is selecting snow/ice pixels and not other colors. Fair warning! Pay attention to the class of your variables, and be sure to use color ranges that are appropriate for each class. 4) Next, to compare the total surface areas of the Ice Caps at these two different times, your code should calculate the total Ice Cap surface area for each image in km 2. The provided images have a resolution of 30 m/pixel. Then your code should calculate the total change in km 2 and rate of change in Ice Cap surface area in km 2 /yr. Your script should print out the following messages to the command window: Total Ice Cap Areas 1998: xx.xx km^2 2010: xx.xx km^2 Total change: xx.xx km^2 Rate of change: xx.xx km^2/yr All of these numbers should come from variables, so if different images were provided, the printed out statements would automatically update. To keep things simple, you can assume that the two images were taken exactly 22 years apart in time. 5) Your script should make 4 figures. The first figure has the raw 1988 image plotted, with a hard coded title of RAW Image The second plot should show the pixels that were detected as snow/ice using the criteria above. The title should be Quelccaya Ice Cap Extent: The next two figures should show the same things but for the 2010 image. Page 5 of 6

6 Part IV: What to Hand in? Like previous labs, you should zip up all of your files and them to me as a single.zip file. Use the appropriate MATLAB command to zip everything into one file. Call your file lab10_lastname.zip. For your convenience, I have provided a list of the required files below. Script Files: Data Files: makeimage.m, processsem.m, compareicecap.m Not needed. I will have copies of the data files in the same directory as your scripts. Page 6 of 6