JPS58223978A - Band compressing system of video signal - Google Patents

Abstract

PURPOSE:To prevent erroneous propagation, by transmitting a picture element value of a picture element in a block formed with prescribed number of adjacent picture elements, obtaining a difference with a prescribed picture element value for other picture elements, quantizing the difference and transmitting it. CONSTITUTION:The value of each picture element of a digital signal is applied to a serial parallel converter 30 via an input terminal 29. The converter 30 outputs five data for five picture elements's share in parallel and the data are applied to a matrix 35 comprising difference calculus devices 31-34. The output of the difference calculus devices 31-34 is subjected to a prescribed quantization at quantizers 36-39 and applied to a parallel serial converter 40. The data of serial bits of a transmission line 41 is converted into a parallel data at a serial parallel converter 42 and applied respectively to a matrix 51 and inverse quantizers 43-46, output values of which are finally equal to input values of the quantizers 36-39. The output is added with a value of a reference picture element at adders 47-50 of the matrix 51.

Description

[Detailed description of the invention] The present invention relates to a band compression method for video signals.

When transmitting and processing video signals, various signal deteriorations occur due to transmission paths and signal processing errors. Therefore, various measures to eliminate such signal deterioration have been studied. One approach is to digitize the signal source. The digitization of information sources and digital information processing have continued to make remarkable technological advances in recent years, with i-semiconductor integration technology developing at a rapid pace. Incidentally, even in the field of handling video signals, video signal transmission and signal processing have come to be performed in digital form depending on the application.

Therefore, let's compare a digital video signal with an analog one. Taking the NTEC system Eirin 1F as an example, the frequency band of the video signal itself is about dM, so the sampling frequency is f3MHz or higher (actually L
1[' due to PF design and relationship with color subcarrier frequency, etc.
], 7MHz [often required]. on the other hand,
For quantization, 8 bits or more per pixel is required, and the total data rate reaches approximately 86 Mbit/sea (assuming the sampling frequency is 10.7 MHI and the number of quantization bits is 8) it. . In this case, 101
The maximum repetition frequency of 111 and 111 is about 43 MHz, which is about 10 times that of the original analog NT8C video signal.

In this way, by digitizing the video signal, signal deterioration can be significantly improved, but on the other hand, the band required for transmission increases significantly.

From this point of view, when a video signal is digitized and transmitted, how to narrow the necessary frequency bandwidth is a serious problem. There are roughly two approaches to narrowing this required frequency bandwidth. One of these is an improvement in the digital modulation method (also called 1-coding'), and the other is an improvement in band compression. The former digital modulation method is a method of transforming the data string to be transmitted and converting its frequency spectrum, while the latter band compression method is a method that distinguishes between truly necessary data and unnecessary (or unnecessary) data to be transmitted. How to convert data speed by separating unnecessary data). Of course, both of these are normally used to make effective use of the transmission path.

The present invention is related to the latter, and band compression of video signals will be described below. Various research and developments have been conducted on band compression of video signals, but they can be roughly divided into predictive coding methods (usually called "DPCM").
There are two methods: and orthogonal transformation method. The former method uses the characteristics of the video signal to delete redundant data. Specifically, in the case of video signals, the correlation between adjacent pixels is very high, so the value of the pixel to be transmitted is predicted from the adjacent pixels, and the difference between the predicted value and the true value is calculated. It detects the difference, then quantizes and transmits the difference.

On the other hand, in the latter method, multiple neighboring pixels are compared with various pixel patterns set in advance, converted into a function of each pixel pattern, and further quantized with the weight of each pixel pattern. This is considered to be the case.

