Bonfire Redux ha detto:
Non mi è ben chiaro il rapporto tra i frame interpolati e la frequenza a 96hz...se i fotogrammi vengono moltiplicati x 4, non ci dovrebbe essere interpolazione. A meno che a partire da due fotogrammi se ne ottenga prima un terzo e poi si raddoppi tutti e tre questi frame ("2:3" X 2, per intenderci).
....
Premesso che per capire appieno la gestione complessiva del movimento nei nuovi Philips probabilmente bisognerà attendere qualche articolo tecnicamente esaustivo e possibilmente il più chiaro possibile, nell' attesa penso che ci possa aiutare leggere qualche articoletto sul nuovo post-processore deputato a queste funzioni, sto parlando del PNX5100:
New Generation Chip
As part of its motion accurate picture processing (MAPP), the new generation of chips such as the PNX5100 video post processor performs motion estimation-compensation, picture improvement and picture
scaling operations with 17-bit sub-pixel accuracy.
To handle the required computing load, it features three on-chip media processors plus a dedicated video back-end that performs functions such as sharpness, contrast and color enhancement as well as graphics insertion.
The MAPP's combination of movie judder cancellation, motion sharpness and vivid color management successfully removes visible halo and blur to deliver an enhanced viewing experience - particularly when watching fast-moving sports and action movies. The chip is also equipped with automatic picture control which adjusts the processing parameters used to obtain optimal improvement on every output frame.
The PNX5100 accepts 1,920 x 1,080p@60Hz LVDS input and produces a 1,920 x 1,080p@120Hz LVDS output. As a result, it fits into an LCD TV's video pipe.
Halo Reduction
One artifact that commonly occurs is a so-called "halo" around moving objects (see Fig). These halos occur because most motion estimation techniques only compare two consecutive cine frames in order to establish a motion vector for each pixel block. If the same pixel block can be identified in both frames, the motion vector is relatively easy to calculate. However, if a matching pixel block cannot be identified in both frames due to something moving in the picture that has caused it to be obscured or uncovered, a reliable motion vector cannot be established. All that motion compensation algorithms can then do is to use a motion vector from surrounding pixel blocks in the hope that it is roughly correct. Since this obscuring or uncovering of pixel blocks occurs at the edges of objects as they move relative to one another, which leaves motion compensation algorithms no alternative but to use unreliable vectors, the halo artifacts appear in the TV picture. Many chip manufacturers try to mitigate the effect by simply blurring the picture on affected areas, but that in itself becomes noticeable to viewers.
Matching Pixels
A new motion estimation-compensation technique which significantly reduces halos has been developed. Instead of identifying matching pixels in two consecutive frames, it matches them across three consecutive frames, block matching between the current frame
and both the previous frame (n-1) and the next frame (n+1).
A pixel block that becomes obscured in frame n+1 will still be identifiable in frames n and n-1, which means that its motion vector for frame n is determinable. Similarly, a pixel block that was obscured in frame n-1 but becomes visible in frame n will be identifiable in frames n and n+1, once again allowing its motion vector to be determined. For frame n, it is therefore possible to establish a complete set of motion vectors for all pixel blocks. Applying the process to a 3-frame rolling window allows a complete set of motion vectors to be established for every cine frame.
The performance of the motion compensation algorithms that create the intermediate frames relies heavily on the accuracy and completeness of the motion vectors. However, one also needs to know whether an individual vector was determined by forward or backward matching in order to apply it correctly, so the motion estimator and motion compensator need to be closely linked. It is also necessary to interpolate the motion vectors themselves so that they temporally match the image data at the point in time where the intermediate frame is being added.
Since the technique identifies pixels that are obscured or revealed in successive frames, the motion compensator also knows which pixels to use and which ones not to use in the interpolated frames. This adds significantly to picture detail and picture sharpness at the edges of moving objects.
With the MAPP technology from the NXP can the halo effect eliminate: For the green and blue blocks in the current (current) frame, can match blocks in the previous frame and the next easily. However, only a three-frame Estimator the matching blocks in places find that the woman in the picture is covered (red) and back up (yellow).
HDTV Contraints
The current motion estimation-compensation challenge for HDTV is to push the frame rate up to 100 or 120 frames per second (fps) while also handling up to five times more pixels per frame than there are in a standard-definition (SD) picture. For standard cine material captured at 24fps, it means interpolating four intermediate frames between each cine frame. That in itself significantly increases the amount of computing power and memory needed. Block matching both forwards and backwards then adds further computing power and memory bandwidth requirement.
Motion estimation-compensation is not the only thing one need do in order to get the most out of HDTV. Most TV formats don't utilize the full range of colors that appear in nature, so to produce truly life-like pictures one needs to intelligently map and process the colors of certain features through techniques such as blue sky enhancement, skin tone correction, blue stretch and green enhancement. Standard TV formats also fail to deliver the picture resolution and sharpness that HDTVs can display, so manufacturers need to scale up picture resolution and improve sharpness at the same time. It has to be done on every frame and on resolutions up to 1,080p@120Hz.
Per chi sa il tedesco trovo questi articoli davvero interessanti
sempre relativamente alla nuova tecnologia di NXP PNX5100:
http://www.elektronikpraxis.vogel.de/themen/hardwareentwicklung/displays/articles/104914/
http://www.hdforum.de/index.php?templateid=news&id=173&sess=4be6cf4dbf81ddeef9c84c659de8274f
