|
Format Conversion
Our image manipulation techniques construct a statistically-determined
idealized representation of a sequence of moving images in transform space. The
resulting representation is the sampled (constructed) at the required rate in
destination space. This technique creates the frames of the resulting image as
if they had been originally captured at the desired resolution or frame rate.
The resulting video has compelling clarity and smoothness of motion.
Format conversion by means of this idealized three-dimensional transform
supports such features as digital television format conversion and HDTV
upsampling.
Go to Top of Page
Gibbs Error Removal
Gibbs error is introduced at boundaries between highly correlated video
sequences (i.e., fast motion on the screen). In such cases, the boundary between
segments is detected, and interpolation proceeds on either side independently.
Go to Top of Page
Predictive Frame Conversion
Markov-PredictionTM synthesizes perfect interpolated frames.
Go to Top of Page
Line Interpolation
Markov-PredictionTM combined with bi-directional Motion Adaptation
give highest-quality progressive frames.
Go to Top of Page
Aspect Correction
Aspect conversion supports preservation of the aspect ratio, or continuous
stretch or truncation.
Audio Pitch Reshaping
The Interpolating Pitch-Invariant DCT is used to reshape audio to the new
sampling rate for time compression.
Go to Top of Page
Lossless Compression
Long run lengths with high pixel correlations lead to better compression. In
this example, examination three different run lengths leads to the optimal
representation.
| Run Lengths |
Compression Ratio |
| Three four-pixel runs |
68 bits (13:1) |
| Twelve pixel run |
52 bits (17:1) |
| Eight pixel run, followed by four pixel run |
43 bits (20:1) |
Go to Top of Page
Correlation-Based Partitioning
Sequences of fields are subdivided into maximally-sized three-dimensional
regions for entropy coding.
Go to Top of Page
Lossy Compression
HVS-based quantization gives higher compression ratios as less information
(area under curve) is transmitted. HVS-based quantization is visually
imperceptable.
Quantization in other transform spaces leads to aliasing, as shown in this
example of quantization in a Wavelet basis.
However, quantization in the DCT basis does not lead to aliasing, and choice
of HVS-based quantizers leads to pleasing, visually identical results.
Pixel correlations are highest in the time direction, motivating the ZPEGTM
three-dimensional compression algorithm.
Go to Top of Page
Table of ATSC Formats
|
# Lines
|
#
Pixels
|
Aspect
Ratio
|
Frame
Rate
|
Progressive
|
|
1080
|
1920
|
1,3
|
1,2,3,4,5
|
1
|
|
3,4,5
|
0
|
|
720
|
1280
|
1,3
|
1,2,3,4,5,6,7,8
|
1
|
|
480
|
704
|
2,3
|
1,2,3,4,5,6,7,8
|
1
|
|
3,4,5
|
0
|
|
640
|
1,2
|
1,2,3,4,5,6,7,8
|
1
|
|
3,4,5
|
0
|
|
List of government-approved (ATSC-listed)
digital video formats.
|
Aspect Ratio Formats:
- 1 Square Pixel
- 2 4:3 Aspect Ratio
- 3 16:9 Aspect Ratio
Frame Rates:
- 1 23.976 Hz
- 2 24 Hz
- 3 25 Hz
- 4 29.97 Hz
- 5 30 Hz
- 6 50 Hz
- 7 59.94 Hz
- 8 60 Hz
Progressive Formats:
- 0 Interlaced
- 1 Progressive
|
