Lab 5 Image Mosaic

Remote Sensing Lab 5

Ethan Nauman

11/13/14

The goal of this lab was to introduce us to important analytical processes in remote sensing. We explored image mosaic, spatial and spectral image enhancement, band ratio, and binary change detection. The process of image mosaicking in this lab is dealing with multiple images and being able to display it as one seamless image. By the end of the lab I was able to complete image mosaicking along with the other skills that I learned.

Part 1: Image Mosaicking

The first part of this lab dealt with being able to combine multiple images into one seamless image covering a wide area of interest. To begin this lab I brought in an image provided for me into Erdas. However, before bringing the image in I had to make sure all the prerequisites fit the image. Since I was bringing in multiple photos, I had to check the boxes in the image file allowing for me to bring in multiple images. After completing this process and uploading the first image, I then had to conduct the same steps allowing me to overlay the first image with a second one. After both of the images were in the viewer, I then had to perform the mosaic on them. For this process I used the simple mosaic express tool. This was located under the raster tools, then under the mosaic express tool. Once the mosaic window appeared, I then had to upload both of the images into the mosaic tool. After uploading both of the images, I selected the file that I wanted the mosaic to store the mosaicked image. This was located in my personal folder. After completing these steps I then ran the mosaic express tool. After running the tool I then brought the two mosaicked images into a new viewer and they appeared as this. 

Section 2: Image mosaic with MosaicPro

In this section I was again going to mosaic the two images I stitched together, however this time I was going to use a more advanced mosaic tool, MosaicPro. To begin this section of the lab I brought in the original two images that I laid over each other. I went into the raster tools and the mosaic tools, instead of using mosaic express this time I used the MosaicPro tool. Once completing this step of getting to the mosaic window, I then had to upload both of the images again into this tool. Before completing the upload for each image I had to change some of the image area options. I used the compute active area button in this option, upon looking over the parameters I accepted the terms and uploaded the first image. I then followed the same steps to upload my second image and accepted all the parameters likewise. After uploading both of the images I had to make sure that the second image I uploaded was the bottom image. Now I wanted to synchronize the radiometric properties at the area of intersection of both images, that way there would be a smooth color transition from one image to the other. To do this I used the color corrections tool. After opening this tool I used the match histogram option, this opened another tab that allowed me to select the set button. The set button allowed me to select the overlap areas button in the dialog box, this allowed for a smooth transition between the two images but didn’t affect the brightness of the rest of the images. I accepted all other parameters in the color corrections dialog box. On the MosaicPro toolbar I then selected the set output options dialog icon to open the output image options window. This window would allow me to change map projections or pixel size if I choose. I accepted all parameters and closed the window. I then clicked on the set overlap function icon; this opened another window for me. I accepted the default parameter of overlay, which used the brightness values of the top image in the area of intersection. After filtering through all these parameters I then ran the tool. The image that appeared had a more smooth transition from one image to the other and the colors matched better. 

Part 2: Band Rationing

In this section of the lab I would perform band ratio by implementing the NDVI found on the original image of Eau Claire. I started with a blank viewer and then upload the original Eau Claire image. I used the raster toolbar and used the unsupervised tool and used the NDVI tool on the drop down. This opened the indices window. My input file was the original Eau Claire image and my output file was saved and located in my personal folder for the class. I also had to make sure that the sensor parameter was Landsat TM and the select function parameter had NDVI selected. I accepted all other parameters and ran the tool. The image that appeared caught me by surprise because it was a mostly white image. The white of the image was highly vegetated areas on the image.

Part 3: Spatial and spectral image enhancement

For this section of the lab I uploaded the Chicago TM image from our image enhancement folder for our class. The image demonstrated some amount of high frequency, which needed to be suppressed. For the first portion of this section I used a 5x5 low pass convolution filter. This was located on the raster tool bar under the spatial icon, which I followed the drop down to convolution. This opened the convolution window. In this window I changed the kernel to 5x5 low pass. My input image was the Chicago TM image and I saved the output image in my personal folder. I accepted all other parameters and ran the tool. After the tool ran, the image didn’t appear much different from the original besides that its quality was not as detailed as the original. Since this image didn’t appear very different I decided to try and improve the brightness quality with a high pass filter on a different image. I started again with a clear image viewer and uploaded the Sierra Leone image from our lab folder. I used the same tools, but instead of using a low pass convolution kernel I selected the 5x5 high pass convolution kernel. The below images are the original image (left) and the high pass convolution (right). 

