JPS6165574A - Picture signal coding system - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a multilevel image signal data compression method, and particularly to a method for encoding prediction errors with high efficiency when a predictive encoding method is used.

[Conventional technology]

There is a predictive coding method as a data compression method for multivalued images. This method predicts the currently input image signal X based on an already input image signal S that is in a predetermined positional relationship with the encoded pixel The error signal e between the signal X and the image signal X is encoded. Conventionally well-known DPC
M (Differential Pulse Cod
e Modulation) method, the predicted signal
Find it as a linear combination of . For example, when three pixels a, b, and c shown in FIG. 4 are used as S, the predicted signal X in linear prediction is determined by X=kl al 2 bl 3 c.

Here, kl and 2 + k3 are coefficients, and the magnitude thereof indicates the strength of the correlation of each reference pixel with respect to X.

The prediction error signal e is e=X−X Find it with

The probability of occurrence P of the prediction error signal e (el has the property of concentrating on small values around e=0 as shown in A short code is assigned to a prediction error that is easy to cause.By doing this, data is compressed efficiently using this DPCM method. -2487
Seven methods have been proposed.

The former method attempts to increase the compression ratio by utilizing the fact that the occurrence probability distribution of e changes depending on S. For example, if S has a probability of e attending school of 90% or more, then Sa (the subscript G indicates the GOOD state) is 90% or more.
If S is less than S, (subscript B indicates BAD state), then the probability of occurrence of prediction error e corresponding to S, SB is P (
elsc), P (elsB) are shown in Figure 7 fan, fb
Since it becomes like l, SC.

For the prediction error of SB, encoding is performed by preparing two code tables to adapt to each prediction error distribution shown in Fig. 8 (al, (bl). ).

L(e) in (b), LG (81,t,,
[el indicates the code length.

An example of a compression circuit for implementing this method is shown in the block diagram of FIG. 9, and its operation will be explained below with reference to the same diagram.

Here, the image signal has 16 values (bits/pixel) from 0 to 15. The image signal is input to a shift register 10 of 4 bits×(β+2) words. Here, β is the number of pixels per line, and the shift register 10 stores (6+2) pixel bone image signals, from which reference pixels a, b
, c are input to the predictor 11 and the mode signal generator 12. The predictor 11 outputs a predicted signal amount. From this and the coded signal X of the shift register 10, the subtracter 3 obtains a prediction error signal e. This prediction error signal e is input to the GOOD mode code table when the predicted hit rate is high and to the BAD mode code table when the predicted hit rate is low, and a code corresponding to the prediction error is output. On the other hand, the mode signal generator 12 determines whether the predicted state is GOOD mode or not.
A 1-bit signal M specifying whether the mode is D is output, and this code is selected. The selected code is input to the parallel-to-serial converter 17, where it is converted into a series of compressed codes and output.

Since this method performs encoding using variable length codes, it is relatively easy to implement the device. Moreover, since the code is selected according to the predicted state, a compression ratio higher than that of DPCM encoding can be obtained. However, since the gradation of one pixel is encoded with one variable length code, there is a drawback that the possible compression ratio is limited by the encoding method. For example, if the image signal is 1
When using 6 values (4 bits/pixel), the compression ratio cannot be higher than 4.

On the other hand, in the latter method, prediction errors are RL encoded in order to eliminate the restrictions imposed by the above-mentioned encoding method. This method uses nonlinear prediction to further increase the compression ratio. An example of a data compression circuit for implementing this method is shown in a schematic diagram in FIG. 10, and its operation will be explained below with reference to the same diagram.

Reference numeral 10 is a shift register which functions similarly to the shift register 10 in FIG. 9 above. In this method, a ranking generator 21 is used to obtain a ranking of how likely X is to appear as a prediction error under conditions a, b, and c from reference pixel signals a, b, and c and encoded pixel signal X. .

The rank generator 21 is created by determining the order of ease of producing X using several images for all combinations of a, b, and c. 4 bits (
Each bit eQ of the rank signal consisting of 1st to 16th ranks),
el, e2. e3 is set in the shift register 22. Here eQ is the least significant bit (LSB) of the order, 8
3 is the most significant bit (MSB).

The mode signal generation circuit 23 generates reference pixel signals a and b.

