Browse Prior Art Database

Method of Rate Control for Multi-Pass MPEG Variable Bit Rate Coding

IP.com Disclosure Number: IPCOM000014939D
Original Publication Date: 2002-Jan-29
Included in the Prior Art Database: 2003-Jun-20
Document File: 3 page(s) / 86K

Publishing Venue

IBM

Abstract

Method of Rate Control for Multi-Pass MPEG Variable Bit Rate Coding Ligang Lu, Cesar A. Gonzales, and Jack Kouloheris What is disclosed is a new rate control algorithm for MPEG multi-pass VBR coding. The primary purpose of the first coding pass in a multiple pass coding scheme is to obtain statistical information of the source so that appropriate coding parameters can be chosen for better picture quality in the later passes. In our method, we gather statistical data for each picture P(k) in the first pass. The two most important statistics collected are the bit count B(1,k) and the average quantization scale Q(1,k) which are used to estimate the parameter of the bit production model or the picture complexity C(k). The other statistics, such as the spatial activity and the temporal activity, may also be obtained for scene change and fading detection. In the following we first introduce a more accurate picture coding complexity estimate. 1. A More Accurate Picture Coding Complexity Estimate

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 41% of the total text.

Page 1 of 3

Method of Rate Control for Multi-Pass MPEG Variable Bit Rate Coding

Method of Rate Control for Multi-Pass MPEG Variable Bit Rate Coding

Ligang Lu, Cesar A. Gonzales, and Jack Kouloheris

What is disclosed is a new rate control algorithm for MPEG multi-pass VBR coding. The primary purpose of the first coding pass in a multiple pass coding scheme is to obtain statistical information of the source so that appropriate coding parameters can be chosen for better picture quality in the later passes. In our method, we gather statistical data for each picture P(k) in the first pass. The two most important statistics collected are the bit count B(1,k) and the average quantization scale Q(1,k) which are used to estimate the parameter of the bit production model or the picture complexity C(k). The other statistics, such as the spatial activity and the temporal activity, may also be obtained for scene change and fading detection. In the following we first introduce a more accurate picture coding complexity estimate.

1. A More Accurate Picture Coding Complexity Estimate

In [2], the bit production model for the k-th picture is assumed as B(1,k) = C(k) /Q(1,k). This model has a problem, especially in the low bit rate coding cases. The bit count B(1,k) includes both the bits used to code the overhead, such as the headers and the motion vectors, and the bits used to compress the DCT (Discrete Cosine Transform) coefficients. However, only the amount of bits used to encode the DCT coefficients depends largely upon the picture coding complexity while the bits used to code the overhead is in general independent of the picture coding complexity. Including the overhead bits in the model results in inaccuracy in the estimation of the picture coding complexity. This inaccuracy will become significant when the picture is relative easy to be compressed, especially in the low bit rate coding cases. The model will tend to overestimate the picture coding complexity.

To more accurately estimate the picture coding complexity and describe the bit production model for better rate control and picture quality, we divide B(1,k) into the bits used in encoding the overhead B(1,k,o) and the bits spent on coding the DCT coefficients B(1,k,c). That is B(1,k) = B(1,k,o) + B(1,k,c). In general, B(1,k,c) depends on the picture coding complexity while B(1,k,o) is independent of it. To more accurately estimate the picture coding complexity, especially for the low bit rate coding applications, we propose a new bit production model B(1,k) = B(1,k,o) + C(k)/Q(1,k) and the picture coding complexity C(k) = B(1,k,c) Q(1,k).

2. Statistic Data Processing

Since in the first pass encoding there was no information about the source characteristics, the coding parameters that were used may not be appropriately chosen. Given the information, we might use more pertinent coding parameters. Therefore before we determine the coding parameters for the second pass coding, we process the

1

Page 2 of...