JPH0265372A - Encoding system for picture - Google Patents

Abstract

PURPOSE:To acieve efficient encoding to various pictures by dynamically changing a parameter, with which an attention picture element is forecasted on the basis of plural picture elements near the attention picture element, in correspondence to the rate of coincident and decidence between the forecasting picture element and attention picture element. CONSTITUTION:A group signal G300, which is outputted from a forecasting circuit part, and an attention picture element signal D301 are respectively inputted to a group generation frequency counter 90 and an inferior symbol counter 91. The generation frequency counter 90 counts how many times the condition of a group is generated by groups. When the count value of each group goes over a setting value S302, an updating signal 303 is outputted. The number of inferior symbols, which are generated from the generation of the preceding updating signal to the generation of the next updating signal, is counted by groups in the counter 91 and outputted as a count value lc304. The output of an LUT92 is latched by the updating signal 303 from the generation frequency counter 90 and a condition is updated to the new condition.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an image encoding device for encoding an image.

[Conventional technology]

A facsimile machine, which is a typical example of a conventional still image communication device, sequentially scans images and
A method is used in which images are encoded and transmitted using H, MR encoding, etc.

With this method, it is necessary to transmit all of the image data in order to grasp the overall image of the image, which takes a long time to transmit, and requires a quick image judgment such as the image database service Videotex. It was difficult to apply this method to image communication services.

Therefore, as an encoding method that can be applied to this service, for example, the pixel of interest is predicted from a plurality of peripheral pixels in the vicinity of the pixel of interest to be encoded, such as arithmetic coding, and the predicted pixel and the actual pixel of interest are It has been proposed to encode images based on the coincidence and mismatch of .

According to this, it is possible to achieve simpler and more efficient encoding than conventional MH, MR encoding, etc. configurations.

[Problem that the invention is trying to solve] However,
In such encoding, the parameters necessary to predict the pixel of interest based on surrounding pixels are determined in advance using a standard image, and the prediction accuracy rate may be low for images with different characteristics from the standard image ( , the encoding efficiency deteriorates, and sometimes there may be a problem that the encoded data becomes longer than the original data.

[Means for solving problems]

The present invention has been made in view of the above points, and predicts the pixel of interest by referring to multiple pixels in the vicinity of the pixel of interest to be encoded, and performs encoding based on the coincidence or mismatch between the predicted pixel and the pixel of interest. The present invention provides an image encoding method in which a parameter for pixel prediction is changed based on the proportion of coincidence and mismatch between a predicted pixel and a pixel of interest.

〔Example〕

FIG. 1 shows an embodiment of an encoder to which the present invention is applied.

First, original image data 1, which is binary data indicating white/black of each pixel of the original image, is stored in the first frame memory IO1. Next, the 10th filter 11 performs smoothing processing, and the first comparator 12 again inputs the threshold value T1 to the binarizer 12, and these values are used to determine the required image quality,
and L) is determined by the controller 9 based on the encoding efficiency. Next, vertically by the first subsampling 13,
A signal 101 representing an image thinned out horizontally by 4 is generated.

This signal 1011'! It is stored in the second frame memory 14.

On the other hand, a signal 102 representing the image A is generated from the signal 101 by the 20th-noise filter 15, the second comparator 16, and the second subsampling 17 in the same way. This signal 1
02 is stored in the third frame memory 18.

Further, as described above, the signal 102 is processed by the 30th filter 19, the third comparator 20, and the third subsampling 21 to obtain a signal 103 representing a 1/8 size image. This signal 103 is stored in the fourth frame memory 22. In addition, the second. The 30th-pass filter has parameters C2 and C3, respectively. The third comparator has a threshold T
2. T3 is input by the controller 9.

Further, the signal 103 is encoded by the first encoder 23 and transmitted as a first stage signal 104.

The second encoder 24 performs encoding while referring to the data in the fourth frame memory 22 and the third frame memory 18.
A second stage signal 105 is transmitted.

