In a seminar about "Numerical Methods in Computational Linguistics and Machine Learning", I had to summarize and present two papers on that topic, namely

• Heisele et al: 'Face Recognition With Support Vector Machines: Global versus Component-Based Approach', Center for Biological and Computational Learning Cambridge, 2001

• Romdhani et al:  'Computationally Efficient Face Detection', Microsoft Research Ltd Cambridge, 2001

At first, I didn't think that I could make it in this seminar because it really contained a lot of math (we also wrote an exam), but in the end I didn't only learn a lot, I also got my credits for it.
Anyway, both papers are really interesting. They propose methods to be more efficient and accurate with face detection (which is far from being perfect nowadays). A good lesson why math is important. (Support Vector Machines are not explained here). So take away:

the summary I wrote in order to get my credits (5 pages, pdf)
the presentation I gave (pdf, contains more cool pictures)