JPH0683444B2 - Image coding method - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to an image coding system for efficiently coding a television signal.

(Prior Art) Conventionally, a vector quantizer has been known as a method for encoding a voice or image signal with high efficiency.

In this method, a block is composed of a plurality of samples, and an input vector whose elements are sample values in the block and a vector having the smallest distance among the predetermined representative vectors are selected for each input vector, By transmitting the identification code, the signal is reproduced from the identification code sent on the receiving side.

The coded information amount R (bits / sample) is M when the number of samples in the block is N and the number of representative vectors is N (= 2 ^L ). Given by, for example, 256 for a (4 × 4) block
In a vector quantizer using (= 2 ⁸ ) representative vectors, R = 8 / (4 × 4) = 0.5 bits / sample.

Compared to the case where the prediction error signal in sample units in predictive encoding as another example of high-efficiency encoding is scalar-quantized, and also in the transform encoding in which the coding process is also performed in block units, the transform coefficients are scalar. The vector quantizer achieves higher coding efficiency than the case of quantizing,
It is known that the coding rate closest to the Rate-Distortion limit can be obtained.

The coding efficiency of the vector quantizer is determined by the amount of coded information expressed by equation (1) and the amount of coding distortion generated by quantization, and a set of representative vectors (codebook) that minimizes the average coding distortion is created. The law influences its quality.

An algorithm that creates a codebook using training data has been proposed.However, in order to create a codebook based on the statistical properties of the input signal in order to obtain a high coding rate, It requires a calculation using training data that is more than 10 times the number of vectors included.

It is known from the Rate-Distortion theory that the larger the number of samples in a block, and thus the higher the number of vector dimensions, the higher the coding efficiency. However, the size of the codebook (2 ') and the training data are increased accordingly. Need to increase the number of.

However, to a large block size from the computing power and the computation accuracy of the state of the computer point is limited, 512 to 1024 (= 2 ^{9-2 10} against (4 × 4) blocks
It is realistic to create a codebook of a moderate size.

As a quantization method for increasing the coding efficiency by increasing the block size and the number of codebooks and reducing the average coding distortion, a representative vector is obtained as the first-stage quantization, and then a distributed vector dictionary or the like is used. There is an invention in which the second quantization is independently performed for each vector component centering on each representative vector (Japanese Patent Application No. 57-204849).

However, in the above invention, the sum of the number of bits assigned to the first-stage quantization and the number of second-stage quantization bits is set to be constant. As described above, when the amount of information changes depending on the magnitude of the motion of the subject, it cannot be applied as it is to the configuration of the encoding method that avoids large image quality deterioration by controlling the amount of information generation per block. There was a flaw.

(Object of the Invention) An object of the present invention is to provide an image coding method using a vector-scalar quantization method, which is characterized in that the average coding distortion is reduced by controlling code allocation in block units. Especially.

(Structure of the Invention) (Differences between Features of the Invention and Conventional Techniques) The present invention relates to an error signal code when a vector quantization error is coded after blocking an image signal and performing vector quantization for each block. The number of bits allocated to the digitization is controlled based on the amount of code generation in the past and the size of the error signal of the current block, and the amount of information generation as a temporal average value is maintained at a constant value. Even if the change occurs, without causing a significant deterioration in the quality of the reproduced image,
The main feature is to minimize the average coding distortion.

It differs from the conventional technique in that the number of bits assigned to the coding of the vector quantization error signal is variable and its control method is different.

(Embodiment) FIG. 1 is a diagram showing a configuration of an embodiment in which the present invention is applied to a motion compensation interframe coding system of a television signal, in which 1 is a signal input terminal and 2 is a scan conversion circuit. 3 is a motion vector detection circuit, 4 is a variable delay circuit, 5 is a frame memory, 6 is a delay circuit, 7 is a subtractor, 8 is a significant block detection circuit, 9 is a vector quantizer, 10 is a subtraction circuit, 11
Is a vector quantization error calculation circuit, 12 is a code allocation control circuit, 13 is a scalar quantizer, 14 and 15 are addition circuits, 16 is a delay circuit, 17 is a multiplexing circuit, 18 is a buffer memory, 19 is a codebook storage The circuit, 20 is a signal output terminal.

First, the digitized image signal input from the signal input terminal 1 according to the raster scanning order is scanned by the scan conversion circuit 2.
According to the block size, for example, 4 × 4, 8 × 8, 16 × 16, etc., they are rearranged in the block scanning order.

The blocked image signal is input to the motion vector detection circuit 3, and the motion vector indicating the displacement of the subject between frames is detected in block units with reference to the reproduced image of the previous frame stored in the frame memory 5. It

The variable delay circuit 4 delays the reference image block from the frame memory 5 used for interframe prediction based on the detected motion vector and outputs the delayed reference image block to the subtractor 7.

On the other hand, the image block from the scan conversion circuit 2 is delayed by a time corresponding to the sum of the operation delay time required for motion vector detection in the delay circuit 6 and the delay time of the variable delay circuit 4, and the reference in the corresponding previous frame is obtained. It is input to the subtractor 7 in a state where the image block and the delay are adjusted.

The subtractor 7 performs subtraction on a pixel-by-pixel basis within the block, and the significant block detection circuit 8 uses the subtraction result to select an image block to be encoded and transmitted.

For example, the cumulative sum of absolute values in the block of the subtraction result is small,
Only the identification code indicating the invalid block is transmitted to the image block identified as the invalid block, and the subsequent encoding process is not applied.

The vector quantizer 9 performs block matching processing on the significant block with the representative vector (codeword) stored in the codebook storage circuit 19 and shows the representative vector with the minimum distance between them. The code (index) is sent to the multiplexing circuit 17 and the selected representative vector is sent to the subtraction circuit 10.

