JPH07120962B2 - Quantization method and apparatus - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Industrial field of application) The present invention relates to a quantization method used in an encoding device for shortening the transmission time of an image signal or an audio signal, or for reducing the storage capacity, and the device thereof. Regarding

(Prior Art) Data compression methods for image or audio signals include information-preserving encoding and information-non-preserving encoding. Information-preserving coding refers to coding that does not include quantization in the coding process, and it is possible to reproduce exactly the same signal as the original signal by the processing of coding and decoding, but with high compression. You can't get a rate.

On the other hand, the non-information-preserving type encoding refers to the type that includes some kind of quantization processing in the encoding process, and includes the quantization noise of the reproduced signal due to the processing of encoding and decoding, but it is highly compressed due to the deterioration of quality. The rate is obtained.

In information non-conservation type encoding including quantization, there are cases where a variety of qualities of reproduced signals from low quality to high quality are required, and due to the limitation of transmission path capacity and transmission time. Transmission at a wide variety of compression rates may be required. For example, there are cases where a rough image is required for the image signal and therefore high-speed transmission is desired, and there is a case where high-quality transmission is required because it may take time.

In order to deal with such a case, conventionally, the statistic amount of the image signal is measured, and the quantization bit number of the quantizer based on the statistic amount (the number of bits after quantization, which is an output representative in the quantization). (Corresponding to the number of points) was changed.

(Problems to be Solved by the Invention) When the number of quantization bits of the quantizer is changed based on the statistical amount of the input signal as in the conventional case, the variance value of the input signal is used as the statistical amount. That is, the quality of the reproduced image when quantized with the number of quantization bits is estimated based on the statistical amount of the input signal, and the number of quantization bits that achieves the required quality is determined. However, since the quality of the reproduced image is estimated by using the statistical amount of the image as described above, there is a problem that the quality of the reproduced image actually obtained cannot be accurately controlled.

In order to solve such problems, the number of quantization bits for each unit that quantizes the input signal (each signal in the case of scalar quantization, each signal included in a vector in the case of vector quantization) There is also a way to determine. However, when such a method is adopted, information indicating the determined number of quantization bits is required for each unit of quantizing the input signal, and this information becomes huge, so a high compression rate cannot be obtained. There is a point.

The present invention obtains the average power of error signals with respect to all input signals, and when this average error power is larger than a preset threshold value, the number of quantization bits is increased and the quantization is performed again to reproduce the reproduced image. The quality can be controlled accurately.

(Means for Solving Problems) A quantization method according to the present invention is a quantization output in which a quantization bit number is variable, and a quantization output and an input signal obtained by quantizing an input signal with a preset quantization bit number. The average error power for all the input signals is calculated by adding the power of the quantization error, which is the error of, and the average error power is compared with a preset threshold value, and when the average error power is larger than the threshold value, Increase the set number of quantization bits,
It is characterized in that the input signal is requantized.

Further, the quantizing device of the present invention is, in a quantizing device using a quantizing means having a variable quantizing bit number, an input storing means for storing an input signal, and a bit number for storing the quantizing bit number in the quantizing means. Storage means, dequantization means for receiving a quantization index output from the quantization means and outputting a quantized output, and means for obtaining an error signal between the output of the storage means and the output of the dequantization means , An average error power calculating means for adding the powers of the error signals to obtain an average error power for all input signals, a comparing means for comparing the average error power with a preset threshold value, and referring to the comparison result When the average error power is larger than the threshold value, the quantization bit number stored in the bit number storage means is increased until the average error power becomes smaller than the threshold value. And having a control means for controlling repeat the process.

(Operation) The quantization method of the present invention will be described.

In the quantization method with variable number of quantization bits, the input signal is first quantized with a preset number of quantization bits. Then, the power of the quantization error, which is the error between the quantized output and the input signal, is added to all the input signals.

From this addition result, the average error power for all input signals is obtained and compared with a preset threshold value. If the average error power is smaller than the threshold value, the quantization process is terminated when the reproduced image of the required quality is obtained.

Also, if the average error power is greater than the threshold,
Since the reproduced image of the required quality cannot be obtained, the number of quantization bits is increased and the quantization process is performed again. Then, by increasing the number of quantization bits and repeating the quantization process until the average error power becomes smaller than the threshold value, it is possible to obtain a reproduced image of required quality.

By changing the value of this threshold value, the quality of the reproduced image can be controlled freely and accurately.

Embodiment An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the present invention. The input signal may be an image signal or an audio signal (for example, a signal of 8 bits per sample) as it is, or a prediction error signal after performing some prediction processing. Orthogonal transform coefficients after orthogonal transform such as DCT (discrete cosine transform) may be used (for example, 8 × 8).
8 bit signal with 1 coefficient after performing DCT with 1 as a block).

This input signal is stored in the input storage unit 1 and then the signal 10
It is read as 1, and is quantized by the quantizer 2.
The number of quantization bits (the number of bits after quantization and corresponding to the number of output representative points in quantization) B (0) in the quantization unit 2 is stored in the number-of-bits storage unit 3 in advance.