The advantages and disadvantages of the two differ depending on the degree of band compression, but
Generally, the degree of band compression is between 1/2 and 1/2 at the boundary.
It is said to be around. For example, draw a pixel B btt
In this case, predictive coding is said to be advantageous when the average number of bits after compression is 3 to 4 bits or more, and conversely, when compressing to a smaller number of bits, alternating transformation is said to be advantageous. Of course,
As the degree of compression increases, the quality of the transmitted image deteriorates, so 4-bit data is usually used. Therefore, in terms of image quality, predictive encoding is slightly superior to compression degree i.

Therefore, a conventional example of DPCM with prior value prediction, which is the most rumored among the predictive coding methods C (hereinafter referred to as 'DPCM'), is shown in a block diagram in FIG. In Figure 1, (1)
ke input terminal, (2) is the difference, % (3) is the decoder (denoted as ke1Q1 in the figure), (4) is the inverse quantizer (denoted as ζτke1Q'' in the figure), (5 ) is an adder, and (6) is a delay device (
IDw in the figure], (7) is a transmission line, (8) is an inverse quantizer (indicated as @Q-11 in the figure), (9) is an adder, and OG is a delay device ( Inside is written as IDw), (11)
is the output terminal. Note that (A) to (G) in the figure will be explained later when used in the explanation.

The analog video signal is sampled, further A/I converted, converted into digital form, and applied to a differentiator (2) K via an input terminal (1). Delay unit (6) in difference unit (2)
The predicted value applied from the input terminal (1), the value added to the input terminal (1), and the q error, that is, the prediction error, are calculated, and the results are added to the quantizer (3). The Divine Converter (3) is a stone that quantizes manually generated prediction errors according to a predetermined procedure. For example, in the DPCM shown in Fig. 1, the number of t)it per pixel is changed from 8 bits to 41 bits.
) it, the quantizer (3) determines which of 16 (24) predetermined representative values the input prediction error should be. 41) It data indicating which of the representative values is the one is sent to the transmission path (7) and the inverse quantizer (4). The inverse M quantizer (4) returns the received 4-bit data to the original representative value, and supplies the representative value to the adder (5) K.

The adder (5) combines the output value of the Sakuen device (6) with the reverse Sabako conversion non-t.
41 output value is added, and the result is sent to the delay device (6) V
Enter large r. On the other hand, the 4-bit data that has passed through the transmission path (7) is similarly restored to the original representative value by the inverse quantizer (8), and added to the output value of the delay device θG by the adder (9).

The output of the adder (g) is sent out via the output terminal (river) and is simultaneously input to the delay device 00.Of course,
If no error occurs in this system, the predicted value (also called 1 local signal 1) output from the delay device (6) will be equal to the value sent out from the output terminal (11) [2 (, terminal i
It will be equal to or very close to the value applied to l+.

In addition, it is well known that the delay device (6) and fl (1) are also called 1 predictor 1, and in the case of Fig. 1, since it is a prediction method, it is constructed by the length of one sample period ρ. I'm being harassed.

Next, the relationship between the prediction method and pixels will be explained with reference to the pixel diagram shown in FIG. In FIG. 2, (countries to t19) represent pixels, and (20) and I21 represent the (n-1)th horizontal scan and the nth horizontal scan. Also, painting! +
13] (141 (17) (18) respectively U, V, W
,X. Now, considering the time point at which image (18) is predicted, in the case of previous value prediction, if the predicted value is Y, then x=W (101). A typical two-dimensional prediction is X=W-1-V-U
...(102) is mentioned. (102
) formula adds the amount of change from pixel (13) to pixel (14) by surface element (17)K to make the preview direct.

Of course, there are various prediction methods depending on the complexity of the video signal being handled (NTSC composite color video signal or black and white video signal) and sampling frequency.

By the way, the biggest problem in the conventional example of DPCM shown in FIG. 2 is error propagation. Although the DPCM method has excellent band compression characteristics, the first reason why it is not often adopted except in cases where a high-quality transmission path is used is because of the occurrence of error propagation.

Therefore, error propagation in DPCMr will be explained together with the @1 table.

