PDF version of this entire document

Further Data Preparation

In addition, due to cases of mis-detection of the nose (about 10% of the time in the case of GIP datasets with 3 examples shown in Figure ), improvements were made to the cropping algorithm, which now has 4 methods implemented. The latest one yields something which best resembles the IJCV paper. This ought to help future implementations involving not just PCA, which is largely implemented but requires good data to operate on.

Images are shown in Figure , which demonstrates what cropping tends to look like.

Figure: Examples of faces that, given the default set of parameters, do not detect the nose correctly (with old-style cropping)

Figure: Improved cropping of faces that takes spatial measurements into account