JP2005295064A - Compression encoding method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To match second and subsequent code generation amount to a target code amount with high precision in multipath compression encoding processing of two path or more. <P>SOLUTION: The compression encoding method comprises a step (a) for generating an encoded stream by carrying out orthogonal conversion, quantization and entropy encoding of a motion picture signal, a step (b) for setting a target code amount for each unit section based on the results of encoding from step (a), and a step (c) for determining a quantization width corresponding to the target code amount with reference to a code generation amount prediction graph for the quantization width and executing processing of step (a). <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a high-efficiency encoding technique for compressing and encoding a moving image signal.

  In recent years, as a next-generation high-efficiency video coding standard, ITU-T (International Telecommunication Union-Telecommunication Sector) VCEG (Video Coding Experts Group) and ISO / IEC (International Organization for Standardization / International Electrotechnical Commission) MPEG ( “H.264” (ISO / IEC14496-10), which was jointly developed by two standardization organizations called “Moving Picture Experts Group”, is drawing attention.

Also, as a moving picture signal encoding technique, variable bit rate encoding for variably controlling the bit rate of an encoded stream and fixed bit rate encoding for controlling the bit rate constant are known. The variable bit rate encoding method of the 2-pass method determines a target code amount for each unit section based on the first-time (first-pass) encoding result, and at the second (second-pass) compression encoding, In this method, the generated code amount for each unit section is controlled to approach the target code amount by changing the quantization width according to the target code amount. A two-pass variable bit rate encoding technique is disclosed in, for example, Japanese Patent Application Laid-Open No. 11-136673.
JP 11-136673 A

  However, the conventional two-pass compression encoding method has a problem that the difference between the generated code amount by the second encoding process and the target code amount determined by the first encoding process tends to increase. In particular, H. The two-pass compression encoding method based on H.264 requires a larger amount of computation than the conventional compression encoding method based on the MPEG method, so that it is difficult to predict the quantization width with respect to the target code amount, and the above-described problems are likely to occur.

  In view of the above situation and the like, the main object of the present invention is to provide a compression code that makes it possible to accurately match the generated code amount for the second and subsequent times to the target code amount in a multi-pass compression encoding process of two or more passes. Is to provide a method.

  In order to achieve the above object, the invention according to claim 1 is a compression encoding method of a multi-pass scheme of two or more passes, wherein (a) a moving image signal is orthogonally transformed and quantized and entropy-coded to obtain an encoded stream. (B) setting a target code amount for each unit section based on the encoding result obtained in the step (a), and (c) predicting the generated code amount for the quantization width And a step of determining a quantization width corresponding to the target code amount with reference to a prediction graph and executing the processing of step (a).

  Embodiments according to the present invention will be described below with reference to the drawings.

  FIG. 1 is a block diagram schematically showing a configuration of an encoder 1 conforming to H.264. The encoder 1 includes a subtracter 10, a conversion unit 11, a quantization unit 12, an entropy encoding unit 13, an inverse quantization unit 14, an inverse conversion unit 15, an adder 16, a loop filter 17, a frame memory 18, and an intra-frame. A prediction unit 19, a motion compensation unit 20, a motion prediction unit 21, a switch unit 22, and an encoding control unit 23 are provided. Before describing the multipass compression encoding method, the operation of the encoder 1 will be outlined first.

  The moving image signal is composed of a plurality of temporal frame images and is input to the encoder 1. The subtracter 10 outputs the difference between the input frame video and the frame video transmitted from the switch unit 22 to the conversion unit 11. The switch unit 22 connects the intra-frame prediction unit 19 and the subtractor 10 when performing intra-frame prediction encoding for an I picture (Intra-coded picture) under the control of the encoding control unit 23. When inter-frame predictive coding is performed on a picture (Bidirectionally predictive-coded picture) and a P picture (Predictive-coded picture), the motion compensation unit 20 and the subtractor 10 are connected.

  The conversion unit 11 performs integer conversion on the input signal in units of 4 × 4 pixels, and outputs the conversion coefficient to the quantization unit 12. The quantization unit 12 quantizes the input transform coefficient with the quantization width (quantization step width) Q supplied from the encoding control unit 23, and the resulting quantization coefficient is inverted from that of the entropy encoding unit 13. It outputs to the quantization part 14. Here, the larger the quantization width Q, the smaller the generated coding amount, and the smaller the quantization width Q, the larger the generated coding amount. The motion prediction unit 21 generates a motion vector with reference to 2 to 5 frame images input to the encoder 1 and supplies the motion vector to the entropy encoding unit 13. The entropy encoding unit 13 A coded bit stream is generated by performing variable length coding on the quantized coefficient based on the motion vector.

  The inverse quantization unit 14 inversely quantizes the quantization coefficient input from the quantization unit 12, and the inverse transform unit 15 decodes the transform coefficient input from the inverse quantization unit 14 into a frame image. . This frame image is added to the frame image transmitted from the switch unit 22 and is stored in the frame memory 18 after block noise is reduced by the loop filter 17. The frame memory 18 supplies the delayed frame image to the intra-frame prediction unit 19 and the motion compensation unit 20.

