PDF version of this entire document

Residue Adjustments

The set is reasonably well handled following the removal of two images from almost 200 images in total, leaving a group of 86 distinct people. Shown in Figure [*] and Figure [*] are two masks that were tested on the group. Shown are just the first few images, not a bunch of cherry-picked examples. Among the 4 binary masks that are used (with different thresholds for depth as well), the latter works better and it is aided by cropping that more or less normalises the region under consideration, making it easier to sample and subsequently compare. Worth noting are the difference around the mouth, which reveal some teeth.

By profiling people's expression residues and then testing to see if these profiles - be they based on a model or not - can be used to detect the identity of the person, we can reason about the approach and compare pertinent, exchangeable parts, swapping them and assessing the effect on overall performance. First, something more basic like sum-of-squared-differences will be tested as a differentiator (for match/target).

Figure: The first 6 images in the set with a narrow mask used to extract and attain a neutral-to-non-neutral residue
Image residues-capture-smile

Figure: Same as the previous figure, but with only 5 images. The top row shows the effect of using a broader mask and the bottom part shows the effect of applying a fixed mask and thresholds to make the data more trivially comparable.
Image residues-capture-smile-mark-aspect-ratio-and-binary

Roy Schestowitz 2012-01-08