JP2001054138A - Color image coder, method, color image decoder and method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain unified coding where reproduced image quality can optionally be controlled by receiving an RGB signal, obtaining intermediate components by the mean value of the RGB signal, the difference between G and R, and the difference between B and G, applying prediction processing to each of output components obtained from the intermediate components and applying entropy coding to the output of the prediction processing. SOLUTION: A quantizer 2 quantizes the outputs O1-O3 of a reversible color transform device 1 to respectively obtain outputs O'1-O'3. A prediction processing section 3 holds the output O'1 right in front and outputs the difference from a new output O'1 to an entropy coder 4 and uses the new output O'1 for succeeding prediction. Although the original image cannot be completely reproduced through quantization, the entropy of an output signal of the quantizer is reduced and the compression rate can be enhanced. A quantization step size controller decreases the entropy of the output signal or the power of the reproduced error under a high speed condition where distortion generated in a decoded image is suppressed within ±p with respect to the pixel value of the original image.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus and method for encoding a color image which can control the quality of a reproduced image, and an apparatus and method for decoding a color image.

[0002]

2. Description of the Related Art Irreversible, which is used for color image coding,
Models of lossless and quasi-lossless compression are shown in FIGS.

FIG. 9 shows a configuration of color still image coding in the case of lossy compression. In the figure, reference numeral 30 denotes a color converter, 31 denotes a DCT converter, 32 denotes a quantizer, 33 denotes an entropy coder, and 34 and 36 denote a thinning processing unit.
Reference numerals 5 and 37 denote color difference signals (Chrominance) 1 and color difference signals 2, respectively.
Corresponding to the luminance signal (Luminance) (3
1, 32, 33).

The supplied RGB signals are supplied to a color converter 3
At 0, it is converted into a luminance signal and two color difference signals, and the luminance signal is DCT transformed, quantized and entropy coded. The two color difference signals are subjected to a thinning process, and thereafter, the same process as the luminance signal is performed.

FIG. 9 shows a JPEG (Joint Photographic Ex).
perts Group), which is a method adopted in DCT-based coding, etc., and is implemented by truncation and rounding errors due to finite word length in color conversion / inversion and DCT conversion / inversion, quantization distortion, thinning out of color difference components, etc. Is always added to the color image output reproduced by. The color conversion employed in this method is RGB → YUV
There is a conversion. This conversion is performed by the following matrix operation. Y corresponds to Luminance, U corresponds to Chrominance1, V corresponds to Chrominance2

[0006]

(Equation 3)

The inverse conversion from the YUV signal to the RGB signal is given by the following equation using the inverse matrix of the above matrix.

[0008]

(Equation 4)

FIG. 10 shows a configuration of color still image coding in the case of lossless compression. In the figure, reference numeral 40 denotes a predictive coding unit, 41 denotes an entropy coder, and 42 and 43 denote processing units that perform the same processing (40, 41) on the signal R corresponding to the signal G and the signal B, respectively. ing.

Each of the supplied RGB signals is subjected to a process of predictive encoding, and then entropy encoded.

FIG. 10 shows a method employed in the lossless mode of JPEG. Since no error occurs in a prediction operation including addition and subtraction, a color image output reproduced by a decoder has no distortion.

FIG. 11 shows a configuration of color still image encoding in the case of quasi-lossless compression. In the figure, reference numeral 50 denotes a color conversion unit, 51 denotes a predictive coding unit, 52 denotes an entropy coder, and 53 and 54 process a luminance signal corresponding to the chrominance signal 1 and the chrominance signal 2, respectively (51, 52). Represents a processing unit that performs the same processing as that described above. 55 is a level adjustment unit,
56 denotes an inverse color conversion unit, and 57 denotes an error evaluator.

The supplied RGB signals are supplied to a color conversion section 50.
Is converted into a luminance signal and two color difference signals. Each of the three signals passes through a level adjustment unit 55 and an inverse color conversion unit 56, and is led to an error evaluator 57. The error evaluator 57 also receives the supplied RGB signals, evaluates them in association with the signal from the inverse color conversion unit 56, and outputs the evaluation result.

The output from the color conversion unit 50 is added to the evaluation result from the error evaluator 57, and is led to the predictive coding unit. That is, in the case corresponding to the luminance signal,
The quantization result is coded and output by the entropy coder 52 through the process of obtaining the difference from the prediction signal and quantizing in the prediction coding unit 51.

FIG. 11 shows a quasi-reversible compression model. The entropy of an input signal is reduced by color conversion, and the compression ratio can be improved as compared with the lossless model.

