JPH04294685A - Method and device for encoding color image signal - Google Patents

Abstract

PURPOSE:To perform the compressive encoding of data with high efficiency in the encoding of a color image where digital color signal data is compression- encoded, and accumulated, and original data is decoded and reproduced. CONSTITUTION:A prediction error value between the prediction value of a prescribed picture element value and an actual picture element value is found at a prediction error arithmetic part 2. Multilevelvalue/binary conversion is performed at a multilevelvalue/binary conversion part 3 so as to decrease an entropy rate by using a conversion table, and data with fixed length can be obtained. After that, arithmetic encoding is performed at an arithmetic encoding part 4, thereby, compression efficiency by the arithmetic encoding can be improved.

Description

[Detailed description of the invention]

0001

[Field of Industrial Application] The present invention relates to a method and apparatus for encoding color image signals, and in particular to a method and apparatus for encoding color image signals, and in particular for compressing and encoding digital color image signals input from cameras, photo input devices, etc. and storing them, and from the stored data. The present invention relates to a method and apparatus for encoding color image signals, which are suitable for image databases and image compression and transmission equipment, and which enable decoding and reproduction of original image signals.

0002

2. Description of the Related Art A reversible encoding method for reproducing the original data by compressing and encoding a digital signal of a color image, storing it, and decoding and reproducing it is known, for example, in the document "JPEG-8-R5". .2 draft document” (Joint
Photographic Experts Group
p, ISO/IEC JTC1/SC2/WG8,
There is a method (hereinafter referred to as the JPEG method) that is described in the SG-15 edition, published in 1989) and is being recommended for standardization.

[0003] The general encoding method of the JPEG system consists of three parts: prediction, multilevel/binary conversion, and arithmetic encoding. In the prediction unit, the n-th pixel (
The prediction error y(
m, n).

y(m,n) =x(m,n)−(x(m−1,
n) +x(m, n-1) )/2 Formula (1
)

As shown in the above equation, the prediction error y(m,n)
is the pixel value x(m,n) and the predicted value of this pixel value (x(
m-1, n) +x(m, n-1))/2.

Generally, the image signal x(m,n) has a strong correlation with surrounding pixels at each pixel position, so the above (1)
The prediction error y(m,n) obtained by the formula becomes data biased near zero.

Next, the multi-value/binary conversion section converts the prediction error y(m,n), which is multi-value data, into a binary data series. Specifically, a binary code with a smaller bit length is assigned to prediction errors that occur frequently (
(variable length encoding), the overall average code length is shortened as a result. The binary code assignment method is based on a decision tree. By inputting such a binary code sequence to the arithmetic coding section, further data compression is attempted.

[0008]

[Problem to be Solved by the Invention] In image encoding using the JPEG method as the prior art, the entropy rate of the data is already close to 1 at the time when the processing in the prediction unit and the multi-value/binary conversion unit is completed. In other words, the condition was close to the compression limit. In other words, the compression effect of image data is mostly achieved by the prediction section and the multi-value/binary conversion section, and the compression effect of the arithmetic coding section at the last stage is not fully utilized. There were problems that needed to be resolved. Furthermore, since the multilevel/binary converter uses a variable length encoding algorithm in which the length of the output code is not constant, there is a problem in implementation that the circuit configuration becomes complicated.

The present invention provides a color image signal encoding method and apparatus that improves the compression effect of an arithmetic encoder by converting prediction error values into fixed-length binary data and performing arithmetic encoding. The purpose is to

0010

[Means for Solving the Problems] In order to solve the above problems, the present invention first predicts the signal value at each pixel position of an image signal, and then calculates the entropy rate in descending order of the frequency of occurrence of the prediction error value. The prediction error is converted into an array of binary data using a conversion table consisting of fixed-length codes determined in the order of decreasing Features.

[0011]

[Operation] According to the present invention, due to the above-mentioned features, the multi-value/binary converter performs conversion that lowers the average entropy rate of the output code after multi-value/binary conversion, rather than compressing data. This makes it possible to enhance the compression effect in the arithmetic encoding section and, as a result, to enhance the overall compression effect. In addition, since the multivalue/binary conversion section processes a fixed length,
Subsequent encoding and decoding operations require only simple table reference processing, making it extremely easy to simplify the circuit and speed up the processing.

0012

DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, one embodiment of the present invention will be described in detail with reference to the drawings. Generally, color images are RGB, YUV,
It is handled as three or four color component signals such as CMYK. In this embodiment, as shown in the configuration example of the RGB signal encoding circuit in FIG. 2, a data compression circuit having a similar circuit configuration is provided in parallel for each color component signal. As shown in FIG.
The pixel data inputted from the pixel data is inputted to analog/digital converters 32, 33, and 34 through signal lines 51, 52, and 53, respectively, and is converted into analog/digital data. Outputs from the analog/digital converters 32, 33, and 34 are sent to the encoding circuit 35 through signal lines 54, 55, and 56.
, 36, and 37, and each RGB color component signal is compressed and encoded. Compression-encoded image data is sent to output terminals 41 and 4 through signal lines 57, 58, and 59.
2 and 43, respectively.