After overlooking the image, the next step was to perform an edge enhancement technique. I again brought in the original Sierra Leone image from our lab folder. I accessed the convolution window again and input the Sierra Leone image. Under the kernel selection, I selected the 3x3 Laplacian edge detection kernel. I checked fill under the handle edges parameter and unchecked normalize the kernel. I accepted all other parameters and ran the tool. 

Section 2: Spectral enhancement

In this section I performed two types of linear contrast stretch. I started with a blank image viewer and brought in the Eau Claire 1991. I used the metadata tab to look at the histogram for the image. I decided a minimum-maximum contrast stretch was best for this image. I clicked on general contrast and followed the drop down to the general contrast tool and clicked on it. This opened the contrast adjust interface. Under the method tab I changed it to Gaussian. I then ran the tool and the image appeared.

 I decided to run a piecewise stretch after looking at the Gaussian stretched image. I clicked on general contrast and followed the drop down to piecewise to access the tool. Under the range specification tab I clicked on middle and changed the dynamic range of brightness values for the last mode to 180. I applied the tool to the image and it appeared as this. 

Histogram Equalization

This process improves the contrast of the image to enhance visual interpretation. I opened the original image; this image was the red band of the Landsat TM. I used the raster toolbar and clicked on radiometric and histogram equalization. This process opened the histogram equalization window. The input image was the original image I brought in and the output image was saved in my personal lab folder. I accepted all parameters and ran the tool. 

Part 4: Binary change detection

In this part of the lab I estimated the brightness values of pixels that changed in Eau Claire County and surrounding area from 1991 to 2011.

Section 1: Image differencing

I began this with opening two viewers in Erdas. I uploaded the Eau Claire images from 1991 into one viewer and the Eau Claire image from 2011 into the other viewer. I then clicked on the raster toolbar to activate it, then clicked on the function tab and scrolled down to the two-image functions tool to access the two-input operations window. In the input file one I inserted the 2011 image and put the 1991 image into the second input file. Under the output options portion I changed it from a plus to a minus. Underneath the first image in the input file I clicked on the layer tab and changed it from all to four, and did the same on the second input image. I accepted all parameters and ran the tool. After bringing in both of the images I then observed their histograms. I used the rule of thumb threshold of (mean+1.5 standard deviation) to determine the cutoff points on the histograms. After viewing the histograms and calculating the changes, my histogram came out looking like this.

Section 2: Mapping change pixels in difference image using spatial modeler

In this section I mapped out the changes of Eau Claire County and the surrounding areas from 1991 to 2011. I started with a blank viewer and didn’t upload any image; instead I opened the model maker by opening the toolbox then clicked on the model maker tab. I constructed a simple model with two input raster objects for my 1991 and 2011 images. After bringing the two image files into the model maker, I then subtracted the 2011 image file from the 1991 image file in the function. For my output raster I saved it in my personal lab folder and ran the model. After running the model and bringing the image into my viewer I then opened the histogram. Upon observing it, the next step was to determine were the upper threshold of the histogram was. To do this I used the mean+(3x standard deviation) function. I again opened the model maker and setup another easy function however this time I only had one input raster, which was the image I saved during the last function. In the function raster option I changed it from analysis to conditional. I then clicked on the either, if, or, or function. This function simply meant that all pixels with value above change/ no change threshold value and mask out those that are below the change/ no change threshold value. My change/ no change threshold value was 202.18. After running the model and bringing it into the viewer I realized that it was hard to read because it was such a dark image. I then opened the arcmap interface from our programs. I brought in the NIR image of Eau Claire from 1991 with the four bands and overlaid it with the Eau Claire image from 2011. In the 2011 image I set the no data to no color so that I could see the areas that changed. I changed the view of the images from data to layout and brought in a legend, north arrow, and scale bar. The final part of this lab looked like this.