[0016]

However, in the case of FIG. 9, since the transform coefficients of the inverse transform matrix have a finite word length, the input signals R, G, and B in the transform equation (2) are completely Is not restored. In addition, in equations (1) and (2), an error occurs due to addition, subtraction, and multiplication of a finite word length.

In this configuration, the RGB signals are not completely reproduced, but the entropy of the input signal is reduced by 1) color conversion and 2) quantization, and the compression ratio can be improved. In color conversion, the redundancy between RGB signals is reduced, and in quantization, entropy is reduced by changing the number of levels and frequency distribution of the signals.

In the case of FIG. 10, there is a limit in the improvement of the compression ratio because there is redundancy between the RGB signals and quantization cannot be performed to guarantee reversibility. Further, in the case of FIG. 11, since the color conversion is used, the distortion cannot be avoided, but the distortion generated in the decoded image is suppressed to ± 1 with respect to the pixel value of the original image, and the number of pixels to which the distortion is added is minimized. It is controlled so that

An object of the present invention is to propose a unified encoding for lossless / quasi-lossless encoding in which the quality of a reproduced image can be arbitrarily controlled.

[0020]

In the quasi-reversible compression model shown in FIG. 11, the distortion occurring in the reproduced image is suppressed to within ± 1 with respect to the pixel value of the original image, and the number of pixels to which the distortion is applied is minimized. A quasi-reversible algorithm has been proposed, and it is possible to limit the magnitude of distortion in pixel units. But,
The quality required for image encoding differs depending on the application, and it is desirable that distortion can be arbitrarily controlled by an encoding algorithm.

The present invention proposes a quasi-reversible process capable of controlling the distortion generated in the reproduced image within an arbitrary value ± p with respect to the pixel value of the original image. Further, a control process for reducing the entropy of the output signal and the power of the reproduction error is newly proposed under the constraint condition that the distortion of each pixel is kept within an arbitrary value ± p. Furthermore, by combining with the previously proposed color lossless conversion, lossless and quasi-lossless encoding are realized by unified processing.

[0022]

DESCRIPTION OF THE PREFERRED EMBODIMENTS To realize a reversible and quasi-reversible encoding in a unified system, consider a configuration in which a quantizer is added to the previously proposed reversible color converter [Japanese Patent Application No. 11-8717]. FIG. 1 shows a configuration diagram of an embodiment of the present invention. In FIG. 1, reference numeral 1 denotes a lossless color converter, 2 denotes a quantizer, 3 denotes a prediction processing unit, 4 denotes an entropy encoder, and 5 and 6 denote signals corresponding to the signals O ′ _{2 and} O ′ ₃ , respectively. A processing unit that performs the same processing as the processing unit for O ′ ₁ is shown.

The quantizer 2 quantizes the outputs O _1, O _{2 and} O ₃ of the reversible color converter 1 to obtain O ′ _{1 to} O ′ ₃ . Further, the prediction processing unit 3 holds the immediately preceding O ′ ₁ , outputs a difference from the new O ′ ₁ to the entropy encoder 4, and uses the new O ′ ₁ for the next prediction. . Although the original image is not completely reproduced due to quantization,
The entropy of the quantizer output signal is reduced and the compression ratio is improved. The quantization step size controller (not shown)
The distortion generated in the decoded image is ±
The quantization step size is selected so that the power of the entropy of the output signal or the power of the reproduction error is reduced under the constraint condition to keep it within p. As shown in FIG. 8, the configuration differs depending on whether the entropy of the prediction error or the power of the reproduction error is reduced.

(A) Reversible color conversion / inverse conversion (Japanese Patent Application No. Hei 11
[Conversion procedure] Step 1:

[0025]

(Equation 5)

Step 2:

[0027]

(Equation 6)

However,

[0029]

(Equation 7)

Represents a maximum integer not exceeding x. At this time, the conversion coefficient, that is, the output signal is given by O ₁ = x ₁ (0) (8) O ₂ = x ₂ (0) (9) O ₃ = x ₂ (1) (10) Average, average slope,
Represents the difference in slope.

[Inverse Conversion Procedure] Conversion coefficient (O _1, O _2, O ₃ )
The procedure for obtaining the input signals (R, G, B) from is given as follows.

Step 3:

[0033]

(Equation 8)

Step 4:

[0035]

(Equation 9)

(B) Quantizer In the present invention, a uniform quantizer such as the mid-tread type shown in FIG. 2 is used. For an output signal of the color reversible conversion, that is, an input signal O _i to the quantizer, an output signal O ′
_i is represented by the following equation.

