JP2962566B2 - Variable-rate video layer coding device - Google Patents

Description

The present invention relates to a variable-rate video coding system using a high-speed packet network, which is designed to suppress image quality degradation even when packets are discarded due to network congestion. The present invention relates to a layer coding apparatus that performs layered coding on a portion having a large effect on a portion and a portion having a small effect on the portion.

"Prior Art" FIG. 5 shows the configurations of a transmission unit 11 and a reception unit 12 of the variable rate video hierarchical coding system described in Japanese Patent Application No. 63-251391.

In the transmission unit 11, the input video signal is subtracted from the signal of the previous frame stored in the frame memory 102 by the subtracter 101, and the inter-frame prediction error signal obtained by the subtraction is blocked by the orthogonal transform encoder 103. The redundancy is suppressed in units, and the data is converted into orthogonal transform coefficients. Orthogonal transform encoder
The orthogonal transform coefficients from 103 are separated by the layering unit 104 into a part MSP that is important in image quality and a part LSP that is not. The separated orthogonal transform coefficients are obtained from only the MSP in the packet transmitting unit 107,
Or, it is packetized only from LSP and the packet is
An identifier indicating which of SP and LSP is included is assigned, and MSP
Is transmitted by a higher priority transmission class in the network and LSP is transmitted by a lower priority transmission class. In addition, among the separated orthogonal transform coefficients, the MSP is used for prediction of the next frame, so that it is inversely orthogonally transformed and decoded by the orthogonal transform decoder 105, and its decoded output is added to the signal of the previous frame by the adder 106 As a result, the moving image signal is reproduced and stored in the frame memory 102.

The receiving unit 12 assembles the coded information from the received packet, and determines whether the information belongs to the MSP or the LSP based on the identifier.

The information belonging to the MSP is decoded by the orthogonal transform decoder 202, and its decoded output is added to the signal of the previous frame stored in the frame memory 205 by the adder 204, and is reproduced as a moving image signal consisting only of the MSP. Are stored in the frame memory 205 for decoding. A moving image signal including only the MSP is output as a moving image signal obtained by adding the signal obtained by decoding the coded information belonging to the LSP by the orthogonal transform decoder 203 and the adder 206.

Since the packet containing the MSP and the packet containing the LSP are transmitted with different priorities, when the network is congested, the packet containing the LSP will be discarded. Therefore, when discarding occurs, the lack of information belonging to the LSP does not lead to large image quality deterioration, and the information belonging to the LSP is not used for inter-frame prediction, so that the image quality deterioration due to the loss does not continue in the subsequent frame. In addition, image quality degradation due to packet discard can be minimized.

"Problems to be Solved by the Invention" As described above, the hierarchical coding scheme uses only the MSP for inter-frame prediction so that the effect of discarding the LSP does not affect subsequent frames. Therefore, the accuracy of inter-frame prediction changes depending on which orthogonal transform coefficient is used as the MSP, and the entire information amount changes.

In the conventional method, when the MSPs and LSPs are hierarchized, the entire information amount is not taken into consideration, so that there is a problem that the entire code amount increases due to the hierarchies.

According to the present invention, when the orthogonal transform coefficients are hierarchized into MSPs and LSPs, whether or not each orthogonal transform coefficient is effective for predicting the next frame in units of orthogonal transform blocks Is determined to be effective for prediction, that is, for example, when the code amount is calculated when the orthogonal transform coefficient is used for prediction and when it is not used, and when the code amount is reduced using the orthogonal transform coefficient, the orthogonal transform MS coefficient
P, otherwise the orthogonal transform coefficient is LSP.

"Operation" According to the present invention, the coefficients effective for inter-frame prediction
And it is possible to suppress an increase in the code amount when the encoded information is hierarchized.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 shows a first embodiment of the present invention and its receiving section.

In the transmission unit shown in FIG. 1, the k-th frame of the input video signal is orthogonally transformed by the orthogonal transformation unit 301, and is output as an orthogonal transformation coefficient Ck. This orthogonal transform coefficient Ck is converted into a quantization index by the following two types of processing.

a. The difference from the prediction value predicted by the coefficient k of the previous frame stored in the coefficient memory 309 is obtained by the subtractor 302 and output as an inter-frame prediction error signal Ck-k. The prediction error signal is quantized by the quantization unit 304, and is output as a quantization index Q [Ck-k].

b. Quantized by the quantizer 303, and the quantization index Q [C
k].

As described above, two types of quantization indexes Q [Ck-k] and Q [Ck] are output for one orthogonal transform coefficient Ck, and these quantization indexes are determined by the inter-frame / intra-frame determination unit.
At 305, one of the quantization indexes is selected as the coded information Ek by the determination algorithm described below.