The encoding circuit shown in FIG. 1 provided for each component signal in FIG. 2 will be explained in detail below.

A video signal of an image photographed by an image input means such as a camera is input from a signal input terminal 11 after analog/digital conversion, and is sent to the input image memory 1 through a signal line 21 and stored therein.

The input image memory 1 is a memory in a two-dimensional array of M rows and N columns, with M vertical pixels and N horizontal pixels corresponding to the number of pixels of a digitized image. In this example, each pixel value is stored as an 8-bit integer value (0 to 255).

The image data stored in the input image memory 1 is sent to the prediction error calculating section 2 through a signal line 22. The prediction error calculation unit 2 calculates all M pixels in the vertical direction and N pixels in the horizontal direction of the image data sent from the input image memory 1 (MxN
) For a pixel, predict a specific pixel value using neighboring pixel values, and calculate the difference between the two, that is, the prediction error between the predicted pixel value and the actual pixel value. For example, when using the two pixels above and to the left of the pixel, this calculation is performed using the following (
2) Executed according to the formula.

y(m,n) = x(m,n) −(x(m−
1, n) +x(m, n-1) )/2
(2) Formula

Here, x(m, n) is the image data at the n-th pixel point on the m-th line in the input image memory 1 read into the prediction error calculation unit 2 via the signal line 22. , x(m-1,n) is the image data at the n-th pixel point on the (m-1)th line in the input image memory 1 (corresponding to the pixel value next to the above pixel). What is read into the prediction error calculation unit 2 via the signal line 22, x(m, n-1) is the input image memory 1
Image data at the (n-1)th pixel point on the mth line (corresponding to the pixel value to the left of the pixel)
are read into the prediction error calculation unit 2 via the signal line 22. However, if m is "zero" and n
When calculating the prediction error y(m,n) when is not "zero", the following (3
) Calculate the prediction error according to the formula.

y(m,n) = x(m,n) −x(m,n
-1)
(3) Formula

Further, when m is "zero" and n is "zero", the initial value x(0,0) of the pixel data itself is used as the initial value when calculating the prediction error.

[0021] Through the above prediction calculation, in the case of natural images in which there is a strong correlation between pixels, it is almost common that the prediction error value whose absolute value is closer to "zero" appears more frequently regardless of the type of image. do. The multi-value/binary conversion described below applies this property.

[0022] Equations (2), (3), and (4) above show an example of prediction using two neighboring pixels of the pixel in question, and the number of pixels used for prediction or the coefficient value can be increased or decreased. It is also possible to change the prediction accuracy by making changes.

[0023] The multi-value/binary conversion unit 3 includes the prediction error calculation unit 2
3 through the signal line 23, and using a multi-value/binary conversion table, which will be described later and shown in FIG. A binary code as shown in the 'Binary code' column is output, which corresponds to the prediction error data as shown in the 'Binary code' column.

The multivalue/binary conversion table, as shown in FIG. 3, consists of a series of combinations of 'multivalues' and corresponding 'binary codes'. In this example, each pixel value is 8
Since the prediction error can also be expressed in 8 bits, the multi-value/binary conversion table is composed of 256 elements. This table defines the binary code corresponding to the 'binary code' column, which corresponds to the prediction error data corresponding to the 'multi-value' column. The multi-value/binary conversion table is created such that the more frequently occurring prediction error data is associated with a binary code having a lower entropy rate, the entropy corresponding to the 'entropy' column. In addition,
Entropy is defined by the following formula. Here Pi
is the probability of occurrence of i.

[0025]

[Math 1]

In this embodiment, the most frequently occurring prediction error (multi-value) is "zero", and "0000 0000
” is given. Below, the prediction error values "1" and "-1" with the second highest probability of occurrence are "0000 0001" and "0000 0010," respectively.
'' is given, and a conversion table is created by sequentially assigning codes in the same way.

By this conversion, a fixed length binary code with a low entropy rate is uniquely associated with a frequently occurring multi-value. As a result, the entropy rate of the entire binary code obtained by converting image information is reduced, and the compression efficiency by arithmetic coding is improved.

[0028] The arithmetic encoding unit 4 sequentially reads the binary code data series from the multilevel/binary conversion unit through the signal line 24, and performs data compression (entropy encoding) operation according to the arithmetic encoding algorithm described later. .

