JPS60177787A - Vector quantizing system of color picture signal - Google Patents

Abstract

PURPOSE:To prevent deterioration of picture quality and obtain high efficiency of encoding by making forecast error encoding (DPCM) for each of luminance signal and color difference signal of picture signals and quantizing obtained differential signals giving priority to those components that give large influence to human visual valuation. CONSTITUTION:Y, I, Q signals are converted to differential signals eY, eI, eQ by a forecasting device 1. A quantizer 2 treats signals eY, eI, eQ as three-dimensional vector signals, and when quantizing, selects those nearest to vector E=(eY, eI, eQ) out of plural quantizing representative values arranged on only basic axes of three-dimensional vector coordinate. As quantizing characteristic, for instance, five values are allotted to eY axis, and three values of quantizing representative value qn(n=0-8) are allotted to eI and eQ axes respectively. In this quantizing characteristic, there is follow-up characteristic similar to normal scalar quantizing for signals that formed maximum change out of luminance signal or two color difference signal. However, in the case where two signals or three signals change largely, following-up becomes slow. However, deterioration of picture quality is hardly detected due to color difference error or luminance error masking. As mentioned above, quantizing is made utilizing man's visual characteristic.

Description

DETAILED DESCRIPTION OF THE INVENTION (Technical Field) The present invention relates to a quantization method for color image signals, and in particular:
The present invention relates to a vector quantization method for a color image signal separated into a luminance signal and two color difference signals.

(Background Art) With the recent widespread use of color image display terminal devices having a frame memory, expectations for high-speed V-digital transmission of color images are increasing. In order to meet this expectation, for example, taking a color image consisting of 512 x 512 pixels, a technology is required that can transmit this image in about 10 seconds. Considering a transmission speed of 64 Kbps, this is about 20 to 25 bits per pixel.

In the conventional method of transmitting a color image with brightness separation, each component signal is subjected to 8-point quantization, and 24 e nodes are used per pixel. If the above-mentioned requirements are to be satisfied using such conventional technology, it is possible to develop it further and reduce the amount of information to 1 in some way.
It must be compressed to 1/10 to 1/12.

Traditionally, predictive coding (
DPCM) method is common, but when this method is applied to image signals, the compression ratio for the amount of information is said to be limited to about 2.0 to 2.5, and the deterioration in image quality at this time is significant. Hey.

(Objective of the Invention) The present invention solves the above-mentioned drawbacks of the prior art, and its purpose is to provide a vector quantization method for color images that prevents deterioration of image quality and provides an extremely high compression ratio for the amount of information. do.

(Structure and operation of the invention) A feature of the present invention is that prediction error coding (DPCM) is applied to each of the luminance and color difference signals of a color image signal, and the obtained difference signal is processed by taking human visual characteristics into consideration. It is thick for visual evaluation (it consists in quantizing with priority given to the components that have an influence on the visual evaluation).

The visual characteristics referred to here are those that are sensitive to changes in the luminance component, but do not significantly affect the visual evaluation of the hue component.

Therefore, as a color image signal, it is easier to handle a signal separated into a luminance signal and two color difference signals.
, φ, A, etc., but none of them limit the scope of application of the present invention, so in the explanation of the present invention, they will be expressed as Y, I, and Q according to the NTSC system. shall be.

Note that when the color image generation source is the LG and B outputs, it may be converted into Y, I, and Q signals using the following equation.

Hereinafter, the present invention will be explained in detail using the drawings.

FIG. 1 is a system diagram of an embodiment of the present invention. 1 is a predictor, 2
is a quantizer, and 3 is an encoder.

The Y, I, and Q signals input to the predictor 1 are digitized signals. This is, for example, Y to be encoded.
, 1. If each of the Q analog signals is converted into PCM at a sampling frequency of approximately 12 to 13 MHz, approximately 512×512 pixels can be obtained as a final digital image signal. In the following explanation, an image signal of 512×512 pixels will be described as an example. These Y, I.

