Microscopic Imager Toolkit

Image Registration

As the IDD moves the MI camera, the image of the rock or soil in view changes. If the scene is planar, then the changes can be described with a homography (an 8 parameter linear transformation). The image registration finds this transformation. An example is shown below. If the scene is not planar, the image registration finds the closest fit.

Original data (movie)	Registered (movie)

Disparity Optimization

If the object is not planar, then the homography does not describe all of the differences between the two views. In this case, the disparity optimizer can find matching pixels from two images. It finds these matches to an accuracy less than a pixel. These matches can then be used for focal section merging or for recovering 3D models.

Planar alignment (movie)	Disparity optimizer (movie)

Focal Section Merging

Because the microscopic imager has a very narrow depth of field, the object in view can go in and out of focus as the camera moves toward the scene. Since it may also be irregularly shaped, different parts are in focus at different times. Focal section merging finds the parts of different images that are the best focused views of different regions and combines them to produce one best focus image.

3D Models

If the camera moves so that there is enough parallax between the views, then the 3D shape of the surface can be recovered. Using a dense set of matched points from the disparity optimizer, a structure from motion algorithm recovers the camera position as well as an accurate estimate of the location of close to a million different points on the surface. These points are used with one of the views to produce a 3D texture mapped model which can be viewed in the appropriate visualization tools.

View 1	View 2	View 3
3D model (movie)