JPH01185087A - Image encoder - Google Patents

Abstract

PURPOSE:To obtain a more satisfactory picture quality even in the same transmission rate as the case where a vector quantization is not used by adding to the vector quantization to a prediction encoding. CONSTITUTION:At the time of executing a transmission by k-bits per 1 picture element, quantizers 50A and 50B quantize the difference signals of inputs by (k+1)-bits or above for the vector quantization in a following stage. Data quantized by the quantizers 50A and 50B are impressed to vector quantizer 52, vector-quantized to 2k-bits (k-bits per 1 picture element) there, and outputted from an output terminal 30 to a communication line. In such a way, at the time of executing the encoding by, for example, k-bits per 1 picture element, the quantizers 50A and 50B execute the quantizations by (k+1)-bits per 1 picture element first, and the quantization is executed to 2 bits (k-bits per 1 picture element) by the vector quantizer 52. Thus, the picture quality can be improved.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an image encoding device, and more specifically to an encoding device using a predictive encoding method.

[Conventional technology]

Predictive coding (hereinafter referred to as DPCM) obtains the predicted value of the next coded sample value from the past coded sample value,
This method encodes and transmits the difference (error signal) between the predicted value and the actual sample value, and the stronger the correlation between the sample values, the higher the information compression rate, which makes it easier to transmit and transmit image and audio signals. It is often used for recording purposes.

However, when the transmission rate is extremely high, predictive encoding processing must also be made faster, but there is a limit to how fast it can be done. Therefore, the sampled video data is
A method has been proposed in which each pixel is sequentially and cyclically supplied to a CM encoder and the measurement processing speed is reduced to one of a plurality of bones.

However, in this method, in order to calculate the predicted value of a pixel in each encoder, the data input immediately before,
That is, data of adjacent pixels is not used. Therefore, for example,
When raster-transmitted video data is one-dimensionally encoded, prediction errors are large because predicted values are generated using pixels spaced apart by a plurality of pixels in the horizontal direction on the screen. This is because the correlation between the pixel and the pixel used to generate the predicted value of that pixel decreases.

Generally, differential data is nonlinearly quantized to reduce the amount of data, but if the prediction error is large, the difference between the representative value of the nonlinearly quantized data and the true value becomes large, and the transmitted image data deteriorates. Resulting in. Therefore, the applicant
Another patent application (Japanese Patent Application No. 62-272282) proposed that parallelization of predictive encoding processing be performed in units of a predetermined period. This process will be briefly explained below.

FIG. 3 shows a block diagram of the configuration of an encoder (or transmitter) that performs parallel processing on a line-by-line basis for high-speed DPCM processing. The digital signal input from the input terminal 10 is
The data is written into the line memories 14A and 14B by a switch 12 that is switched to contacts a and b for each line (horizontal scanning line). In this case, since the input signal is divided into two, the time axis is expanded twice on the memories 14A and 14B. In other words, in the case of n division, it is multiplied by 0. Each data expanded twice in the memories 14A and 14B is subtracted by a predicted value, which will be described later, in subtracters 16A and 16B.
It is input to 8A and 18B.

The quantizers 18A and 18B quantize the input value into data of bits per pixel and output the data.

The inverse quantizers 20A and 20B are the quantizer 18A.

The difference value (actually, its representative value) is decoded from the 18B quantized data. Adders 22A and 22B are inverse quantizers 20
The predictor 24A restores the input signal by adding the predicted value to the representative difference values from A and 20B.

24B predicts the next pixel value from the human input signal restored by the adders 22A and 22B, and uses the next pixel value as the predicted value.
subtracters 16A, 16B and adder 22A in M cycles;
22B. Predictive encoding is performed sequentially through this operation loop. This is called DPC old rape.

Output of leaf CM processed in parallel (quantizer 18A).

18B) are respectively line memories 26A, 2
6B and line memory 14A.

In order to restore the time axis that was doubled in 14B, the time axis is compressed to 172 here. Then switch 28
By selectively connecting the a and b contacts in the order of the a and b contacts, DPCM data is output from the output terminal 30 in the order of input at the input terminal 10.

FIG. 4 shows a block diagram of the configuration of the receiving side (decoding side) corresponding to FIG. 3. An encoded signal is manually input to the input terminal 32, and this encoded signal is sent to the line memories 34A and 34B alternately at the a and b contacts by a switch 33 that switches for each line, similar to the encoding side. written. In the line memories 34A and 34B, the time axis is expanded twice. The DPI dequantizers 36A and 36B decode the DPCM code and output a representative difference value. adder 38A,
38B adds a predicted value, which will be described later, to the representative difference value,
Restore the original value. The predictors 40A and 40B are adders 3
The predicted value of the next pixel is calculated from the outputs of 8A and 38B, and the adder 38A calculates the predicted value of the next pixel from the outputs of 8A and 38B.

Return to 38B. This series of operations is called a DPCM decoding cycle.