Using a predictive coding method for the Q signal, differential signals ef, e1 . Convert to e. Examples of predicted values of these differential signals for the Y, I, and Q signals are shown in FIG. 2 for the Y signal. In the figure, 4,5°6 is an addition element, 7 is a division element, 8,10 is a 1-image delay element, and 9 is a 5
11 pixel delay element, (Iy is the quantized value of ey quantized by quantizer 2. Here, the prediction function for calculating the predicted value Y is selected from pixels A and B from among the pixels shown in FIG. 3. , C
shall be selected. Figure 3 shows arbitrary two lines L14n-1 extracted from 512x512 pixels. Pixel F, the pixel located directly above and one line before X, and the pixel one pixel before B is D.
, B is shown as A, and the pixel two pixels after B is shown as E.

In FIG. 2, 1-pixel delay elements 8, 10 and 511-pixel delay elements extract pixels A, B, and C as prediction functions. That is, when trying to predict the pixel X in FIG.
The value of 511 Di wide pixel 9 output is the value to the pixel,
There is a time relationship in which the value of pixel B appears at the output of the one-pixel delay element 10.

These three values are added in addition element 5 and divided into three values in divider 7.
The predicted value YX of the pixel X expressed by the following equation is obtained.

△ YX−+(′i″A+YB×Y.)・・−(31Figure 2
In VC, YX is input to the addition element 4, and a difference signal eY from the input signal Y is obtained. Further, b is added to the quantized value qY of ey by the addition element 6, and this added value is treated as the value of X. At this time, the addition value of wolf and ey may be calculated by adding element 6, but since the value transmitted to the receiving side is qY, it is better to use Qy in order to reduce the error between transmission and reception. That's why.

As mentioned above, the method of determining Y is exactly the same for (2) and QX, and the predictor 1 in FIG. 1 uses Y.

The prediction circuit shown in FIG. 2 is provided for each of I and Q. Note that I and Q are expressed by the following equations, similar to equation (3).

Next, the quantizer 2 in FIG. 1 will be explained in detail.

[As described in 111, the feature of the present invention is particularly in this quantizer, and the three difference signals ey obtained by the predictor 1
, QX, and CQ are each considered as a vector element, and the whole is treated as one three-dimensional vector signal. In this quantization, multiple quantized representative values placed only on the fundamental axis of the three-dimensional vector coordinates are Vector b-(
, e y + e k, eQ).

Here, as the quantization characteristic of the quantizer 2, for example, FIG.
Prepare the items shown below. In Figure 4, the ey axis is 5 values,
The e- and eQ-axes are each expressed by a three-value quantized representative value qn (
n=o~8). The numbers 0 to 8 in the figure indicate the positions of the quantization representative values qn, and their respective coordinates are shown in Table 1, for example.

Here, the number of quantization steps of the luminance difference signal eY is
The reason why the number of quantization steps is larger than that of the color difference difference signal er + ec; is because human visual characteristics are sensitive to changes in brightness and are not very sensitive to changes in color difference. It attempts to quantize finely and coarsely color difference changes.

Now, the input vector signal E''= (ey + 81
, ``Q'' is represented by one quantized value according to the quantization characteristics shown in FIG.
, In ' qQn ), the input vector E and the vector q are determined by the following equation. Vector q with the minimum squared error with ~8. All you have to do is select.

(ey QYn )2+ (e□-qln)2+(e
Q ”Qn >2 (5) In addition, vector q11(i
You may select according to the l-th equation (MIN MAX method)
.

rninrnax ley qynl + lel q
lnl , leQ'qQn' Thus, the difference signal ey
By vector quantizing +eI+eQ, the quantized value can be extracted as a scalar quantity.