[0029] Arithmetic coding operations can be performed using, for example, the L - By R-type arithmetic coding algorithm.

The encoded data arithmetic encoded by the arithmetic encoder 4 is output to the output terminal 12 through the signal line 25.

The above description is of the encoding (data compression) circuit. On the other hand, decoding the image (reproducing the original data)
In the above, each compression encoding operation is performed in the reverse order.

First, the coded data sequence is input to an arithmetic decoding circuit, and an arithmetic decoding operation is performed to obtain a binary code data sequence. This arithmetic decoding operation is described, for example, in the literature “
The LR type arithmetic decoding algorithm described in "Image Coding Algorithm (Entropy Coding)" (written by Shigeo Kato, Journal of the Television Society, December 1989 issue, page 64) is used.

Next, by using the table in FIG. 3 in the opposite way to the encoding calculation, the prediction error data (FIG. 3) corresponding to the binary code data series (corresponding to the 'binary code' column in FIG. (equivalent to the 'multi-value' column).

Finally, pixel values are restored from the prediction errors decoded in this way. The nth pixel value x(m,n) on the mth line is calculated by the inverse operation of equation (2), and the prediction error y(m,n) and the already decoded pixel value x(m-
1, n) and x(m, n-1), the following (
5) It is obtained by the calculation shown in the formula.

x(m,n) = y(m,n) +(x(m-
1, n) +x(m, n-1) )/2
(5) Formula

Here, the pixel value x(m,n) when m is "zero" is calculated according to the following equation (6) since there is no decoded data adjacent above.

x(m,n) = y(m,n) +x(m,n
-1)
(6) Formula

Furthermore, when n is "zero", the pixel value x(m, n) is calculated according to the following equation (7) since there is no decoded data on the left.

x(m,n) = y(m,n) +x(m−1
, n)
(7) Formula

[0040] Furthermore, the pixel position at the upper left end of the entire screen, that is, the pixel value x (m, n) when n is "zero" and m is "zero", is the pixel value x (m, n) of the prediction error value data of the relevant pixel. Let the initial value y(0,0) itself be the pixel value.

[0041]

[Effects of the Invention] Since the present invention has the structure as described above, it exhibits the effects as described below. The signal value at each pixel position of the image signal is predicted and the difference from the actual signal value, that is, the prediction error value, is calculated. Next, the prediction error values are determined in order of decreasing entropy rate in descending order of appearance frequency. The prediction error is converted into a binary data array using a conversion table consisting of fixed length codes, and the binary data array is data compressed by arithmetic coding. Therefore, it is possible to enhance the compression effect by the arithmetic encoding section. Furthermore, since multivalue/binary conversion requires only simple table reference processing and fixed length processing, it is possible to simplify the circuit and speed up the processing.

[Brief explanation of the drawing]

FIG. 1 is a schematic configuration diagram of a data compression circuit according to a color image encoding device of the present invention.

FIG. 2 is a diagram showing the overall configuration of a color image encoding device according to the present invention.

FIG. 3 is a diagram showing a conversion table from multi-value data to binary code data stored in the multi-value/binary conversion unit of FIG. 1;

[Explanation of symbols]

1 Input image memory 2 Prediction error calculation unit 3 Multi-value/binary conversion unit 4 Arithmetic coding calculation unit 11 Input terminal 12 Output terminal 31 Cameras 32, 33, 34 Analog-digital conversion unit 35,
36, 37 Encoding circuit 41, 42, 43 Output terminal

Claims (4)

[Claims] 1. A color image signal encoding method for compressing and encoding an input digital color image signal, the method comprises predicting a signal value at each pixel position of the image signal and calculating the signal value at each pixel position. and the predicted signal value, convert the prediction error value into fixed-length binary data according to its frequency of occurrence, and arithmetic encode the converted binary data. A method for encoding a color image signal, characterized in that data is compressed by compressing the data. 2. The conversion of the prediction error value into the fixed length binary data is characterized in that the prediction error value that occurs more frequently is associated with a binary code having a lower entropy rate. 1. The method for encoding a color image signal according to 1. 3. A color image signal encoding device for compressing and encoding an input digital color image signal, the device predicting a signal value at each pixel position of the image signal and calculating the signal value at each pixel position. and a prediction error calculation means for calculating a prediction error value between the predicted signal value and the predicted signal value, and converting the prediction error value calculated by the prediction error calculation means into fixed-length binary data according to the frequency of occurrence thereof. A color image signal encoding device comprising: binary data converting means; and encoding means for compressing data by arithmetic coding the binary data converted by the binary data converting means. 4. The code of the color image signal according to claim 3, wherein the binary data conversion means associates a binary code with a lower entropy rate with the prediction error value that occurs more frequently. conversion device.