Table 1 In Table 1, 0) to (G) represent each part of (A) to (G) in Figure 1, and (n-)) to (n+3) represent the input terminals (
1) indicates the pixel number applied to 0.128.
256 each indicates a value. In Figure 1, b per pixel
The number of it is 8)it, and 256 of pixels O~255
Indicated by gradation. Of course, if the dynamic range of the image signal is 1 pp, it goes without saying that one gradation will be 1/256 V. Below, the value of each pixel will be expressed as a number from 0 to 255.

The value applied to the input terminal +11 in FIG. 1 is shown in (a) of Table 1, which means that the 1st shift of 128 is always input. (b) is the output value of the differentiator (2). By the way, the output value of the inverse quantizer (3) is not necessarily equal to the output of the subtractor (2) because of the intervening quantization characteristics, but to simplify the explanation, the two will be explained as being equal. . Therefore, (b) shows the output 1h of the subtractor (2) and the inverse quantizer (4). (c) is an adder (
5), (2) is the output value of the delay device (6), (E) is the output value of the inverse quantizer (8), (F) is the output value of the delay device (G), and (G) is the adder. The output values of (9) (1, that is, the values output from the output terminal (11)) are shown, respectively.
Inverse quantization: If there is no error in the output value of (8), inverse quantization.

The output value is the same as in (4).

When the pixel number is in the range of (n-2') to (n+6), '128' is always input to the input terminal (1) as shown in (a), and the predicted value of the delay device (6) is Output is also constant”
128', and the prediction error is always 10 as shown in (b).
”.

By the way, assuming that the same error does not occur during transmission up to pixel number (n-1), the output value of the inverse quantizer (8) is equal to pixel number (n-1), as shown in (e). ) until IO
Full of W. Naturally, the output of the delay device (10) (1) and the output of the adder (9) (g) continue to output "128".

By the way, when the pixel number is n, an error occurs in the transmission process, and the output value (e) of the inverse quantizer (8) is originally "01".
If the output value (g) of the delay device αO is incorrectly set to “128” when it should be “128”, the output value (g) of the adder (9) will output “128”, the value at pixel number (n-1). After that, the output value (g) of the adder (9) changes to '256'.

Next, when the pixel number is (n+1), the output value (e) of the inverse quantizer (8) has returned to 10I, but the output value (e) of the delay device (1) is The output value ()) of the first adder (9) is 1, which is "256", and the output value (g) of the adder (9) is maintained at '25S'. "256" is then sent out from the output terminal (11).In the end, the error when the pixel number is n causes all subsequent values to be lost.This is error propagation.

Of course, it goes without saying that the pattern of such error propagation differs depending on the prediction method, but a description of cases of other prediction methods will be omitted.

Normally, when transmitting gage digital information, error correction IF encoding is applied as a measure against error occurrence, so that even if an error occurs on the transmission path or in the data processing process, the original correct data can be restored. However, the error correction ability of this error correction coding is at stake.Normally, it is effective against random errors, but has little effect on burst errors.On the other hand, when DPCM is used, Then, random errors become burst errors K due to error propagation, and the error correction ability is significantly reduced.

There is currently no measure to completely prevent this error propagation. However, as a way to reduce it to some extent, there is a method to reset the DPCM to the initial value at regular intervals.

(ll) A method of separately transmitting the true value before band compression with DPCM.

Examples include.

However, in the above-mentioned method, (a) the error propagation cannot be corrected and remains as it is to a large extent.

(b) DPCM control becomes complicated.

(Q) The prediction immediately after the above-mentioned (1) reset becomes incomplete.

This includes my father-in-law.

Therefore, the present invention executes band compression using the correlation between 14 adjacent pixels based on the same concept as the conventional DPCM, but since the receiving side does not perform complete i-integration as in the case of DPCM, errors may occur. Almost no propagation occurs, and DP
The purpose is to enable a simple hardware configuration instead of a complicated node like CM.