The time axis of the decoded signals processed in parallel is compressed to 1/2 in the line memories 42A and 42B, and the lines are output from the output terminal 46 alternately by the switch 44.

According to the above configuration, the prediction error is thickened (without
It becomes possible to perform high-speed predictive encoding processing.

[Problem to be solved by the invention]

In the case of only one-dimensional predictive encoding, the image quality is almost determined by the number of quantization levels, and if you want to further improve the image quality, you can use higher-order predictive encoding such as two-dimensional or three-dimensional. A prediction circuit is required for this purpose, and the circuit scale becomes enormous.

Therefore, an object of the present invention is to provide an image encoding device that can achieve high image quality without increasing the circuit scale.

[Means to solve the problem]

The image encoding device according to the present invention is characterized by comprising predictive encoding means for predictively encoding sample values of an image, and vector quantizing means for vector quantizing the predictive code by the predictive encoding means. do.

[Effect]

By performing vector quantization on a plurality of highly correlated pixels, high image quality can be achieved while maintaining a high compression ratio. Therefore, by performing vector quantization on highly correlated predictive codes, it is possible to achieve higher image quality even at the same transmission rate.

〔Example〕

Hereinafter, the present invention will be described in detail with reference to the drawings.

FIG. 1 shows a configuration block diagram of a transmission system of an embodiment in which the present invention is applied to a -dimensional predictive encoding device.

The same components as in FIG. 3 are given the same reference numerals.

The signal input to the input terminal 10 is written to the line memories 48A, 48B line by line by the switch 12. In the line memory 48A or 48B, for vector quantization, the time axis is adjusted or expanded so that pixels located above and below on the screen have the same timing. The outputs of line memories 48A and 48B are respectively
The adders 16A and 16B convert the signal into a difference signal, which is applied to the quantizers 50A and 50B. When it is desired to transmit bits per pixel, the quantizers 50A and 50B quantize the input difference signal to (k+1) bits or more for vector quantization in the subsequent stage. The data quantized by the quantizers 50A and 50B is applied to the vector quantizer 52, where it is vector quantized into 2 bits (bits per pixel) and output from the output terminal 30 to the communication path. .

The output of the vector quantizer 52 is sent to the inverse vector quantizer 54
is also applied. The inverse vector quantizer 54 decodes the vector quantized data into DPCM encoded data, and
It is applied to inverse quantizers 56A and 56B. The DPCM dequantizers 56A and 56B calculate the difference representative value, and the adder 22
A, 22B. Adders 22A, 22B and predictors 24A, 24B determine the predicted value of the next pixel.

For example, when encoding with bits per pixel, the third
In the conventional encoding device shown in the figure, the quantizers 18A and 18B quantize bits per pixel, whereas in this embodiment, the quantizers 50A and 50B quantize bits per pixel (k+1
) bits or more, and the vector quantizer 52 quantizes the image into 2 bits (bits per pixel), thereby improving image quality.

FIG. 2 shows a block diagram of a decoding system that decodes the transmission code shown in FIG. 1. The same components as in FIG. 4 are given the same reference numerals. The encoded data input to the input terminal 32 is decoded into a DPCM code by the inverse vector quantizer 60. That is, inverse vector quantizer 60 functions in the same way as inverse vector quantizer 54 of FIG. The DPCM code decoded by the inverse vector quantizer 60 is decoded into a differential representative value by the inverse quantizers 62A and 62, and then the adders 38A and 38B and the predictors 40A and 40B perform decoding processing similar to the conventional example. will be held. The signal decoded in this way is
The data is written to memories 64A and 64B.

In the line memories 64A and 64B, processing is performed to restore the time axis adjustment made on the encoding device side (line memories 48A and 48B in FIG. 1), and the line memories 42A and 64B
, 42B are supplied line by line to an output terminal 46 by a switch 44.

At the output terminal 46 the decoded signal is obtained.

In the above description, -dimensional predictive coding was taken as an example, but the present invention is not limited to this, but can also be applied to high-order predictive coding such as two-dimensional coding. In addition, although the case where two systems of predictive encoding means are used is taken as an example, it is also possible to have three or more predictive encoding means. Alternatively, vector quantization may be performed by temporarily storing the data in .

〔Effect of the invention〕

As can be easily understood from the above explanation, according to the present invention, by adding vector quantization to predictive coding, better image quality can be obtained at the same transmission rate as when vector quantization is not used. . Furthermore, there is no need to provide a line memory for compression/expansion of the time axis on the encoding side and the decoding side.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of the configuration of the transmitting side according to an embodiment of the present invention.
FIG. 2 is a configuration block diagram of the receiving side, FIG. 3 is a configuration block diagram of a transmitting device that performs parallel processing, and FIG. 4 is a configuration block diagram of a receiving device that performs parallel processing.

Claims (1)

[Claims] An image encoding device comprising: predictive encoding means for predictively encoding sample values of an image; and vector quantizing means for vector quantizing the predictive code by the predictive encoding means.