This video shows the result of a genetic evolution to find the best encoding profile for multiple types of content. Each frame of this video is composed by 4 frames of each one of the 522 videos encoded with different compression parameters that lead to progressive improvement during the evolution. Soccer content were one of the hardest we had to encode. The combination of the noisy grass engulfing the small players and even smaller ball, along with all the full-scene movement brought the necessity of a close-to-optimum encoding profile. To test all the combinations of parameters were unfersible -- as there were so many parameters sometimes with numbers ranging from millions to fractions -- so I used genetic algorithms to improve the quality and lightness of compression, those were tested with several sample videos of every content type we needed to produce (our production run around thousand clips per day of soap operas, news and sports).

The GREEN graph you see beside each frame is the signal-noise ratio of this frame using one of the profiles evolved on the sample video[1]. [note as at the end of the evolution we can almost read the name of the goal keeper on the upper-right frame]. All the YELLOW graphs are the same: the timerate[2] of all the profiles found through the evolution with this sample video[1], one of the main variables being optimized. [note how, even when quality stabilizes, the time to encode keep dropping, this were extremely important for productions of the volume we have. also note that it's not hard to improve quality using extreme processing time]. The images on the left (00000001 and 00000671) are KEYFRAMES, those on the right (00000176 and 00000902) are FRAMES in between. [note the balance of quality between left and right as one of the essential tuning for reaching and ideal profile]. Also note that it's a high definition (720p) video to better display the compression artifacts, all the output frames are in full size (480x360) and I recommend to see this video on fullscreen, unscaled, if possible.

[1] Source sample: 30 seconds segment of a high definition (720p) soccer game from FIFA World Cup 2006, copyright: ESPN (under fair-use, I hope :)

[2] time that the process encoding took divided by the total duration of the output video. timerate of 6 means thet the video took six times the duration of the content to encode (e.g., a 10 seconds video would take one minute of processing time)

NOTE: this video were a sample evaluated through the evolution process, so it should be one of the best quality encoded with this profile.

PS: search for GeneCoding to find other videos on the subject.