Signal to noise ratio optimization for video compression bit-rate control
View Patent ↗A method of determining a quantization parameter for video compression. A block characteristic for blocks in a video frame is quantized and then used to obtain a candidate quantization parameter. With the candidate quantization parameter, a number of bits for the blocks in the video frame is summed, producing a candidate bit rate. If the candidate bit rate is greater than a predetermined bit rate threshold, the candidate quantization parameter is set as the quantization parameter.
1. A method of determining a quantization parameter for video compression, the method comprising:
quantizing a block characteristic for blocks in a video frame, producing a quantized block characteristic;
using the quantized block characteristic to obtain a candidate quantization parameter;
using the quantized block characteristic as an index to a look-up table of possible quantization parameter values, wherein producing the look-up table comprises defining a joint probability density function of a quantization parameter and a picture type, applying vector quantization to convert the joint probability density function to a set of code vectors, quantizing the block characteristic, and producing a look-up table with the quantized block characteristic as a table index and the code vectors as the values;
summing a number of bits for the blocks in the video frame producing a candidate bit rate, wherein the quantized block characteristic and the candidate quantization parameter determine the number of bits per block; and
setting the candidate quantization parameter as the quantization parameter for the frame, if the candidate bit rate is greater than a predetermined bit threshold.
2. The method of claim 1 , wherein the method further comprises obtaining a new candidate quantization parameter and performing the summing, if the candidate bit rate is less than the predetermined bit threshold.
3. The method of claim 1 , wherein the method further comprises the quantized block characteristic being the sum of absolute differences for the block.
4. The method of claim 1 , wherein using the quantized block characteristic to obtain a candidate quantization parameter further comprises using the quantized block characteristic as an index to a look-up table of possible quantization parameter values.
5. The method of claim 1 , wherein the code vectors are produced from estimates of a conditional mean value of the quantization parameter.
6. The method of claim 1 , wherein the method further comprises:
perturbing the quantization parameter to obtain a predetermined maximum change, producing a perturbed quantization parameter;
applying vector quantization to obtain a pixel signal-to-noise ratio to quantization parameter relationship for each block of the video frame for the perturbed quantization parameter;
calculating a signal-to-noise efficiency for each value of the perturbed quantization parameter;
selecting the perturbed quantization parameter based upon the signal-to-noise efficiency and the predetermined bit threshold.
7. A video encoder, comprising:
a port to receive video signals;
a memory to store look-up tables of quantized block characteristics indexed to quantization parameter values, wherein creation of the look-up tables comprises defining a joint probability density function of a quantization parameter and a picture type, applying vector quantization to convert the joint probability density function to a set of code vectors, quantizing the block characteristic, and producing a look-up table with the quantized block characteristic as a table index and the code vectors as the values; and
a processor to:
use the quantized block characteristic to obtain a candidate quantization parameter;
sum a number of bits for the blocks in the video frame producing a candidate bit rate, wherein the quantized block characteristic and the candidate quantization parameter determine the number of bits per block; and
set the candidate quantization parameter as the quantization parameter for the frame, if the candidate bit rate is greater than a predetermined bit threshold.
8. The video encoder of claim 7 , wherein the processor is also operable to obtain a new candidate quantization parameter if the candidate bit rate is less than the predetermined bit threshold.
9. The video encoder of claim 7 , wherein the processor is also operable to:
perturb the quantization parameter to obtain a predetermined maximum change, producing a perturbed quantization parameter;
apply vector quantization to obtain a pixel signal-to-noise ratio to quantization parameter relationship for each block of the video frame for the perturbed quantization parameter;
calculate a signal-to-noise efficiency for each value of the perturbed quantization parameter;
select quantization parameters for each block based upon the signal-to-noise efficiency and the predetermined bit threshold.
10. An article containing machine-readable code that, when executed, cause a machine to:
quantize a block characteristic for blocks in a video frame, producing a quantized block characteristic;
use the quantized block characteristic to obtain a candidate quantization parameter;
sum a number of bits for the blocks in the video frame producing a candidate bit rate, wherein the quantized block characteristic and the candidate quantization parameter determine the number of bits per block;
set the candidate quantization parameter as the quantization parameter for the frame, if the candidate bit rate is greater than a predetermined bit threshold;
perturb the quantization parameter to obtain a predetermined maximum change, producing a perturbed quantization parameter;
apply vector quantization to obtain a pixel signal-to-noise ratio to quantization parameter relationship for each block of the video frame for the perturbed quantization parameter;
calculate a signal to noise efficiency for each value of the perturbed quantization parameter; and
select the perturbed quantization parameter based upon the signal-to-noise efficiency and the predetermined bit threshold.
11. The article of claim 10 , wherein the article contains further code that, when executed, causes the machine to obtain a new candidate quantization parameter and perform the summing, if the candidate bit rate is less than the predetermined bit threshold.
12. The article of claim 10 , wherein the code causing the machine to use the quantized block characteristic to obtain a candidate quantization parameter further comprises code that causes the machine to use the quantized block characteristic as an index to a look-up table of possible quantization parameter values.