Kaiousama
26th September 2002, 15:05
Maybe those are stupid considerations, but i'll try to explain myself:
There is a way to implement in the codec an eye focus detection system that helps to gain compressibility not where is much movement, but only where the movement isn't in the screen portion seen with attetion by the eye? (my considerations are based and applicable first to anime, maybe their'll be useful even with normal films)
I'll try to explain myself better, immagine two situations:
First a cloudy landscape where is a slow clouds' movement (a storm) as background and over that there is an airplane that has a banner attached with some words and flyes faster than the clouds' movement; so, if i've well understood the actual codec take in considerations (exclude the luma situation for now) the movement in the scene, so the clouds gets more bits than the banner because their movemement is slower, but the attention of the watcher is all on the plane because the 90% of the scene information (i mean the conceptual information not the bits) is flying with the plane.
The second example is a scene where one person, in the centre of the frame speaks (only the face is on the big scene and takes a big amount of the frame)moving his face. after this person there are two children playing in a little part of the screen, the attention of the watcher goes mostly in the background to watch the playing children, and the person in the centre is considered as a redoundant information of the scene ('cause in this case the sound brings all the information of this person).
A right question is..... "so, ok, but how can i implement an algorythm that imagine what are the human tastes in watching a frame?"
An idea is to give more bitrate to those part of the image that has small-dimensions&fast-moving and smaller bitrate to the big-dimension&slow-movements things
Another concept, strictly relative to anime is to identify the edges of what is moving (in a way close to what actually make the Msmooth Donald graft's filter for example), analize the movement of each thing separately and give bitrate to the parts that are changing often its movement speed coefficient (what changes speed takes easily the attention of the watcher)
A little questions on the chroma versant (that if i've well understand a change in the chroma values changes only a little the compressibility gainability) The actual codec it's weak in compressing the transient zone between two full colours (i mean a pot that is yellow on the base and red in the top with gradual gradient in the centre between this two colours) there is a way to recognise this gradient part of the frame and give to him more bitrate?
a question i've never really understand: why the codec is weak in compressing a dark red, or blue, org green moving scene instead a more complex multicolour scene (the classical macroblocks in red scenes)?
Tnx for your spent time, and excuse my italian ^_^
There is a way to implement in the codec an eye focus detection system that helps to gain compressibility not where is much movement, but only where the movement isn't in the screen portion seen with attetion by the eye? (my considerations are based and applicable first to anime, maybe their'll be useful even with normal films)
I'll try to explain myself better, immagine two situations:
First a cloudy landscape where is a slow clouds' movement (a storm) as background and over that there is an airplane that has a banner attached with some words and flyes faster than the clouds' movement; so, if i've well understood the actual codec take in considerations (exclude the luma situation for now) the movement in the scene, so the clouds gets more bits than the banner because their movemement is slower, but the attention of the watcher is all on the plane because the 90% of the scene information (i mean the conceptual information not the bits) is flying with the plane.
The second example is a scene where one person, in the centre of the frame speaks (only the face is on the big scene and takes a big amount of the frame)moving his face. after this person there are two children playing in a little part of the screen, the attention of the watcher goes mostly in the background to watch the playing children, and the person in the centre is considered as a redoundant information of the scene ('cause in this case the sound brings all the information of this person).
A right question is..... "so, ok, but how can i implement an algorythm that imagine what are the human tastes in watching a frame?"
An idea is to give more bitrate to those part of the image that has small-dimensions&fast-moving and smaller bitrate to the big-dimension&slow-movements things
Another concept, strictly relative to anime is to identify the edges of what is moving (in a way close to what actually make the Msmooth Donald graft's filter for example), analize the movement of each thing separately and give bitrate to the parts that are changing often its movement speed coefficient (what changes speed takes easily the attention of the watcher)
A little questions on the chroma versant (that if i've well understand a change in the chroma values changes only a little the compressibility gainability) The actual codec it's weak in compressing the transient zone between two full colours (i mean a pot that is yellow on the base and red in the top with gradual gradient in the centre between this two colours) there is a way to recognise this gradient part of the frame and give to him more bitrate?
a question i've never really understand: why the codec is weak in compressing a dark red, or blue, org green moving scene instead a more complex multicolour scene (the classical macroblocks in red scenes)?
Tnx for your spent time, and excuse my italian ^_^