[0037]

(Equation 10)

Where Δ _i is a positive integer and represents the quantization step size of O _i . The inverse quantization O is the output O _i of _{". I = (2Δ i +1} ) O ' is calculated by _i (18) dequantization signal after reduction O" _i and the color reversible converter O _"i - The relationship of Δ _i < O _i < O ″ _i + Δ _i (19) holds, and the quantization error by the quantizer of FIG.
Is within Δ _i.

(C) Analysis Regarding Error Between Reconstructed Image and Original Image A reproduced image is obtained by inversely transforming the inversely quantized signal of Expression (18) by Expressions (11) to (16). Here, the error between the reproduced image and the original image will be considered. An error between the signal O ″ _i after inverse quantization and the output O _i of the color reversible converter, that is, a quantization error is defined as δ _i by the following equation.

Δ _i = O ″ _i −O _i (20) From equation (19), δ _i is a value in the range of −Δ _i < δ _i < Δ _i . At this time, the reproduced images R ′ and G ′ , B 'can be expressed by the following equation.

R ′ = R + p ₁ (δ _1, δ _2, δ ₃ ) (21) G ′ = G + p ₂ (δ _1, δ _2, δ ₃ ) (22) B ′ = G + p ₃ (δ _1, δ _{2 ,} δ ₃ ) (23) Here, p _i (δ _1, δ _2, δ ₃ ) is a reproduction error, that is, an error between the reproduced image and the original image, and is given by the following equation.

[0042]

[Equation 11]

[0043] Here m, n are integers, the maximum error p _i = Max when the quantization step size _{Δ i | p i (δ 1} , δ 2, δ 3) | (29) for δ i = -Δ _{i, ...,} Motomari in Δ _{i i} = 1,2,3, in order to suppress the reconstruction errors within the maximum p is _{p = Max (p 1, p} 2, p 3) (30) and so as to delta _1, delta _2, may be set delta _3. However, |
x | represents the absolute value of x. As an example, p = 1 and p
= In the case of _{2 (Δ 1, Δ 2,} Δ 3) and the corresponding maximum error (p
_1, indicating _{p 2, p 3), 1} ) the case of p = 1

[0044]

(Equation 12)

There are the following six patterns, and in any case, the reproduction error can be suppressed to within ± 1.

2) When p = 2 When p = 2, the following 14 patterns exist.

[0047]

(Equation 13)

[0048] p> 3 of so as to satisfy the relational expression Likewise formula case _{(30) (Δ 1, Δ} 2, Δ 3) by setting, it is possible to suppress reproduction error within ± p Become. Note that the lossless coding can be realized by setting _{(Δ 1, Δ 2, Δ} 3) = (0,0,0).

(D) Quantization step size controller The step size (Δ _1, Δ
_2, Δ ₃ ) exist in several types. Therefore, an appropriate step size _{(Δ 1, Δ 2, Δ} 3) to define the criteria for selecting. Here, the minimization of one of the following two quantities is adopted as the evaluation criterion.

1. 1. Entropy of prediction error Power of Reproduction Error FIGS. 3 and 4 show configurations based on the above evaluation amounts. Reference numerals 1 to 4 in the figure correspond to FIG. 1, 10 is an entropy calculator, 20 is an inverse quantizer, 21 is an inverse color converter,
Reference numeral 22 denotes an error evaluator. Since these metrics do not give essentially the same step size, it is necessary to decide which criterion to choose. The procedure for setting the step size will be described below.

(E) Step Size Setting Procedure 1) Determine how much distortion generated in the decoded image is allowed with respect to the pixel value of the original image. That is, p is set.

2) Under the constraint condition of equation (30), a step size that minimizes the selected evaluation criterion is adopted. The quantization step size is controlled so as to reduce the entropy of the output signal or the power of the reproduction error under the constraint condition that the distortion generated in the decoded image is set to ± p with respect to the pixel value of the original image by the above procedure. It becomes possible. Note that the additional information is a step size Δ _1, Δ _2, Δ ₃ alone, the overhead is small.

(F) Coding Characteristics In order to verify the effectiveness of the present invention, a simulation was performed on the temple image shown in FIG. Table 1 shows the distribution of the reproduction error when the distortion generated in the reproduced image is set to p = 1.

[0054]

[Table 1]

Here, the SNR (SN ratio) was calculated by the following equation.

