Picture Quality Indicator
Original Publication Date: 1998-Jun-01
Included in the Prior Art Database: 2005-Apr-04
Greenfield, JD: AUTHOR [+4]
Disclosed is an efficient way of indicating Picture Quality on a picture by picture basis within an Encoder using a minimum amount of hardware and microcode. The disclosed method requires much less calculation and circuitry than the equivalent SNR measurement.
Picture Quality Indicator
Disclosed is an
efficient way of indicating Picture Quality
on a picture by picture basis within an Encoder using a minimum
amount of hardware and microcode. The disclosed method requires
much less calculation and circuitry than the equivalent SNR
MPEG-2 is a
lossy compression technique in which the
regenerated image differs from the original image. The differences
may or may not be visually noticeable. In order to measure these a
method called a Signal-to-Noise Ratio (SNR) may be used. SNR is
defined as the magnitude of the signal divided by the amount of
unwanted noise that is interfering with the signal. SNR is usually
described in decibels "db". Generally, a larger SNR results in a
higher quality picture. The SNR will vary from picture to picture
depending on picture content, complexity and other encoding
equation for Signal to Noise ratio is: SEE
o MSE = Mean Squared Error
o M = Horizontal image size
o N = Vertical image size
o 255 = Peak to peak value of the original image
Given this, SNR
is an ideal way to measure picture
quality. It requires, however, an exact floating point calculation
to produce an accurate signal to noise ratio number. In many encoder
designs a true floating point calculation cannot be performed so
another technique of measuring or indicating picture quality must be
luminance and chrominance macroblock data is
processed and sent to hardware unit 'A'. This data represents the
original image. Reference luminance and chrominance macroblock data
is calculated and also sent to hardware unit 'A'. IDCT (Inverse
Discrete Cosine Transformer) data is also received and is used to
help calculate reconstructed data out of this unit. See Fig. 3.
'A' uses the current macroblock data and the
Reference macroblock data along with the IDCT data to calculate the
absolute value of the sum of the differences of the Current
macroblock minus the reconstructed macroblock. See Fig. 4.
In an intra
macroblock no reference data is calculated
so the reconstructed data in this case is simply the IDCT result.
See Fig. 5.
o Reconstructed = 00 () IDCT
In an non-intra
macroblock, reference data is used so the
reconstructed data has the reference data factored in the result.
o Reconstructed = Reference () IDCT
adds up all the macroblock differences and
then divides by the numbe...