Presentation of my paper "Optimization of Video Compression Parameters through Genetic Algorithms", published on the WebMedia Symposium at 2008, Vila Velha, Espirito Santo, Brazil ... so it is in Brazilian Portuguese. But the paper is here in english: slideshare.net/zed9h/optimization-of-video-compression-parameters-through-genetic-algorithms-2008

This video were uploaded to vimeo with a lower resolution and framerate to make it smaller, but the final, quality-comparison videos were also uploaded separately at:
Codec Comparison: vimeo.com/2119639
Frame Evolution: vimeo.com/2120250
Frame Evolution Graph: vimeo.com/2120356

And the slides: slideshare.net/zed9h/presentation-of-the-paper-optimization-of-video-compression-parameters-through-genetic-algorithms-on-the-webmedia-symposium-2008

I have made two mistakes on this presentation:

a) On the last video I referred to the PSNR' of the profiles as the CYAN marks, but in fact the PSNR' are the MAGENTA marks and the CYAN are the profile's timerates;

b) Unfortunately I turned off the audio before the questions. There were one question remarking that PSNR were a very limited method to assert image quality. The example given were: if there is an image with rocks and a blue sky, noise on the rocks is not so bad as noise on the sky. Then I remarked that PSNR were a free comparison, since the compression tool returned it on the process and if I wanted a more sophisticated comparison technique I could use SSIM (Structural Similarity). But, anyway, PSNR only caused me problem when I were comparing more than one transformation at once, like the blurriness of scaling the image and and the blocking of compressing it. But if I kept my testing to were it were blocking more or less, the PSNR were quite close to image quality and were able to bring great results.

PS: the audio should be normalized, because I was walking back and forth to the projected screen... but here comes laziness again.