Similarly, the third encoder 25 performs encoding while referring to the data in the third frame memory 18 and the second frame memory 14, and transmits the third stage signal 106. Similarly, in the fourth encoder 26, the second frame memory 14 and the first
Encoding is performed while referring to the data in the frame memory 10, and a fourth stage signal 107 is transmitted. As described above, the first to fourth encoders encode image data having mutually different resolutions.

In this way, by encoding and transmitting the image data from the first stage to the fourth stage in order from the lowest resolution image data, the receiving side can quickly identify the entire image before transmitting the high resolution image data. , If that data is unnecessary, it is possible to stop the transmission of subsequent high-resolution image data.

This reduces communication time for unnecessary data and enables efficient image communication services.

FIG. 2 shows an embodiment of a decoder to which the present invention is applied.

The first stage signal 104, which is reduced to 1/8 by the encoder, is decoded by the first decoder 27 and sent to the fifth frame memory 31.
is memorized. After this signal is converted into high-resolution data by the first interpolator 35 through an eight-fold interpolation process, the controller 42 switches the selector 38 and stores it in the video memory 39. The video memory 39 is composed of a two-port memory that can input and output in parallel. Therefore, images decoded on the receiving side are displayed on the monitor 40 at any time.

Further, the second stage signal 105 encoded by the decoder based on the I/4 and 1/8 images is stored in the fifth frame memory 3.
The data is decoded by the second decoder 28 while referring to the data of No. 1, and is stored in the sixth frame memory 32. Further, this data is subjected to quadruple interpolation processing by the second interpolator 36, and is stored in the video memory 39 by switching the selector 38.

Similarly, in the encoder, the third stage signal 106 encoded based on the images of % and 〃 and the fourth stage signal 107 encoded based on the original image and the image of 3. Fourth decoder 29. 30 and 7th. After being decoded by the eighth frame memory 33, 34 and the third interpolator 37, it is stored in the video memory 39 and displayed on the monitor 40.

On the other hand, the signal of the eighth frame memory 34, which is the decoded image signal of the fourth stage, is output to the printer 41 to obtain a hard copy.

FIG. 3 shows the first encoder used in the encoder shown in FIG. 2nd゜3rd
The filter coefficients of a 3x3 pixel size low-pass filter used as a 0-bus filter are shown, with the weighting coefficient of the center pixel being C, the weighting of 2 for the 4 pixels closest to the center pixel, and 1 for the next closest pixel. The coefficient is given.

This changes the value of the center pixel to Dij (1=1~M
.

j=l-N: M, N are horizontal and vertical pixel sizes), then the average density W is: W = (D+-1, H-+ + 2D; 1-+ +
Di++1-+ + 2D+-+1 + CDI+ +
2D+++1+ D +-+g++ +2 D +1
++ +D; ++, H).

1st. Second. In the third comparator, this value W is associated with a correspondence relationship such as a threshold value.

FIG. 4 is a block diagram of a low-pass filter and a comparator that perform the above operations. The input signal is the latch 44a,
b and c are each held with a delay of one pixel clock. In addition, input signals delayed by one line are held in the line memories 43-a and 43-b, respectively, and the latches 44-d and 44-e
, f, and latches 43g, h, and i, signals whose pixel positions correspond to those of latches a, b, and c are obtained. As a result, the pixel data shown in FIG. 3 is obtained. Latch 44a.

The output signals from c, g, and i are summed by an adder 45a, and multiplied by a constant (x1) by a multiplier 46a.

Further, the output signals from the latches 44b, d, f, and h are summed by an adder 45b, and multiplied by a constant (×2) by a multiplier 46b. Further, the output signal from the latch 44e, which is the median value, is multiplied by a constant (×C) by the multiplier 46c. The value of C can be set using the external controller 9, but the standard value is C=4.

The output signals of the multipliers 46a, b, and c are summed by an adder 47, and then compared with a threshold value T by a comparator 48. get the signal. This threshold value T is also controlled by an external controller 9.
However, as mentioned above (as a standard value T=
It takes the value of (12+c)/2.