That is, in the present invention, a block is formed by a predetermined number of adjacent pixels, at least one pixel in the block transmits its pixel value, and other pixels transmit the difference from a predetermined pixel value in the block. The intended purpose was achieved by converting the differences into data, converting them into data, and transmitting them.

An embodiment of the present invention will be described below based on the drawings. Hey, now I'm going to reveal the principle of this invention with @ro1v.

In the case of DPCM, the number of pixels (181 as shown in Fig. 2) is predicted by the neighboring pixel 1131 (+41 or (I7), and the prediction error is quantized and transmitted at the same time.
The quantized prediction error is added to the predicted value and the result is stored. Further, subsequent pixels are predicted based on these stored values. Therefore, DPCM
When we are inverting the data, we have to keep accumulating data values from the distant past. As mentioned above, this is the root cause of error propagation.

On the other hand, in the case of the present invention, a predetermined number of adjacent pixels are divided into blocks, and the reference pixel is transmitted as is,
Other pixels are used to detect differences between neighboring pixels or with a reference pixel, quantize the difference, and transmit the detected difference.

The situation will be explained with reference to the pixel diagram shown in Figure 3.
In FIG. 3, Q2 to Q2 represent pixels, ω7) represents a block, and 1281 represents the n-th horizontal scan. This figure shows a case where a block is composed of five consecutive pixels during the same horizontal scanning period. For example, the value of pixel (24) is transmitted in its size (the value of the pixel input manually is not changed, it is The pixel to be transmitted will be referred to as 1 reference pixel 1). Then, between the pixels outside this reference pixel + 24) μ (23
125) and 弼 detect the difference with the reference pixel t241,
After performing appropriate quantization, the signals are transmitted. Of course, since the differences between adjacent pixels are concentrated at zero or near zero because the pixels have a strong correlation, it cannot be denied that band compression is possible.

Next, one embodiment of the present invention is shown in a block diagram in FIG. In Figure 6, 8!9) is the input terminal A1 (B).
is a series-parallel converter (denoted as 'S/P'), South 1 is a matrix, C (II~+341 is a differencer, ■~ appended ψ converter (denoted as IQ and I~'Q14, respectively), (41 parallel-to-serial converter (denoted as IP//S1), (41) is a transmission line, 142 is a series-to-parallel converter (denoted as Is/PI)
, 143~(461-digit inverse quantizer (Fuo IQ-, If
,,, denoted as Iζ1'), (511-digit matrix,
14η~sake is an adder, 1li21 is a parallel-serial converter cl
(denoted as □1), is the output terminal. The value of each pixel expressed in digital form is added to the serial-to-parallel converter (Serial to Parallel Converter) via the input terminal (9).If 8 bits are originally assigned to each pixel, then the parallel B bit pixel values are sequentially converted to serial to parallel converter (9). Added to parallel converter (30).Series parallel converter (30)
simultaneously outputs data for five consecutive input 8-bit pixels, that is, five pieces of 8-bit data in parallel.

The values of the five strokes output in parallel are respectively supplied to the example matrix. The pixel value applied to the matrix value (matrix value) is the pixel value of the pixel (221-6) in FIG.
Pixels in the figure (23112t'il and (A) are respectively led to differentiators (311 to +341 K), and pixel (24) is led to all differentiators 1311 to □□□ and a direct parallel-to-serial converter (41. , pixel (22
The difference between and pixel (24) is the pixel area in the subtractor G21 and the pixel (241) is the pixel (i) in the subtractor G21.
The difference between the pixel (24) and the difference between the inside of the pixel and the pixel (24j) is calculated in the subtractor (product).The subtractor 131
The outputs of 1 to 1341 are respectively subjected to predetermined digitization by the digitizer C(~(39), and then added to and subtracted from the parallel-serial converter 4(M).

A modulator and a demodulator are inserted before and after the parallel-to-serial converter (usually one transmission line), which converts the input data into a serial bit string and sends it out to the transmission line 141. (omitted in Figure 4).

