Surface Change Detection From Mars Orbital Imagery

Recently identified surface changes on Mars have provided evidence of water present on the surface within the past decade. These and previously studied changes such as the appearance of new dark slope streaks and dust devil tracks have been identified through the manual process of visually comparing two images taken of a region at different times. This tedious and time-consuming process does not scale to the vast amount of historical Mars data currently available from Viking, Mars Global Surveyor, Mars Odyssey, and Mars Express, and continuing to be collected by Mars Odyssey, Mars Express, and the Mars Reconnaissance Orbiter. Automated change detection methods have yet to be effectively applied on a wide scale to robustly identify significant changes on Mars. One important challenge is how to reliably remove changes in appearance that do not represent actual surface alteration, such as changes in viewing geometry, lighting, and atmospheric effects. We present work on the development of methods to provide rapid identification of surface changes on Mars from orbital imagery. We have focused on methods and image matching techniques which, for the most part, do not require the computation of a pixel-based correspondence between images. Our approach is to identify landmarks and changes based on descriptive statistical models of surface regions. We show examples from two methods. The first method uses an intensity histogram to describe a region. Interesting features are identified using the KL-divergence between regions. We applied this method to 12 MOC (Mars Orbiter Camera) image pairs, in which the later image is known to contain evidence of one or more new dark slope streaks. The dark slope streaks, as well as a variety of other surface features, were identified as landmarks. The second method employs compact covariance descriptors to describe regions. It has been shown to effectively identify regions of surface change while being relatively robust to viewing conditions. We applied this method to 5 MOC images in which new gullies have previously been discovered. In each case, the gully was quickly detected, providing a demonstration that automated change detection algorithms could in principle make similar discoveries, while covering many thousands of images in the time it would take a human to examine only a few. Using automated change detection methods, we can identify surface changes that may otherwise be overlooked in the large Mars data sets that have been archived and are continuing to be collected.