FIG. 5 is an explanatory diagram of the subsampling operation used in the encoder shown in FIG. By extracting pixel data indicated by diagonal lines in the figure at every other timing in the main scanning and sub-scanning directions, a sub-sampled image of A size (2 in area) is formed. This can be easily achieved by adjusting the latch timing of image data.

Next, encoding by the encoders 23 to 26 shown in the first diagram will be explained. In this example, encoding is performed using arithmetic codes to predict the value of the pixel of interest from surrounding pixels, and the symbol when the predictions match is the dominant symbol (1), and the symbol when the predictions do not match is the dominant symbol (0). , and let p be the probability of occurrence of an inferior symbol, and refer to "Signal processing for this information" by Nikkan Kogyo.

In other words, if the binary arithmetic code for the code sequence S is C(S) and the auxiliary amount is A(S), then encoding proceeds by the arithmetic operation of A (nu l I!) = o, lt・t. It is something.

By approximating p(S)=2''.'s), multiplication can be completed by shifting only binary numbers.Q is called the skew value, and by changing this parameter, the arithmetic It becomes possible to use codes dynamically.

For decoding, the binary signal sequence S is S'xS', and when S' is restored, C(S) and C(S' ) + A(S' O) are compared, and C(S)>C(S')+A(S'O) then x
=1. Otherwise, decode as X=O.

In this embodiment, dynamic encoding is realized by dynamically determining the above-mentioned skew value Q and the dominant symbol m (or the inferior symbol I) based on previously encoded data. .As the encoding for this method, Q
, m is determined in advance from a standard image, and the two types have different characteristics.

A configuration for achieving the above encoding will be described below.

FIG. 6 shows the second part of the first diagram. Third. FIG. 3 is a block diagram of a circuit portion that predicts a pixel of interest in a fourth encoder.

The frame memory A51 is a memory in which at least one screen worth of image data to be encoded is stored. Further, the frame memory B52 stores at least one screen worth of image data that is subsampled to % of the image in the frame memory A51, which is the image sent one stage earlier. Each memory is composed of two-dimensional memory. X
If the address clock is φ1 and the X address clock is φ2, frame memory A51 has φ1. φ2
, %φl+ of each frequency is stored in the frame memory B52.
V2φ2 is given, which causes frame memory B5
For one pixel of frame memory A51, the pixel of frame memory A51 is 2×
Corresponds to 4 pixels of 2.

Each data is stored in line memories 53a, b and 5
At 4a and 4b, the data is delayed by l lines and is input to latches 55 and 56. In this latch, data delayed one pixel at a time is held.

When the output of each latch is correlated with the pixel position in Fig. 7, the pixel of interest (*) is latch 55'', and similarly No.
3 is latch 55hSNo, 4 is latch 55f, No.
5 is the latch 55e, No. 6 is the latch 55b, No.
7 is the output of the latch 55d.

In addition, the pixel position in FIG. 8 is No. 8 is the latch 56e1N.
o, 9 is latch 56f, No, IO is latch 56d,
No.

11 is the latch 56c, No, I2 is the latch 56h, N
o, 13 is the output of the latch 56g, and No, 14 is the output of the latch 56i.

Note that pixel N098 in FIG.
A 2-bit pixel position signal 59 for identifying four states (upper right, lower left, and lower right) is outputted by a counter 60 to a clock φ1. φ2
Generate based on.

The pixel position signal 59 and the pixel signals from the latches 55 and 56 are input to an LUT (look-up table) 57, and a group signal G300 indicating the group number of each state (pattern) is output.

This grouping is done based on the following three policies.

(1) Group patterns with similar predicted match rates.

(2) Group patterns with similar pattern shapes together.

(3) Summarize characteristic image patterns.

First, the predicted matching rate k for each state (pattern) is determined using a standard image according to method (1), and the image is divided into N groups based on the value of k. Among them, the predicted pixels (mps) are divided into NX2 groups with white ones as 0 and black ones as 1.