Furthermore, as is clear from FIG. 4, since the selected quantization representative value qn is predetermined, it can be expressed not as a value but as a position. For example, the numbers 0 to 8 in the figure correspond to this, and in order to transmit this information, 4 bits are sufficient even if equal length encoding is used, and an information compression rate of 1/6 can be obtained here. There is. Furthermore, in the encoder 3 of FIG. 1, if variable length encoding is performed in consideration of the appearance rate of each quantization representative value, the information compression rate can be improved, and in this case, the number of bits used can be reduced. It can be reduced to 2.5-30 bits.

The characteristics of the quantization method described above can be summarized as follows.

The quantization characteristics shown in Figure 4 always have the same tracking characteristics as normal scalar quantization for the signal with the largest change among the luminance or two color difference signals, but two of these signals Alternatively, if all three signals change sharply at the same time, 1ri tracking may be slow. This is because the vector quantization characteristic in FIG. 4 has the property of preferentially quantizing the maximum change signal, and as a result, for signals whose change amount is in the second or third order, 1/2 or 1
This is because the result is equivalent to /3 subsampling.

However, in actual encoded image quality, deterioration in image quality due to such quantization operations is hardly detected.

This is color difference error masking (], umi
nance masking of chromina
nce) includes luminance error masking (
Chro-rninance rnasking of
This is due to a visual characteristic called luminance. In addition, due to this property, conversely, as shown in Fig. 4, the quantization characteristics are such that even if the quantization vectors other than those on the YIQ difference signal axis are omitted, the resulting encoded image quality is Almost no deterioration is detected.

As described above, this embodiment can be said to be a quantization method that utilizes the permissible range of human visual characteristics such as color difference error masking and luminance error masking.

Note that a more specific configuration of the predictor 1 and quantizer 2 in FIG. 1 is shown in FIG. 5. Since this configuration can be easily understood from FIGS. 1 and 2, the contents of the quantizer 2 will be described here.

When the quantizer 2 has the quantization characteristics shown in FIG.
For example, if the quantization representative value qnk is determined in advance, the optimal quantization representative value can be previously determined for all combinations of the values of the difference signals ey, e□+eQ. Therefore, it is possible to tabulate the difference signal ey + eq) and the quantized representative value. This means that if the table is stored in Rαd and the content of the ROM is read out using the combination of the values of the difference signal ey + e□ + e (a as an address), the quantization representative value can be found. The ROM is constructed based on this idea.

Next, other embodiments of the present invention will be described.

In the embodiment of FIG. 1, the Y, I, and Q signals all had the same sampling rate, but in this embodiment, subsampling is further performed on the I and Q signals, and vector quantization is performed on them. It is characterized by

That is, as mentioned above, changes in the color difference signals I and Q have lower visibility and higher tolerance than changes in the luminance signal Y, so using this, the transmitting side can change the I and Q signals. The components are sent out mutually, that is, at a 1/2 sampling period, and the receiving side interpolates and interpolates the (2) or Q signal components that are not sent.

FIG. 6 shows a subsampling pattern, FIG. 7 shows a configuration example of a predictor and a quantizer, and FIG. 8 shows a quantization characteristic of the quantizer.

In this embodiment, based on the premise that ■ and Q signal components are sent to each other, the quantizers are configured to use Y and 1 or Y and Q signal components.
Two-dimensional vector quantization can be performed using For this reason,
As shown in FIG. 7, 3□, 3. These quantizers are provided with two quantizers, and their quantized outputs are switched and outputted at each sample timing by a switch SW1. SW2 is a switch linked to SWI.

The characteristics of the receiver 3Q are shown in FIG. 8(a), and the characteristics of the quantizer 31 are shown in FIG. 8(b), both of which are expressed in two-dimensional coordinates.
The representative quantization values are also 7 points from 0 to 6.

Therefore, only three pits are required to encode the selected quantization representative value. If this is subjected to variable length encoding as in the previous embodiment, it can be compressed to 2.0 to 2.5 bits.

In the configuration of FIG. 7, the method of creating the predicted values Y, I, Q of the y, I, Q signals and the difference signals eY, e□+% is basically the same as in FIG.