The signal output from the transmission line C41) is converted to 1 at the serial/parallel converter 14.
Converts q-series data into parallel data, parallel-to-serial converter! 401 and applied to the matrix 611. The input value is restored to 16 (24) predetermined values in the 401 and applied to the matrix 611. 4 indicating which representative value was converted to a representative value.
It is sent as bit-structured data. Therefore, the inverse quantizers (43) to 14 perform the opposite operation.
It is one line, and from the data of d bit 41 which is manually input, it is determined which representative value is meaning [2], and the representative value is converted into a matrix 61) in the form of IT configuration.
Apply to. After all, dequantization! to ~ (The output value of Tsukuda is equal to the input value of the fan fan +161-s, and the output of the matrix +511 is composed of adders +471-15n, and the inverse quantizers 1431 to 146) The values are added to the reference pixel in the respective adders 1, 171 to 171.

In this way, the five input values of the parallel surface series converter (521) are respectively restored to values that are equal to or close to the five output values of the series/parallel converter (15).
From 21, five consecutive pixel values are sent out in an 8-bit configuration.

Needless to say, since the correlation between adjacent pixels is very high, the output values of the differentiators 311 to (34) are concentrated at or near zero. Therefore, here the reduction in the number of bits is f
iJ function, and bandwidth can be compressed. Series + 1 m column converter 10) and '421, parallel-serial converter - 4 (1) and (52) can be easily configured with a combination of registers and latches, Vessel 14:1
1~+461 f'iROM (Read○nly)
4emory) can be easily configured.

Next, the situation when an error occurs in the embodiment of the present invention shown in FIG. 4 is shown in LJ and ii+? below. , reveal. In the case of Fig. 4, 5 pixels constitute one block, 4 pixels are 4 bits, 1 pixel f
Since it is transmitted in 8 bits, 24 bits per block.
t configuration (it) it X 1 pixel + B bit X 1 pixel = 241) it) % average 1 pixel drawing 4.8 t + it assignment. Also, what is the relationship between error bits and error pixels?) If there is an error in the data of the reference pixel...an error occurs in all five pixels (11) If there is an error in data other than the reference pixel...
If only that pixel has an error, then the average number of error pixels in one block is equivalent to the average bit error of random errors in the transmission frame.

On the other hand, in order to compare with the case of conventional DPCM, in order to compare with this embodiment, suppose that a method is adopted in which the true value is transmitted every 5 pixels, and then DPCM is executed using this as the initial value. , 1 block consists of 5 pixels, 24 bits
(8 bits for the true value, 4 t for each of the other 4 pixels) it
Let us take as an example the case where it is composed of

(1) If there is an error in the initial value pixel data...an error occurs in all pixels (iD If there is an error in the n-th (n=2.3./i, 5) pixel...(5-n+1) ) Error occurs in pixel.

Therefore, if the average random bit error rate is increased by 4, the average number of error pixels within one block becomes 6.63 pixels, which is better in the present invention. Regarding the quality of the image obtained, the present invention is also superior, since in the conventional method prediction becomes incomplete for a while after transmitting the true value.

By the way, there are various ways to use the subtractor 1 between pixels within a block. Therefore, as another embodiment, a case in which the difference is taken in a different manner is shown in FIG. 5. In FIG.
Except for the connections of 1), (34), (37, and (), the configuration is the same as that in FIG.