Next, in line with the above policy (2), we extract a pattern in which all the pixels marked with a circle have the same value (the states of other pixels are not considered) from the predicted pattern as shown in Figure 12(a). and group A.

Next, a pattern in which all pixels have the same value as shown in FIG. 6(b) other than group A is set as group B.

In addition, in accordance with the above policy (3), a pattern like the one shown in Figure (C), which tends to characteristically appear in dithered images etc. (pixels marked with ○ are the same value and pixels marked with the opposite value), is ,
Those other than group B are grouped as C group. Also, A,
Patterns other than B and C are considered D grooves. FIG. 11 shows such grouping. G.
, G2. ..., G8N becomes the group number of each pattern. Moreover, k+ (i=l−n) is the probability at the division point. It goes without saying that the number of groups and the similar pattern shapes shown here are not limited to those of this embodiment.

In this way, by grouping not only based on predicted matching probability but also based on pattern similarity, separation of characteristic patterns, etc., it becomes possible to adaptively encode images other than standard images.

Note that the prediction circuit in the first encoder 23 shown in the first figure is the sixth one.
In the illustrated configuration, line memories 54a, b and latches 56 a-i process the output of frame memory B52.
and the counter 60 is removed.

FIG. 13 is a block diagram of a dynamic adaptive circuit for dynamically changing the skew value Q and the inferior symbols in the encoder. Group signal G300 and target pixel signal D301 output from the prediction circuit section shown in FIG.
are input to the occurrence frequency counter 90 and inferior symbol counter 91 of the group, respectively. These counters are provided with as many internal counters as the number of groups so as to perform counting operations for each group, and are switched by a group signal G.

The occurrence frequency counter is 90, which counts how many times the state of that group has occurred for each group, and when the count value of each group exceeds the set value 5302, an update signal 3 is sent.
Outputs 03. The number of inferior symbols generated from the generation of the previous update signal until the generation of the next update signal is counted for each group by the counter 91 and output as a count value 1c304. In other words, counter 9
Based on the output of 1, it is indicated that lc out of S occurrences of a certain group state are inferior symbols.

In the following explanation, the state of 5=16 will be explained as a representative.

The LUT 92 includes an inversion signal 306 and a zero count (CT) 307 that instructs to invert the next encoding parameter QG 305 and the previous mps (dominant symbol) value in response to the occurrence of lc inferior symbols. The data referred to is stored in advance.

Zero count means inferior symbol I! It is a value that represents how many times in the past there was a state of O in which c did not occur among three,
This represents the number of consecutive occurrences of the dominant symbol. In other words, in principle, if the initial setting is CT=0,
When the S width lc is 0, it is updated to CT=1, and if it continues two or three times, CT=2. It has been updated to CT=3.

FIG. 14 shows an example of the LUT 92.

The initial state is set to CT=O, and a new value of Qr, . . . the next CT is determined based on each value of lc. For example, CT=O and I! When c=o, Qc=4. CT=1. Next time the update signal 303 comes, CT=1 and j!
When c=, Qq=5. CT=2.

Also, when CT=O and Ac=1, Qa=4, CT=1
will be updated.

The calculation formula for creating this table is lc Qc s -1og23x(.T+1)2"CT# (2) Formula expresses the probability of occurrence of an inferior symbol when 8 states continue (CT+1) times in a power of 2. The exponent part when approximated is Qc.

In addition, in equation (3), CT calculates the generation of inferior symbols by 172
When it is assumed that °', the number of eight 1c=o pairs is recalculated. Since 2''-0 is the number of dominant symbols, the value divided by S is the CT.

Also, CT=O and j! When c > -+ 1, it is treated as a special case, and the mps inversion signal is output, and the value of the conventional inferior symbol is inverted (that is, it is set to 0#1).The subsequent state is CT=O.

In cases other than Rc sin-1, the encoding process is performed normally without changing the inferior symbol.

In FIG. 13, the latch 93 is a conventional QG305, m
Where the ps inversion signal 306 and Cr2O2 are held,
By the update signal 303 from the occurrence frequency counter 90, L
The output of the UT92 is latched and updated to the new state.