The subtraction circuit 10 performs pixel-by-pixel subtraction between the input and output vectors of the vector quantizer 9, and the vector quantization error signal resulting from the subtraction is sent to the scalar quantizer 13 at the next stage.

At the same time, the vector quantization error signal, in the vector quantization error calculation circuit 11, for example, the error signal power in the block is calculated, using this and a signal indicating the state of the information storage amount of the buffer memory 18, in the code allocation control circuit 12 The total number of codes to be assigned to the block being encoded and the bit allocation table for allocating each of the pixels in the block are determined.

The scalar quantizer 13 quantizes the vector quantization error signal on a pixel-by-pixel basis based on the bit allocation table, and sends a code representing the quantization level to the multiplexing circuit 17.

The representative vector selected by the vector quantizer 9 is added to the quantized output signal by the adder circuit 14 to reproduce the motion-compensated interframe difference signal for the significant block.

Further, in the adder circuit 15, the reference block in the motion-compensated previous frame is added via the delay circuit 16 having a delay time corresponding to the encoding processing time per block,
The input signal is reproduced. The reproduced image block is stored in the frame memory 5 and used as a motion-compensated inter-frame prediction reference image block of the next frame image.

The multiplexing circuit 17 time-division-multiplexes the motion vector, the significant / random block identification vector index, the bit allocation control code, and the coding indicating the quantization level for each pixel, and sends the time-division multiplexed data to the buffer memory 18.

The buffer memory 18 stores a code generated non-uniformly in block units, and sends the code to the transmission path via the signal output terminal 20 at a constant speed.

It also monitors the amount of information stored in the buffer memory and sends a control signal indicating that state to the code allocation control circuit 12.

In the code allocation control circuit 12, a variable B ₀ indicating the state of the information storage amount of the buffer memory 18 and a vector quantization error calculation circuit 11
From the function (B ₀ , ε _VQ ) shown in Eq. (1) with the error power ε _VQ obtained as
Ask for.

η = (B ₀ , ε _VQ ) (1) FIG. 2 is a diagram showing the relationship between the scaling coefficient that determines the bit allocation, the vector quantization error signal, and the buffer memory storage information amount.

Equation (1) is qualitatively shown in FIG. 2, and the larger ε _VQ ,
Further, the smaller B _{0, the} larger the scaling coefficient η.

FIG. 3 is a diagram showing a method of controlling scalar quantized bit allocation by scaling, in which sample data is obtained, each element of the bit allocation table of a predetermined representative vector is multiplied by a scaling coefficient η, and anew. Create a bit allocation table for each block.

In the above-described embodiments, any form can be used as the distortion measure function used for the block matching of the vector quantizer.

Further, the scaling coefficient as the bit allocation control code transmitted to the receiving unit is not necessarily limited to the equation (1), and it can be determined only by the state of the buffer memory, and the control cycle thereof can be determined. It can be easily inferred that the frame unit can be used instead of the block unit.

(Effects of the Invention) As described above, according to the present invention, the number of bits allocated to the second-stage scalar quantization can be adaptively controlled by providing the two-stage quantization process of vector quantization and scalar quantization. , A high-efficiency, high-quality coding method that enables coding processing to follow an input signal without causing significant image quality degradation even when applied to a television signal inter-frame coding method that includes intense subject movement. There is an advantage that can be realized.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration of an embodiment in which the present invention is applied to a motion compensation interframe coding system of a television signal, and FIG. 2 is a scaling coefficient for defining bit allocation, a vector quantization error signal and a buffer. FIG. 3 is a diagram showing the relationship with the amount of information stored in the memory, and FIG. 3 is a diagram showing a method of controlling the scalar quantization bit allocation by scaling. 1 ... Signal input terminal, 2 ... Scan conversion circuit, 3 ... Motion vector detection circuit, 4 ... Variable delay circuit, 5 ... Frame memory, 6 ... Delay circuit, 7 ... Subtractor, 8 ... Significant block detection circuit, 9 ... vector quantizer, 10 ... subtraction circuit, 11 ... vector quantization error calculation circuit, 12 ... code assignment control circuit, 13 ... scalar quantizer, 14, 15 ... Adder circuit, 16 ... Delay circuit, 17 ... Multiplexing circuit, 18 ... Buffer memory, 19 ... Codebook storage circuit, 20 ... Signal output terminal.

Claims (4)

[Claims] 1. A vector quantizer for a vector constituted by a sample value in a block as an element in an image coding method in which an image signal is divided into blocks in units of a plurality of samples and an encoding process is executed in units of the blocks. Scalar quantizer that quantizes vector quantization error for each vector component, equipped with buffer memory and code allocation control circuit for matching generated code amount and transmitted code amount to transmission line, past coded Controlling the number of bits allocated to the scalar quantization of the block in the code allocation control circuit using the state of the buffer memory based on the information generation amount of the block and the magnitude of the vector quantization error of the block currently being encoded. An image coding method characterized by. 2. The image coding method according to claim 1, wherein the image signal is a prediction error signal obtained by performing intra-frame prediction. 3. The image coding method according to claim 1, wherein the image signal is a prediction error signal obtained by performing inter-frame prediction. 4. A variance value of a vector component is obtained for each representative vector of a vector quantizer using sample data in a set to which an input signal belongs, and a bit distribution for a vector quantization error component is calculated using the variance value. It is possible to control the number of bits assigned to the scalar quantization for each block by increasing or decreasing the value of the bit allocation using a scaling coefficient determined by the state of the buffer memory and the vector quantization error. Claims (1), characterized by:
The image coding method according to item (2) or (3).