Scribd
Upload
36 views

The Video Compression Problem: Products For The Broadcast Industry

The document discusses video compression technologies. It summarizes that conventional methods like motion estimation and wavelets have limitations, while Futureware's approach uses a human visual model to statistically guarantee removal of all non-visible information, achieving the highest compression ratios with minimal artifacts. It describes how their technology uses lossless compression through entropy encoding of decorrelated video data, and visually lossless quantization by defining quantizers that result in no visual loss under given viewing conditions. Their system continuously monitors the subjective quality of compressed video by comparing it to the reference and signaling errors above a threshold in "visibels".

Uploaded by

nagarjunagolamari
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
36 views

The Video Compression Problem: Products For The Broadcast Industry

The document discusses video compression technologies. It summarizes that conventional methods like motion estimation and wavelets have limitations, while Futureware's approach uses a human visual model to statistically guarantee removal of all non-visible information, achieving the highest compression ratios with minimal artifacts. It describes how their technology uses lossless compression through entropy encoding of decorrelated video data, and visually lossless quantization by defining quantizers that result in no visual loss under given viewing conditions. Their system continuously monitors the subjective quality of compressed video by comparing it to the reference and signaling errors above a threshold in "visibels".

Uploaded by

nagarjunagolamari
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

The Video Compression Problem

Products for the Broadcast Industry

HD, SD Format Conversion


Subjective Quality Monitor
Time Compression
Video Compression

Overview

Conventional compression technology is based on one of three technologies:

• Motion Estimation. This technology reduces the redundancy between frames of


motion video by representing a match of an object between the frames as a vector
to the object. This technology is limited in the compression ratios it may attain
without degrading picture quality, and is computationally prohibitively expensive.

• Function matching (fractals). This technique attempts to describe a region of


video as a mathematical formula. This technique has also been unsuccessful in
obtaining large compression ratios without visual artifact.

• Subband (wavelet). This technique is limited in its practical utility because


quantization in subbands have side effects in other bands.

Futureware's ZPEGTM compression technology is based upon its VisibelTM Human Visual
Model. This model statistically guarantees that all information not visible to the human
eye is removed, or alternatively, that the highest compression ratios are reached with the
minimum of visible artifact.
Lossless Compression

The value of a given pixel in typical video data (figure 1) may be used to predict the
value of the next pixel with a probability of .96 and above (over .99 in the time
dimension for SD video). This property is known as correlation between pixels, and is
modeled with the Markov probability model.

Figure 1.PCM (Pixel) Representation of Video Sequence.

The optimal representation of video data, in the sense of maximal concentration of energy
into the minimal number of coefficients, is the representation in decorrelated K-L space
(figure 2). The statistics for the video sequence are used determine the properties of this
transform, which is invertible and can be undone when received.

Figure 2.Decorrelated (K-L) Representation of Video Sequence.

After the maximal length of the sequence has been determined (through a technique
invented by Futureware), entropy encoding of this representation will give lossless
compression ratios of 10:1.
Quantization

Quantization is the process of introduction of error into a representation of a video


sequence for the purpose of increasing compression, hopefully without visible artifacts. A
transform representation of video data may be independently quantizable (figure 3, DCT
or similar transform) or retain some correlation between the representational basis (all
other transforms, wavelet in this case).

Decorrelated Basis Correlated Basis


(DCT) (Wavelet))

Reconstruction after
Quantization

Figure 3. Quantization in decorrelated bases give similar reconstructions;


those in correlated spaces show side-effects in the reconstruction.

The advantage of quantization in a decorrelated basis is that quantization of each basis


element can be done without affecting other basis elements. An example of how
important this property is can be seen in the recent unsuccessful efforts to harness the
wavelet transform as a compression transform - the unpredictable effects of quantization
made the technique unworkable.
Visually Lossless Quantization

Once an independently quantizable transform has been defined, the optimal visually
lossless quantizer for each basis element may be define. This quantizer is a function of
screen resolution, temporal resolution, and viewing distance.

The set of (largest) quantizer values which under the defined viewing conditions result in
no visual loss is defined as 0 VisibelsTM, Vb) Every factor of two applied to each of the
quantizers (i.e., each doubling of the viewing distance) is defined to add 6 to the Visibel
value, so that the relationship between decibels and visibels is constant.

Visually lossless compression is obtained by quantizing video data at the 0 Vb level; and
minimum-artifact error is obtained at higher compression rates by specifying a higher
quantization level.
broadcast

reference compressor decompressor

Subjective Quality Monitor

Video format + Viewing


Distance in Vbs Alarm

Figure 4. The Subjective Quality Monitor compares compressed and reference


data streams and reports artifacts above a given Vb. threshold.

Subjective quality of a compressed data stream may now be measured against a reference
stream (its source, figure 4). The reference and decompressed streams are continuously
compared against one another, generating the error between them in Vbs. If the error
exceeds a predefined threshold, a warning is signaled.

Futureware, Inc.
475 Wall Street
Princeton, NJ 08540
Phone (609) 924-4269
www.futureware.com
[email protected]

You might also like