The input of LUT92 is the count signal ZC of inferior symbols.
and CTTa 205 indicating the number of previous dominant symbols and updated according to the table of FIG.
CT and, if necessary, an m ps inverted signal is output. The mps holder 95 holds the dominant symbol (0 or l) used in encoding up to now, and this state is inverted by the mps inversion signal.

The mps/i'ps signal that is the output of the holder 95 is sent to the inferior symbol counter 91, and the inferior symbol counter 91 counts the pixel data D corresponding to the inferior symbol represented by this mps/lps signal as an inferior symbol. .

Encoding will be performed using the Qc and mps/Tps determined here.

FIG. 9 is a block diagram of an arithmetic encoder. If Qc305 from the latch 93 shown in FIG. conversion data 1
92 is obtained.

In this way, the skew value Q which is the encoding parameter
By dynamically determining and dominant symbols based on previously encoded data and performing encoding operations, efficient encoding can be achieved even for various images with different characteristics from standard images.

FIG. 10 shows the first part shown in the second figure. Second. Third. 4th decoder 2
7.28.29.30 block diagram. Similarly to the encoder, the decoder 191 performs a decoding operation using the inferior symbols LPSd 197 and the received data 195 from FIG. 6 and FIG. 13 to obtain decoded data 196.

Another way to determine the contents of the UT92, f shown in Figure 13, is to calculate lc/ from the number of inferior symbols lc in the S pixel.
There is also a method of finding S and determining a new Qc from the relationship shown in FIG. The initial value is Qc=1, and Qc is updated according to the value of lc/S. Q updated from second time onwards
c and j! Qc is determined sequentially using c/S. The value QG' at the time of updating is calculated using the following calculation formula and stored in a table. Also, Q=1, pc/s
>'/z 500 inverts the dominant and inferior symbols.

FIG. 16 shows an example of a dynamic adaptive circuit using this LUT, in which the Qc signal 305 is input to the LUT 92 and the Qc to be updated is determined (.

Further, in this embodiment, a dynamic arithmetic code has been described as the encoding method, but the present method can be applied to other encoding methods that use dynamic parameters.

〔Effect of the invention〕

As described above, according to the present invention, parameters for predicting a pixel of interest based on multiple pixels in the vicinity of the pixel of interest are dynamically changed depending on the proportion of coincidence and mismatch between the predicted pixel and the pixel of interest. It is possible to achieve efficient encoding for images.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an embodiment of an encoder, FIG. 2 is a block diagram of an embodiment of a decoder, FIG. 3 is a diagram showing coefficients of a low-pass filter, and FIG. 4 is a block diagram of an embodiment of a low-pass filter. , FIG. 5 is an explanatory diagram of subsampling, FIG. 6 is a block diagram of the prediction circuit, FIG. 7 is an explanatory diagram of reference pixels on the code plane, and 8th rI! J is an explanatory diagram of the reference pixel of the previous image, FIG. 9 is a block diagram of the arithmetic code encoder, FIG. 10 is a block diagram of the arithmetic code decoder, and FIG. 11 is a table for grouping. FIG. 12 is a diagram showing an example of a characteristic pattern; FIG. 13 is a block diagram of a circuit that dynamically determines encoding parameters Q and l; FIG. 14 is a diagram for determining parameters. 15 is a diagram showing an example of a table used in the second embodiment; FIG. 16 is a diagram showing the second embodiment; 51 is a frame memory A; 52 is a frame memory B 157 is LUT,
90 is an occurrence frequency counter, 91 is an inferior symbol counter, and 92 is an LUT.

Claims (1)

[Claims] An image encoding method that predicts a pixel of interest by referring to a plurality of pixels in the vicinity of the pixel of interest to be encoded, and performs encoding based on a match or mismatch between the predicted pixel and the pixel of interest, the method comprising:
An image encoding method characterized by changing parameters for pixel prediction based on the proportion of coincidence and mismatch between a predicted pixel and a pixel of interest.