[0056]

[Equation 14]

Here, σ ² _R, σ ² _{G, and} σ ² _B indicate the power of the reproduction error of each component of the RGB signal. Note that all the predictors were fixed to the seventh predictor of Lossless JPEG (average of pixels immediately before and immediately above) in consideration of the amount of calculation.

From Table 1, it is found that the reproduction error is suppressed within ± 1.
Can be confirmed. Also, Δ ₁Δ_TwoΔ_ThreeTo change
As a result, the entropy of the prediction error and the SN ratio of the reproduced image
It can be seen that it changes.

[0059] From Table 1, when it is selected to minimize the entropy of the prediction error as an evaluation criterion for the step size selection _{(Δ 1, Δ 2, Δ} 3) = (0,0,4) is a playback error If you choose to maximize minimize namely SN ratio of the power _{(Δ 1, Δ 2, Δ} 3) = (0,0,1) is understood to be selected.

FIG. 6 shows two evaluation criteria: entropy of prediction error (Eval. 1) and power of reproduction error (Eval. 1).
2 shows the characteristic of distortion p versus entropy obtained under the evaluation criterion that minimizes val. 2).

FIG. 6 shows that the entropy is reduced by increasing the distortion. Also, it can be seen that the increase or decrease in entropy can be controlled while keeping the distortion p constant. It should be noted that when the YUV conversion is calculated by a floating-point calculation, it is known that an error of ± 1 occurs in the reproduced image.

FIG. 6 shows that when the minimization of entropy is selected as the evaluation criterion, the entropy of the proposed method at p = 1 shows almost the same value as the YUV conversion.

FIG. 7 shows the characteristics of the distortion p to SN ratio obtained based on the above two evaluation criteria.

From FIG. 7, it can be confirmed that when minimizing the power of the reproduction error is selected as the evaluation criterion, it is possible to keep the SN ratio high. Also, when the same distortion is allowed, that is, when p = 1, the SN ratio is higher than that when the YUV conversion is used.

FIG. 8 shows characteristics of entropy to SN ratio obtained based on two evaluation criteria. From FIG. 8, it is confirmed that the encoding characteristics of the present invention are slightly inferior to those of the case using the YUV conversion. However, in the present invention, the distortion of the reproduction error can be arbitrarily controlled from reversible to quasi-reversible,
The advantages are great, for example, encoding according to the required quality is possible. Although the DPCM is used as a predictor in consideration of the amount of calculation in this example, the coding characteristics can be improved by using a high-efficiency reversible coding method for other monochrome images.

[0066]

As described above, according to the present invention, an encoding method capable of continuously controlling lossless / quasi-reversible encoding of a color still image can be proposed. In quasi-reversible coding, the quantization step size is set so as to reduce the entropy of the prediction error or the power of the reproduction error under the constraint that the distortion occurring in the decoded image is limited to ± p with respect to the pixel value of the original image. Can be controlled, and the effectiveness has been verified by simulation.

The features of the present invention are as follows: (1) Realization of lossless / quasi-lossless encoding by a unified algorithm (2) Realization of high compression rate in lossless encoding (3) Lossless encoding for grayscale images (4) Realization of controllability of reproduced image quality in quasi-reversible coding (5) Small amount of additional information is sufficient.

[Brief description of the drawings]

FIG. 1 shows a configuration diagram of an embodiment of the present invention.

FIG. 2 shows characteristics of a quantizer.

FIG. 3 shows an example of control by minimizing entropy.

FIG. 4 shows an example of control by minimizing a reproduction error.

FIG. 5 shows an image to be processed.

FIG. 6 shows distortion p versus entropy characteristics.

FIG. 7 shows distortion p-to-SN ratio characteristics.

FIG. 8 shows entropy-to-SN ratio characteristics.

FIG. 9 shows a conventional example in the case of irreversible compression.

FIG. 10 shows a conventional example in the case of lossless compression.

FIG. 11 shows a conventional example of quasi-lossless compression.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Reversible color converter 2 Quantizer 3 Prediction processing part 4 Entropy encoder 10 Entropy calculator 20 Inverse quantizer 21 Inverse color converter 22 Error evaluator

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Junji Suzuki 2-3-1 Otemachi, Chiyoda-ku, Tokyo F-term (reference) in Nippon Telegraph and Telephone Corporation 5C057 AA11 BA14 CE06 DA02 DA03 EA01 EA02 EL01 EM09 EM13 GE08 GF03 GJ02 5C059 KK25 KK39 MA00 MA23 MA45 MC33 MC35 MC38 PP15 TA30 TA53 TC08 TC22 TD14 UA02 UA05 5C078 AA09 BA32 BA57 CA00 CA22 DA01 DA02 5J064 AA01 BA09 BA16 BB03 BB14 BC16

