Dense Image Space

Example DIS set, Mark's bookshelf 1

    This is documentation and notes for the creation and display of a particular set of photographic images for the purpose of demonstration and illustration of the DIS concept.

Photographic details and geometry

    The example photographs were acquired using a Canon Rebel XT using a 50 mm lens (Leitz Wetzlar Summilux-R 1:1.4/50, with a Leica/R for EOS Canon adapter), at F/8, 1/5 sec, with ISA = 800 (I forgot to reset this to a lower sensitivity, like gazillions of other people). Lighting was all diffuse daylight from a nearby glass door on a clear afternoon, but with little direct sunlight, inside or out.

    The image set consists of a set of 10 photographs taken from 10 viewpoints in a linear horizontal row. The optic axes of all images are roughly parallel. Here are the images, from left to right (in reading order, of course):
Marks_bookshelf_1_y01_x01
Marks_bookshelf_1_y01_x02
Marks_bookshelf_1_y01_x03
Marks_bookshelf_1_y01_x04
Marks_bookshelf_1_y01_x05
Marks_bookshelf_1_y01_x06
Marks_bookshelf_1_y01_x07
Marks_bookshelf_1_y01_x08
Marks_bookshelf_1_y01_x09
Marks_bookshelf_1_y01_x10

All raw images in a zipped file: Marks_bookshelf_1_DIS_raw_images.zip (21 MB)

    The spacing between viewpoints is about 2", and to reduce the perspective distortion and other geometry errors between shots, I used a tripod that I slid along a steel rule taped to the floor (the front two feet of the tripod along the rule edge, and a pencil mark extension). [Sorry I didn't use metric units, as the rule I used happened to be marked in inches.] This is a context photo, with the tripod at the first photo location:
Marks bookshelf 1 context

    This is a plan view, of the photo geometry, approximately to scale.
[[ geometry illustration ]]

    The angular field of view with this camera and lens is about [[θ = ??° = ??rad.]].  In most cases a wider angular field of view is desireable for DIS; the range of angles from which a subject point will viewable with DIS is about this angular field of view. I have a 17-55 mm zoom lens, but the spherical ("barrel") distortion is significant at short focal lengths, and I didn't want to go to the trouble here correcting each image for this distortion. StereoMaker, the software used below for coregistration, does include a good tool for barrel distortion correction. In general, a wider field of view will also require more images/overlap for an equivalent "smooth coverage" of image space -- since you are closer to the subject, the perspective change relative to the camera shift is higher.

    The fractional overlap at the subject distance, Sd (chosen to be at spines of the books), with this geometry (d = 2", Sd ~=  38") is about [?? = ??%]].

    Each native pixel subtends about [.??mm]] at this subject distance.

Mutual image coregistration

    In order to view the images in allignment, they need to be coregisetered. This involves warping each image such that the overlapping regions are projected onto the same imaginary at a fixed subject distance (in this case, near the book spines), and finding the relative offsets between each image at the subject distance. I took some care to avoid acquisition position and and rotation errors, and approximately alligned the optic axes using the steel rule and tripod on a flat floor. But there were still significant error involved, and I was interested in how accurate this method was. Ideally I'd like to be able to "eyeball" the aquisition of  image sets like this, in which case automated image coregistration will definitely be necessary.
    StereoView could be used to provide this coregistration information, but I chose to use the StereoMaker tools. Here's a brief description of the steps I used.

[[ document coregistration procedure ]]
[[ coregistration parameters and relative position text ]]
All coregistered images in zipped file: Marks_bookshelf_1_DIS_corrected_images.zip (19 MB)

DIS demo

    See Dense Image Space (DIS) description for a conceptual description of how these images can be arranged such that they can be viewed in the context of a DIS. Here I describe a slow and dirty method for visualizing a coregistered DIS set using existing software (Space) that was not designed for this task. This is not an easy or slick demonstration. It requires following obscure instructions and imagining how it might look and feel with software built for the task. Panning is currently not available, although different parts of the image can be examined using the "windowing" functionality. Update times are painfully slow (particularly at magnifications < 1), as multiple large overlays are all resampled and rendered independently; in the final software, fast "blitting" of only the relevant "top" image will speed things up dramatically. 500 MB RAM minimum is required for the full resolution demo. I might make a smaller version (resampled by x 1/2) available.

[[ description of how coregistered images are loaded and converted to coregistered images ("volumes") in Space ]]


All images as coregistered volumes (Space native format) in zipped file: Marks_bookshelf_1_DIS_corrected_images.zip (200 MB. Sorry this is so large; images don't compress well with .gz/.zip, but they need coregistration info that can't be stored with TIFF/JPEG formats. Ideally a GIS format would be used that would integrate the two.) To view them in Space software, unzip, drag all volumes onto a Space window, and use the overlay viewer to select which image is "on top". [Warning: Display update is painfully slow for purely technical, and correctable, reasons.]

[[ description of DIS navigation and navigation limitations in Space ]]

Demo animations, from screen captures

[[ description of next steps for improving demo and streamlining the whole process ]]