First, let N be the number of orthogonal transform coefficients in the orthogonal transform block. For example, if orthogonal transformation is performed on a block of 8 pixels × 8 lines, N = 64.

Next, orthogonal transform coefficients in one orthogonal transform block numbered from 1 to N according to a certain order are denoted by Ck, i (i = 1
…, N). As a numbering method, for example, a method of scanning discrete cosine transform coefficients recommended by CCITT, H.261 shown in FIG. 2 is used. Request t _k which minimizes the code amount V in the block from the input two quantization indices.

Here, CL [] is a function that gives a code length.

The t _k that minimizes equation (1), t _k or less, the quantization index Q of performing inter-frame prediction [Ck-k], otherwise the coded information Ek the Q [Ck].

The coded information Ek is inversely quantized by the inverse quantization unit 306, and the inversely quantized orthogonal transform coefficient is added to 'k by the adder 307, and stored in the coefficient memory 309 for prediction of the next frame. Note that 'k is used by the coefficient control unit 308 to determine the inter / intra-frame determination unit 305 of the inter-frame prediction signal k.
And the coefficient selected as Q [Ck] as coding information Ek is 0
Level. Encoding information Ek is stored in buffer 310
Is temporarily stored in the next frame (the (k + 1) th frame)
Is determined by the interframe / intra-frame determination unit 305, and when t _{k + 1} is determined, the hierarchical indexing unit 311 sets the quantization index equal to or smaller than t _{k + 1} to MSP as in Expression (2), Let the other quantization indexes be LSPs.

The information hierarchized into MSP and LSP are respectively subjected to variable-length coding by variable-length coding
It is packetized at 4,315 and transmitted.

 Next, the operation of the receiving unit will be described.

The received packet is assembled into the original information in the packet receiving unit 401, and the variable length decoding unit 40
In 2, after the variable-length decoding, the inverse quantization unit 403 performs inverse quantization, and the inversely quantized output is added to the coefficient of the previous frame stored in the coefficient memory 406 by the adder 404 to obtain an orthogonal transform coefficient. Is output. At this time, the coefficients of the previous frame stored in the coefficient memory 406 are at the 0 level except for the coefficients predicted between the frames, as will be understood from the following description. A conversion coefficient is output. This orthogonal transform formation is stored in the coefficient memory 406 for prediction of the next frame,
At this time, the orthogonal transform coefficient received as LSP by the coefficient limiting unit 405 is set to the 0 level. This is because the LSP orthogonal transform coefficient is not predicted between frames in the next frame, as described in the operation of the transmission unit. The orthogonal transform coefficient from the adder 404 is simultaneously subjected to inverse orthogonal transform by the inverse orthogonal transform unit 407 and output as a video signal.

FIG. 3 shows a second embodiment of the present invention and the configuration of the receiving section.

In FIG. 3, the input video signal is subtracted by a subtractor 50.
In 1, a difference from the symbol of the previous frame stored in the frame memory 510 is obtained and is set as an inter-frame prediction error signal.The prediction error signal is set as an orthogonal transform coefficient in an orthogonal transform unit 503, and the orthogonal transform coefficient is The quantization unit 505 sets the quantization index. At the same time, the input video signal is used as an orthogonal transform coefficient in the orthogonal transform unit 502 without being subjected to inter-frame prediction.
At 504, the quantization index is set. As for these two types of quantization indexes, the inter-frame / intra-frame determination unit 506 selects the smaller one of the code amounts for each block as coding information. This information is inversely quantized into an orthogonal transform coefficient by the inverse quantization unit 507, and the orthogonal transform coefficient is converted into a video signal or an inter-frame prediction error signal by the inverse orthogonal transform unit 508. If the inverse orthogonally transformed signal is an inter-frame prediction error signal, the signal is added to the orthogonal transformation coefficient of the previous frame stored in the frame memory 510 in the adder 509 to obtain a video signal, and the inverse orthogonally transformed signal is obtained. If the signal is a video signal, the signal of the previous frame is set to 0 level by the signal limiting unit 517. Therefore,
In any case, the output of the adder 509 becomes a video signal, and this video signal is used to predict the next frame.
Store to 0. At the same time, the above encoded information is stored in buffer 5
When the quantization index stored in the buffer 11 and subjected to inter-frame prediction is selected as the encoding information of the corresponding orthogonal transform block of the next frame, the information stored in the buffer 511 is used as the MSP and the inter-frame prediction is performed. If no quantization index is selected, the information stored in the buffer 511 is hierarchized as an LSP.