C and eo, el, e2. which constitute a ranking signal. Input e3 bits and select whether each bit of the ranking signal has a high probability of being 0 (GOOD mode) or a low probability of becoming 0 (BAD mode).
A mode signal indicating which mode is selected is output and set in the shift register 24. Here, M.O.

Ml, M2, M3 are each bit of the mote signal,
Each bit e of rank. +el+82+e
3(7) mode is shown. Next shift register 22.24
If the serial output of 24 is in the GOOD mode, the demultiplexer 25 is changed to the GOOD mode RL encoder 26.
Switch to . Also, if the serial output of the shift register 24 is in BAD mode, the demultiplexer 25 is set to BA
Switch to the D mode RL encoder 27.

By doing this, each bit of the order has 6000 modes.
The RL is counted and encoded in BAD mode. Next, the RL codes are rearranged in the output order by the output order control circuit 29 in the order used at the time of restoration, and are selected and output by the multiplexer 28.

This method has the advantage that a compression ratio close to the theoretical upper limit can be obtained because the GOOD mode RL can be long. However, this method requires counting RLs, requires two RL encoders that take into account the RL distribution of each mote, and requires control of the output order of variable-length RL codes. , the disadvantage is that the logic is complex and the hardware becomes large.

[Problem that the invention seeks to solve]

As mentioned above, conventional prediction error encoding methods have problems such as not being able to obtain a sufficiently large pressure t1d ratio, or when attempting to obtain a large compression ratio, the logic for controlling the output order becomes complex and the hardware becomes large. was there.

[Means for solving problems]

SUMMARY OF THE INVENTION An object of the present invention is to solve the above-mentioned problems and to provide an efficient image signal encoding method that can achieve a relatively large compression ratio with simple encoding.

This purpose is to improve the ranking of prediction errors in the present invention.
mode, the probability of occurrence of the first rank is particularly high (becomes,
Moreover, taking advantage of the property that this first rank appears consecutively,
Block the rank signal into a predetermined plurality of blocks, (1) When all the ranks in one block are the first rank, encode that block with an identification code "0° 1 bit," (2) ■ Block If it includes a rank other than the first rank,
ID code "1'" and then go to m ranks in the block.
This is achieved by encoding with a variable length code having a common rank for OD mode and BA mode.

[Effect]

In the present invention, the ranking of prediction errors is
For the GOOD mode, the ranks are coded together into multiple ranks, and a flag bit is added only when a rank other than the first rank is included in one block. is encoded with 1 bit, and in BAD mode, the prediction error is encoded without attaching a flag bit to each pixel, making it possible to obtain a relatively large compression ratio with simple encoding.

〔Example〕

FIG. 1 is a block diagram showing an example of a data compression circuit for implementing the image signal encoding method according to the present invention, and one embodiment of the present invention will be described below with reference to the figure.

In FIG. 1, shift register lO5 rank generator 21
．． The mode signal generator 12 is the same as the conventional one.

However, based on the combination of S, the mode signal generator 12 selects GOOD when the probability of occurrence of the prediction error of the first rank is high.
The BAD mode is set when the probability of appearance of the first rank is low, and a signal M instructing this is output. Next, the ranking signal e is inputted to the shift register 31 side by the demultiplexer 30 in the GOOD mode, and
In the mode, the data is written to the FIFO memory 33. Here, in this embodiment, an example will be described in which the GOOD mode ranking signals are divided into blocks every four.

The configuration of the shift register 31 is (number of bits of rank signal)
x (number of units of flocking), and in this embodiment, the rank signals are divided into blocks every four, so the structure is 4 bits x 4 words.

By repeating the above operation, when four rank signals to be encoded are stored in the shift register 31, the detector 32
It is detected whether all of these rankings are the first ranking.

If the rankings are all first, multiplexer 3
8 and outputs 1 bit of identification code “0”. If there is a signal other than the first rank among the rank signals of one block stored in the shift register 31, the multiplexer 38 is switched to output the identification code '1' 1 bit, and then the shift register 31 is Shift by one,
The ranking is entered into the code table 36 via the multiplexer 34 and the corresponding code is output. In the code table 36, as shown in FIG. 2, for example, a code is assigned to each rank. The codes corresponding to the above rankings outputted from the code table 36 are outputted by the parallel-to-serial converter 37 through the multiplexer 38 as a series of compressed codes.

The four ranks of the GOOD mode are encoded by repeating this operation of outputting the codes from the code table 36 four times.