In the embodiment shown in FIG. 5, pixel (24) is used as a reference pixel in the pixel diagram shown in FIG. Therefore, among the five values inputted in the matrix in Fig. 5, the true pixel (value 241 is passed directly to the parallel to serial converter (4G), and the value 22 is passed directly to the serial converter (4G);
11! ' is subtracted by the value of pixel I23), pixel (phantom value is subtracted by the value of pixel (C) by subtractor @,
5; When the subtractor is used, the value of the pixel I24) is subtracted, and the pixel (A) is subtracted by the value of the pixel 12Fil when the subtractor is used. The outputs of the subtractor C (I+' and C341' are quantized by the subtractor C and the subtractor C, respectively).
Tsukuda and 琲 are the same as quantization system in Figure 4) and 翶, and they do not necessarily have the same characteristics.The correlation is ge(, so it is rather a bit more specific.Receiving side First, the output values of the inverse quantizers 1441 and 1451 are added to the value of the reference pixel in the adder Nawate and t49, respectively, and then the output values of the adder Nawate and I49 are applied to the adder 14π and t49, respectively, and the inverse quantizer and the output value of 1461.

The present invention has been explained above with reference to two embodiments.
and inverse quantization characteristics ~ (As mentioned above, the characteristics of 461 should be set to appropriate quantization and inverse quantization characteristics depending on how to take the difference between each pixel. , the respective differentiators (31) (31) (32) (
Since the entropy of the values output from 33) (34) and (34) is different, it is better to set the number of output bits of each −鍿子加KI田〜(濶) according to the respective entropy.In the end, from the matrix ■ When each of the output difference values is compressed through the quantizers (1) to (2), the bit allocation to each difference value is not necessarily the same, but it is preferable to allocate it so that the compression efficiency is optimal.

Further, in the embodiments of FIGS. 4 and 5, as shown in FIG. 6C', one block is exemplified by five consecutive pixels within the same horizontal scanning period. Naturally, the block configuration method depends on factors such as the nature of the video signal (NTSG composite color video signal or black and white), sampling frequency (positional relationship of transmitted pixels), etc. If possible, it is also possible to consider a method of making it two-dimensional or using thinned-out pixels as a constructor.

Also, matrix rain) and (511 can be easily configured with a normal adder, but they can also be configured with ROM, @Ktft
It is not impossible to configure it with several large-capacity ROMs, including the solidifier and inverse quantization characteristic ~+4[il.

Further, since the reference pixel is related to compression efficiency and the number of error-generated pixels, it is usually a good idea to use the pixel at the center of the block as the reference pixel, but other pixels may be used as the reference pixel.

In addition, regarding the method of calculating the difference (matrix construction method), the efficiency of band compression and the number of error-generating pixels are almost contradictory.
Ru. A detailed explanation on this point will be omitted.

In the foregoing explanation, the effects of the present invention are remarkable in the case of a digital VTR, which is a highly specific transmission path.

As can be seen from the above description, the present invention enables relatively high-quality band compression similar to the conventional DPCM method, and also significantly improves error propagation, which is a fatal drawback of DPCM.
This paper proposes a band compression method with an extremely simple hardware configuration.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing a conventional example of DPCM, and FIG. 2 is a block diagram for explaining FIG. 1. ii A basic drawing, FIG. 3 is a pixel diagram for explaining the present invention in detail, FIG. 4 is a block diagram showing one embodiment of the present invention, and FIG. 5 is a block diagram showing another embodiment of the present invention. It is. (221-(26)...pixel, 侃η...block,
, +011421...Serial-to-parallel converter, +qn ~t
341+3trta41'...Difference 5, (a ~ +n
-Softener, hit (521...parallel-serial converter, 14~146...inverse quantizer, +471~F+fll
14π sake'...adder agent Yoshi Morimoto
Hiroshi Figure 1 Figure 2-27 Figure 4

Claims (1)

[Claims]] A block is formed by a predetermined number of adjacent pixels, at least one pixel in the block transmits its pixel value, and other pixels transmit the difference from a predetermined pixel value in the queen block. 1. A video signal band compression method characterized in that each of the differences is converted to 1- and the difference is quantized to 1- and transmitted. 2. The video signal band compression method according to claim 1, wherein the parsing characteristics are individually set according to the properties of a plurality of differences, and the bit allocation after the bits is weighted.