  Next, a multipass encoding method using the encoder 1 will be described. As shown in FIG. 2, in the first encoding process (first pass), the optimal encoding prediction unit 30 supplies a control signal including a quantization width Q and the like to the encoder 1. The encoder 1 generates an encoded stream by compressing and encoding a moving image signal under the control of the optimal encoding prediction unit 30. At this time, the optimal encoding prediction unit 30 designates the quantization width Q for each frame video or each GOP (Group of picture) based on the initial information.

  The optimum encoding prediction unit 30 acquires a feature amount from the first encoding result. As the feature amount, for example, the type of intra-frame encoding or inter-frame encoding for each frame image, the generated code amount for each frame image, the original moving image signal before encoding, and decoding after encoding this The error with the converted video signal, the number of pixel blocks used for intra-frame coding, and the like can be mentioned.

  Next, in the second encoding process (second pass), as shown in FIG. 3, the feature amount analysis unit 31 analyzes the feature amount acquired in the first time and converts the input moving image signal into a scene. Divide by unit. Here, the scene means a group of frame images between scene changes where the correlation between the frame images becomes extremely small. In addition, the feature amount analysis unit 31 sets a weighting coefficient W described later based on the generated code amount for each video shot. The feature amount analysis unit 31 supplies the weighting coefficient W, video shot information, and the like to the prior code amount prediction unit 32.

  The prior code amount prediction unit 32 predicts the occupation amount (data accumulation amount) of the buffer memory on the reception side of the encoded stream based on the previous encoding result. FIG. 4 is a diagram schematically illustrating an occupation amount of the buffer memory on the reception side before the feature amount analysis unit 31 analyzes the feature amount. FIG. 5 is a diagram schematically illustrating the occupation amount predicted by the prior code amount prediction unit 32. In the case of FIG. 4, since the occupation amount exceeds the upper limit value (buffer upper limit) of the buffer memory, the buffer memory fails. In order to avoid such a situation, the occupation amount is predicted for each scene as shown in FIG. At this time, the prior code amount prediction unit 32 calculates the prediction code amount for each scene using the prediction graph shown in FIG. The horizontal axis of this prediction graph represents the quantization width x, and the vertical axis represents the generated code amount f (x) related to the quantization width x. The predicted code amount P (n + 1) is obtained by multiplying the previous generated code amount P (n) by the weighting coefficient W and (f (Q) −f (Q + dQ)) / f (Q). . That is, P (n + 1) = P (n) × W × (f (Q) −f (Q + dQ)) / f (Q). The prediction result by the prior code amount prediction unit 32 is supplied to the optimal encoding prediction unit 30. Note that the prediction graph can obtain the relationship between the quantization width and the generated code amount as shown in FIG. 5 by statistically processing the generated code amount for the encoding process.

  The optimal encoding prediction unit 30 determines the quantization width Q according to the target code amount for each scene, using the prediction graph shown in FIG. 6 based on the prediction result by the prior code amount prediction unit 32. The determined quantization width Q is supplied to the encoder 1. Moreover, the optimal encoding prediction part 30 acquires a feature-value from the 2nd encoding result. This feature amount can be used for the third encoding process.

  As described above, according to the multi-pass compression encoding of the present embodiment, the target code amount is calculated based on the prediction graph (FIG. 6), so the generated code amount after the second time is accurately converted to the target code amount. It becomes possible to match well. Further, it is possible to quickly converge to a constant bit rate by repeating the above-described compression encoding process for three or more passes, and it is possible to efficiently execute the fixed bit rate encoding.

H. 2 is a block diagram schematically showing a configuration of an encoder compliant with H.264. It is a figure for demonstrating the encoding process of the 1st time (1st pass). It is a figure for demonstrating the encoding process after the 2nd time (2nd pass). It is a figure which illustrates roughly the occupation amount of the buffer memory on the receiving side. It is a figure which illustrates roughly the occupation amount estimated by the prior code amount prediction part. It is a figure which shows roughly the prediction graph which estimates the generated code amount regarding a quantization width | variety.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Encoder 10 Subtractor 11 Conversion part 12 Quantization part 13 Entropy encoding part 14 Inverse quantization part 15 Inverse conversion part 16 Adder 17 Loop filter 18 Frame memory 19 Intra-frame prediction part 20 Motion compensation part 21 Motion prediction part 21 22 switch unit 23 encoding control unit

Claims (3)

A multi-pass compression encoding method of two or more passes,
(A) orthogonally transforming and quantizing the moving image signal and entropy encoding to generate an encoded stream;
(B) setting a target code amount for each unit interval based on the encoding result obtained in step (a);
(C) referring to a prediction graph for predicting a generated code amount with respect to a quantization width, determining a quantization width according to the target code amount, and executing the process of step (a);
A compression encoding method comprising: A compression encoding method according to claim 1, wherein
Obtaining a feature amount based on the encoding result obtained in step (a);
Determining the unit section based on the feature amount;
A compression encoding method, further comprising: A compression encoding method according to claim 2, wherein
Based on the encoding result obtained in the step (a), further comprising the step of predicting the occupation amount of the buffer memory on the receiving side that receives the encoded stream;
The step (c) includes a step of determining the quantization width according to the occupation amount.

Images