Claims (8)

[Claims] 1. A color image encoding apparatus using RGB as an input, wherein an average of RGB is x ₁ (0), a GR difference is x ₁ (1), and a BG difference is x ₁ (2). was the difference between the intermediate result component _{x 1 (0) ~x 1 (} 2) mean of x ₁ (1) and x ₁ (2) upon obtaining the O ₂ x _{1 (2)} and x ₁ (1) A conversion unit that outputs output components O _{1 to} O ₃ with O ₃ x ₁ (0) as O ₁ and the components O _{1 to} O ₃ are input, and (However, delta _i is a positive represents the quantization step size of the O _i an integer) with and a and the output components O quantizer for outputting a _'1 ~O' _3, the component O _'1 ~ O' _3. A color image encoding device, comprising: a prediction processing unit that performs prediction processing for each of ₃ and an entropy encoder that receives an output of the prediction processing unit and performs entropy encoding. 2. An apparatus according to claim 1, further comprising an entropy calculator for branching and receiving an output of said prediction processing unit, and controlling a quantization step size in said quantizer based on an evaluation result by said entropy calculator. 2. The color image encoding apparatus according to claim 1, wherein: 3. An inverse quantizer that branches and receives an output of the prediction processing unit, an inverse color converter that performs inverse color conversion based on an output of the inverse quantizer, and an output of the inverse color converter. And said RGB,
2. A color image according to claim 1, further comprising an error evaluator for performing error evaluation, and controlling a quantization step size in said quantizer based on an evaluation result by said error evaluator. Encoding device. 4. An apparatus for decoding a color image which outputs RGB, comprising: an entropy decoder for performing entropy decoding on a supplied signal; and performing a prediction process based on an output of the entropy decoder. A prediction processing unit that outputs the output components O ′ _{1 to} O ′ ₃ , the components O ′ _{1 to} O ′ ₃ as input signals, and the output components O _{1 to} O 3
₃ and an intermediate quantized component x ₁ (1) with the components O ₂ and O ₃ as inputs.
And x ₁ (2) as outputs and the component O ₁ and the intermediate result component x
A color image decoding device, comprising: an inverse conversion unit that receives ₁ (1) and an intermediate result component x ₁ (2) as input and outputs RGB. 5. A method for encoding a color image using RGB as input, wherein an average of RGB is x ₁ (0), a difference between GR is x ₁ (1), and a difference between BG is x ₁ (2). was the difference between the intermediate result component _{x 1 (0) ~x 1 (} 2) mean of x ₁ (1) and x ₁ (2) upon obtaining the O ₂ x _{1 (2)} and x ₁ (1) A conversion process is performed in which output components O _{1 to} O ₃ with O ₃ x ₁ (0) as O ₁ are output, and the components O _{1 to} O ₃ are input and (Where Δ _i is a positive integer and represents the quantization step size of O _i ). A quantization process of outputting the output components O ′ _{1 to} O ′ ₃ is executed, and the components O ′ _{1 to} O ′ ₃ are executed. A color image encoding method, wherein a prediction process for performing a prediction process for each of the above and an entropy encoding is performed in response to an output of the prediction process. 6. The method according to claim 1, wherein an output of said prediction processing is branched and received, an entropy calculation is executed, and a quantization step size in said quantization processing is controlled based on an evaluation result by said entropy calculation. 6. The method for encoding a color image according to claim 5, wherein 7. An inverse quantization is performed by branching and receiving an output of the prediction processing, and performing an inverse color conversion based on an output of the inverse quantization, and outputting the output of the inverse color conversion and the RGB. 6. The color image encoding method according to claim 5, wherein the method further comprises receiving and performing an error evaluation, and controlling a quantization step size in the quantization processing based on an evaluation result obtained by the error evaluation. 8. A method for decoding a color image which outputs RGB, performs entropy decoding on a supplied signal, performs a prediction process based on the result of the entropy decoding, and performs output components O ′ ₁ to O ′ ₁ . O ′ ₃ is output, the components O ′ _{1 to} O ′ ₃ are input, the output components O _{1 to} O ₃ are output, and inverse quantization is performed. The intermediate results are obtained using the components O ₂ and O ₃ as inputs. Component x ₁ (1)
And x ₁ (2) as outputs and the component O ₁ and the intermediate result component x
A color image decoding method characterized by performing an inverse conversion process using ₁ (1) and an intermediate result component x ₁ (2) as input and RGB as output.