The above describes how to further compress the amount of information using the subsampling method. However, when considering the ■ and Q signals further, the visibility for the Q component is lower than the visibility for the ■ component. Therefore, it becomes possible to further reduce the sampling rate of the Q signal. In Figure 9, ■
A sampling pattern of 2/3 sampling for the signal and 1/3 sampling for the Q signal will be shown. In this case, SW, , S% of the example of FIG. 7 is shown in FIG.
All you have to do is switch according to the sampling pattern.

The quantization method on the transmitting side has been explained in detail above.

Although the quantization characteristics in each embodiment have been explained as having five values for the luminance component and three values for the color difference component, the quantization characteristics are not limited to this number and can be changed depending on the desired image quality. It is sufficient to increase or decrease the quantization representative values, or to appropriately arrange quantization representative values on other than the basic three axes shown in FIG. Furthermore, although the case where the quantization step is fixed has been described, it is also possible to change it in response to local fluctuations in the image signal. FIG. 10 shows the configuration of a decoder corresponding to the encoder of FIG. 7. Vector quantized representative vector position O~
The information indicating 6 is simultaneously input to the inverse quantizers I and 20 corresponding to the quantization characteristics shown in FIGS. 8(a) and 8(b).

In the inverse quantizer, the numbers and coordinates of the quantized representative vectors are determined according to the correspondence shown in Table 1, and these are input to the decoder 21, 22, 23 manually. At this time I, Q
The two quantization characteristics are appropriately switched according to the alternation of the quantization characteristics corresponding to the subsampling applied to the signal. These useless switches are shown as S, SY, and SQ. P□, P
Y and PQ are exactly the same as shown in FIG.

The decoded outputs 24, 25, and 26 are digital YIQ signals, but after D/A conversion, they become analog YIQ signals and become input signals to a display, which is an output terminal.

In FIG. 10, these circuits are omitted.

(Effects of the Invention) As described in detail above, according to the present invention, by collectively encoding luminance and color difference signals in consideration of visibility characteristics, it is possible to achieve high encoding without significantly deteriorating image quality. efficiency can be achieved. The present invention can be widely used in high-bandwidth transmission of color still images where particularly high coding efficiency is required, high-efficiency coding of moving image signals in TV conferences, etc.

[Brief explanation of the drawing]

FIG. 1 is a system diagram of an embodiment of the present invention, FIG. 2 is a diagram showing a specific example of a predictor for Y signals, FIG. 3 is a diagram for explaining the prediction function, and FIG. 4 is a quantization characteristic according to the present invention. A diagram showing
5 is a diagram showing a configuration example of a predictor and a quantizer, FIG. 6 is a diagram showing an example of a sub-sampling pattern, FIG. 7 is a diagram showing another example of a predictor and quantizer, and FIG. ) and (b) are diagrams showing another example of quantization characteristics, FIG. 9 is a diagram of another example of a subsampling pattern, and FIG. 10 is a diagram showing an example of the configuration of a decoder. 1; Predictor; 2; Quantizer; 3; Encoder. Figure 1 Figure 2 Diagram Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 (a) I Figure 9 Figure 10 Figure 2 Mu procedure amendment (method) % formula % 2. Title of the invention Vector quantization method for color image signals 3 Person making the correction: Relationship to the matter Patent applicant name (121) International Telegraph and Telephone Corporation - Rijin Sansho 1-5-12 Nishi-Shinbashi, Minato-ku, Tokyo 105 Japan Tampapil.

Claims (1)

[Claims] a prediction means that inputs a color image signal separated into a luminance signal and two color difference signals and calculates a difference signal between each signal and a predicted value; 3 defined by three differential signals
quantization means that arranges quantization representative values only on the basic axis of the dimensional vector coordinates, and selects a value indicating the maximum change in the quantization value of changes in the three difference signals to generate the quantization representative value. A vector quantization method for color images characterized by: