Here's the panorama created from the test sequence. Everything works as expected.

"Test Sequence" by Justin McManus, CSE 455 Winter 2012 (full size, 360° viewer)

Not a lot went right with this panorama. I wasn't able to get good feature matching due to lighting differences in a couple of the images. You can see that some sections of the panoarama are clearly much darker than the rest. We forgot to put the camera in manual mode.

"The Quad" by Justin McManus, CSE 455 Winter 2012 (full size, 360° viewer)

This sequence turned out much better than the Kaiden head sequence because the camera doesn't perform any auto-adjustments when in panorama mode

"Some Random Place Near the Quad" by Justin McManus, CSE 455 Winter 2012 (full size, 360° viewer)

Here are the results of applying the Harris operator to two different images. Brighter pixels indicate a stronger response to the Harris corner detection operator.

The Harris image generated from the second graf image.

The Harris image generated from the second yosemite image.

For an ROC curve, the area under the curve shows the accuracy of the test. The next two images show ROC curves for differenct combinations of feature detection and matching algorithms.

ROC comparison for Yosemite - MOPS+SSD (red), MOPS+Ratio Test (green), SIFT+SSD (blue), SIFT+Ratio Test (purple)

SIFT performed exceptionally well on the Yosemite images, to the point where the difference in matching algorithms was almost negligible.

ROC comparison for graf - MOPS+SSD (red), MOPS+Ratio Test (green), SIFT+SSD (blue), SIFT+Ratio Test (purple)

The relative accuracies of detection/matching combinations is the same as with the Yosemite images. SIFT outperforms MOPS, and given a choice of detection algorithm, Ratio Test outperforms SSD for matching features. Accuracies are all much lower for this image set. This is likely due to the significant rotation in the second image.