FIG. 2 is an explanatory diagram showing the quantization processing in the quantization unit 2. As shown in FIG. 2, when 2-bit linear quantization is performed on the dynamic range a of the input signal, the quantization index 102, which is the output of the quantization unit 2, is any one of 0 to 3. Next, this quantization index is inversely quantized by the inverse quantization unit 4, and the quantized output 103 of the level shown in FIG. 2 is output. Then, in the subtraction unit 5, the quantized output 103 and the signal 1 stored in the input storage unit 1
The difference with 01 is obtained and the difference signal 104 is output.

Upon receiving the difference signal 104, the average error power calculator 6 calculates the power of the difference signal 104, adds the powers, and obtains the sum. The power of the difference signal 104 corresponds to the power of the quantization error. At the same time, the average error power calculator 6
The result of dividing the total power of 104 by the number of samples of the input signal is output as the average error power 105.

Then, the comparison unit 7 compares the average error power 105 with a preset threshold TH. At the time when the quantization processing of all input signals is completed, the control unit 8 causes the comparison result 10
6, when the average error power 105 is larger than the threshold value TH, the quantization bit number B set in the bit number storage unit 3 is set.
(0) is read out, a preset increment d is added to this value, and a new quantization bit number B (1) is written. Then, the control unit 8 outputs a control signal to the average error power calculation unit 6 to set the total power to 0. After that, the control unit 8 outputs a control signal to the quantization unit 2 to start the quantization processing with the new quantization bit number B (1).

By repeating such processing, the average error power 105 becomes the threshold T
When it becomes smaller than H, the quantization process ends. That is, the average error power 105 is the threshold value TH set in advance.
The quality of the reproduced image is guaranteed because it is smaller than

In the above description, the case where the quantizer 3 is a scalar quantizer that quantizes signals one by one has been described, but the same processing can be performed in vector quantization in which a plurality of signals are collectively quantized. . For example, in the case of performing vector quantization by the binary tree search on the transform coefficient after DCT transformation in the encoding of the image signal, the square of the distance between the input vector and the quantized output vector is taken as the error power, and the average error power Ask for. Then, the depth (corresponding to the number of quantization bits) of the binary tree search may be controlled by comparing the average error power and the threshold value.

Also, when performing vector quantization for all searches, similar control can be performed by switching the codebook in which representative vectors are stored in correspondence with the number of quantization bits.

(Effects of the Invention) As described above, by using the quantization method and the apparatus thereof according to the present invention, the reproduced image is much more accurately compared to the case where the number of quantization bits is determined by using the statistical amount of the input signal. You can control the quality of.

Further, as compared with the case where the number of quantization bits is determined every time the input signal is quantized, the amount of information to be transmitted or stored is small, so that a high compression rate can be obtained.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention, and FIG. 2 is an explanatory diagram showing the quantization processing in the quantization unit 2. 1 ... Input storage unit, 2 ... Quantization unit, 3 ... Bit number storage unit, 4 ... Inverse quantization unit, 5 ... Subtraction unit, 6 ... Average error power calculation unit, 7 ... Comparison unit , 8 …… control unit, 101 ……
Signal, 102 ... Quantization index, 103 ... Quantization output, 104 ... Difference signal, 105 ... Average error power, 106 ...
Comparison result.

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI technical display location // H03M 1/12 C (56) Reference JP-A-56-66929 (JP, A) JP-A JP-A-58-168344 (JP, A) JP-A-58-179897 (JP, A) JP-A-62-51828 (JP, A) JP-A-59-72244 (JP, A) JP-A-55-135421 (JP , A) JP 62-3535 (JP, A) JP 58-78200 (JP, A) JP 62-169530 (JP, A) JP 63-220300 (JP, A) JP 59-127440 (JP, A) JP 60-131 (JP, A) JP 61-34697 (JP, B2) JP 5-31331 (JP, B2) JP 4-14360 (JP, A) B2) Japanese Patent Publication 3-8136 (JP, B2) Japanese Patent Publication 4-45014 (JP, B2) Special Table Sho 59-500077 (JP, A)

Claims (2)

[Claims] 1. A quantization method in which the number of quantization bits is variable, the power of a quantization error, which is an error between a quantized output obtained by quantizing an input signal with a preset number of quantized bits, and an input signal is added. The average error power for all input signals is calculated, and this average error power is compared with a preset threshold value. If the average error power is larger than the threshold value, the set number of quantization bits is increased to A quantizing method characterized in that a signal is quantized again. 2. A quantizer using a quantizer having a variable quantizer bit number, an input storage unit for storing an input signal, and a bit number storage unit for storing the quantization bit number in the quantizer. Dequantizing means for receiving a quantization index output by the quantizing means and outputting a quantized output; means for obtaining an error signal between the output of the storage means and the output of the dequantizing means; and the error signal Of the average error power for all input signals by adding the powers of the above, an average error power calculating means, a comparing means for comparing the average error power with a preset threshold value, and the average error power referring to the comparison result. Is greater than the threshold value, the quantization process is repeated by increasing the quantization bit number stored in the bit number storage means until the average error power becomes smaller than the threshold value. Quantization apparatus characterized by a control unit for performing controls.