The MSP of the hierarchized information is converted into a variable-length coded sequence by a variable-length coding unit 513, further packetized by a packet transmitting unit 515, and transmitted together with an identifier indicating that the information is the MSP. On the other hand, among hierarchical information, LSP is
The variable-length encoding unit 514 converts the information into a variable-length encoded sequence, packetizes the packet with a packet transmitting unit 516, and transmits the packet together with an identifier indicating that the information is an LSP.

 Next, the operation of the receiving unit will be described.

The received packet is assembled into the original information in the packet receiving section 601, and the variable length decoding section 60
In 2, variable-length decoding is performed, and the decoded output is inversely quantized by an inverse quantization unit 603, and the inversely quantized output is inversely orthogonally transformed by an inverse orthogonal transformation unit 604 to be an inter-frame prediction error signal or a video signal. You. This signal is added to the signal of the previous frame stored in the frame memory 607 by the adder 605 and output as a video signal. At this time, the signal of the previous frame stored in the frame memory 607 is
As will be understood from the following description, since signals other than the signal predicted between frames are at the 0 level, the correct video signal is output from the adder 605. This video signal is stored in the frame memory 607 for prediction of the next frame.
At this time, the signal received as the LSP by the signal control unit 606 is set to the 0 level. This is because the LSP signal is not inter-frame predicted in the next frame as described in the operation of the transmission unit.

[Effect of the Invention] According to the present invention, the coefficients effective for inter-frame prediction
And it is possible to suppress an increase in the code amount when the encoded information is hierarchized. Fig. 4 shows the original picture as CCI
FIG. 4 shows a comparison result between the conventional method (method in which the number of orthogonal transform coefficients used as the MSP is fixed) for the medium motion type medium based on R, Rec601 by simulation. For the sake of simplicity during simulation, CL [] in equation (1) was evaluated using the power of the orthogonal transform coefficient, the quantization step was fixed, and the SN ratio of the encoded image was kept constant. The horizontal axis in FIG. 4 is the entropy per bit of MSP (code length when ideal variable length coding is performed), and the vertical axis is LSP
Shows the entropy per bit of. From this figure,
It can be seen that the system of the present invention can reduce the entropy of the LSP by about 15% when the entropy of the MSP is the same as that of the conventional system of Japanese Patent Application No. 63-251391, that is, the code amount can be reduced.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention and an example of a receiving section thereof, FIG. 2 is an explanatory diagram of an example of ordering of orthogonal transform coefficients, and FIG. 3 is another embodiment of the present invention and a receiving section thereof. FIG. 4 is a diagram showing encoding characteristics, and FIG. 5 is a block diagram showing a conventional hierarchical encoding device and its decoding unit.

────────────────────────────────────────────────── ─── Continuing on the front page Application for Patent Law Article 30 (1) is filed. Published on “Third International Work shop on Packet Video” on March 22, 1990. (72) Inventor Fumio Kishino Inaya, Seika-cho, Kyoto Mihiraya (Nanbanashi) AT R Co., Ltd. Communication Systems Research Laboratories (58) Field surveyed (Int.Cl. ⁶ , DB name) H04N ^7/ 24-7/68

Claims (1)

(57) [Claims] An orthogonal transformation means for orthogonally transforming an input video signal for each frame and outputting an orthogonal transformation coefficient Ck, and a difference between the orthogonal transformation coefficient Ck and a coefficient k of a previous frame stored in a coefficient memory is quantized. Quantized index Q
First quantizing means for outputting [Ck-k], and quantizing the orthogonal transform coefficient Ck to quantize
Second quantization means for outputting [Ck], the above-mentioned quantization indexes Q [Ck-k] and Q [Ck] are input, one frame is divided into N blocks, and the code amount in each block is minimized. the order so as to be in t _k number of Q [Ck-
k] and N-t _k-number of select in the Q [Ck], they inter-frame / inner judging means to be encoded information Ek, the inverse quantization means for inversely quantizing the coded information Ek, the An adding means for adding the inversely quantized orthogonal transform coefficient and the coefficient stored in the coefficient memory and storing the result in the coefficient memory for prediction of the next frame; and Q inside [Ck]
At the time of addition with the corresponding inversely quantized orthogonal transform coefficient,
When the coefficient tk _{+ 1} that minimizes the code amount in the next frame is determined by the coefficient limiting means for setting the coefficient stored in the coefficient memory to the 0 level and the inter / intra-frame determination means, Information corresponding to tk _{+ 1} Q [Ck-k] in information Ek is set as information MSP having a large effect on image quality, and information corresponding to other N-tk _{+ 1} Q [Ck] is used. Layering means for layering as information LSPs having little effect on image quality; first variable length coding means for performing variable length coding on encoded information Ek determined to belong to the information MSP and outputting the information; And a second variable-length coding unit that performs variable-length coding on the coded information Ek determined to belong to and outputs the result.