On the other hand, the order of the BAD mode is encoded by the output order control circuit 35 in consideration of the order used during restoration. This is done, for example, as shown in FIG.

In the figure, the numbers inside circles indicate the order of encoding and output. GOOD mode ranking is 411? Since the block is encoded in blocks of j, if the first block of GOOD mode appears before the BAD mode, like the block of ■, then the BAD mode will be ranked after the block of ■ is output. ■, ■, ■ are output in order. In this way, the ranks of GOOD mode and BAD mode are output in the order in which they are used during restoration, so the ranks of GOOD mode and BAD mode are not confused during restoration.

The FIFO 33 is a temporary storage means for adjusting the output order of BAD mode rankings. Output order control circuit 5
If at least one G00 low mode rank is stored in the shift register 31, it refrains from outputting the BAD mode rank from the PIFO 33 until four GOOD mode ranks are stored. After outputting the four GOOD mode rankings through the multiplexer 34, the PIFO
All BAD mode rankings stored in 33 are outputted via multiplexer 34. If no GOOD mode order is stored in the shift register 31, BA
Similarly to the GOOD mode, the D mode ranking is outputted as a series of compressed codes via the code table 36 and the ps converter 37.

Although not specifically mentioned here, the restoration of this compressed encoded data is realized by a circuit that operates in the opposite manner to the above.

In the embodiment shown in FIG. 1, a ranking is obtained as a prediction error by non-linear prediction, but in this case, linear prediction is performed as shown in FIG. A differential signal may also be used.

〔Effect of the invention〕

According to the present invention, in the GOOD mode, a plurality of prediction errors are collectively encoded, so that a compression ratio higher than that of a method in which prediction errors are simply variable-length encoded can be obtained. Further, since the pre-2111J error is encoded with a variable length code and the same code table is used for both the GOOD mode and the BAD mode, the circuit configuration is simplified.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of a data compression circuit as an embodiment of the present invention, FIG. 2 is a code diagram showing an example of the assignment of codes to the ranks in the code table 36 of FIG. 1, and FIG. 1 is a diagram for explaining the order in which the GOOD mode order and the BAD mode order in the data compression circuit of FIG. 1 are controlled and output from the multiplexer 34 by the output order circuit 35. FIG. Figure 5 is a diagram showing the positional relationship of encoded pixels, Figure 5 is a diagram showing the probability of occurrence of a prediction error signal, Figure 6 is a diagram showing the code length assignment for this prediction error signal, and Figure 7 is a diagram showing the prediction error signal based on the prediction error rate. Figure 8 is a diagram showing the probability of occurrence of a prediction error signal when divided into GOOD mode and BAD mode.
FIG. 0 is a block diagram showing the configuration of a conventional data compression circuit. In the figure, 10.31 is a shift register, 11 is a predictor, 12.23 is a mode signal generator, 13 is a subtracter, 1
4, 15.36 is a code table, 16.28.34.38 is a multiplexer, 25.30 is a demultiplexer, 17.
37 is a parallel-to-serial converter, 21 is a rank generator,
22.24 is a register, 33 is a FIFO memory, 29.
35 is an output order control circuit. 112 Figure 113 Figure 4 Figure Myo-snoring－■−■−−−−１ ■−■−−−１−−− @Figure 5 Figure 6

Claims (1)

[Claims] A prediction rank of the actual pixel value of the encoded pixel is determined using a plurality of encoded pixels having a predetermined positional relationship with respect to the encoded pixel as reference pixels, and the predicted rank of the actual pixel value of the encoded pixel is calculated using the predicted rank. In a predictive coding method for a multivalued image signal that encodes a value, a first mode in which the predictive value of the encoded pixel has a higher predictive value than a predetermined value based on the pixel values of the plurality of reference pixels; It is determined for each pixel which of the second modes it belongs to which is lower than a predetermined value, and the prediction order of the first mode and the prediction order of the second mode are each divided into blocks for a predetermined number of blocks. If all the prediction errors in the block are 0, the block is encoded with a 1-bit predetermined identification code, and if all the prediction errors in the block are not 0, the block is coded with 1 bit different from the predetermined identification code. and a variable length code corresponding to each of the prediction errors of the block, and a prediction error in the block in the second mode is encoded with the corresponding variable length code. Encoding method.