JPH09149417A - Dynamic image signal decoder - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve a picture by filtering a decoded dynamic image through the use of a filter updating a filter coefficient by each frame to remove mosquito distortion appearing the decoded dynamic image without remarkably increasing a circuit scale. SOLUTION: A filter part 4 executes filter arithmetic to the decoded dynamic image outputted from a frame memory part 15 and removes distortion to output. A filter control part 2 determines a threshold value optimum for removing mosquito distortion by each frame through the use of an inputted coding parameter information, quantizing information and inverse descrete cosine coefficient distribution information and outputs the threshold value to a filter arithmetic part 3. The filter arithmetic part 3 executes filter arithmetic through the inputted threshold value to the output of the frame memory part 15 and outputs the decoded dynamic image free from mosquito distortion. Thereby mosquito distortion appearing at the decoded dynamic image is removed without remarkably increasing the circuit scale, to improve the picture.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a moving picture signal decoding apparatus used when decoding a predictive-coded moving picture signal.

When transmitting and storing a digitized moving image signal, it is indispensable to reduce the amount of information to improve the transmission efficiency and the storage capacity. As a high-efficiency coding method for reducing the amount of information as described above, using the coded pixel value, a prediction value for a pixel to be coded is made, and a prediction that is a difference between the input signal and the prediction value. Predictive coding for coding and transmitting error signal: -The image is divided into blocks for each pixel, discrete cosine transform is performed for each block to transform from the pixel domain to the frequency domain, and transform coefficients are coded and transmitted. Transform coding, variable length coding that varies the code length based on the statistics of the codeword itself, quantization, and the like are known.

On the other hand, the above-mentioned high-efficiency coding method is ITU-T H.261, which is an international standardized method of moving picture compression coding, ISO /
IEC MPEG-1 (Moving Picture Experts Group Phase 1)
And ISO / IEC MPEG-2.

However, when encoding is performed using the above-mentioned high efficiency encoding method, noise which is not present in the original moving image occurs in the decoded moving image. This noise is due to the lack of discrete cosine coefficient information due to quantization after performing discrete cosine transform in the encoder.

That is, although the moving image has frequency components from the low frequency region to the high frequency region, when the discrete cosine transform and the quantization are performed, the frequency range becomes a finite and discrete value, so that the moving image and A difference occurs in, and this difference becomes noise.

Further, in a decoded moving image, for example, block distortion which is a unit for performing discrete cosine transform, for example, discontinuity of signal value at the boundary of a square block of 8 × 8 pixels, or intra-block It is called mosquito distortion found in the flat part around the edge of the pixel where the signal value changes greatly.
"Haze" occurs and visual quality deterioration occurs.

The above block distortion and mosquito distortion mainly consist of high frequency components.

[0008]

2. Description of the Related Art As described above, when a high-efficiency coding method is used when transmitting or storing a digitized moving image signal,
Block distortion and mosquito distortion occur in the decoded moving image.

Therefore, in order to remove these distortions, a low-pass filter is applied to the decoded moving image because of the property that the frequency components of block distortion and mosquito distortion are concentrated in the high frequency region. realizable.

However, if a low pass filter is used,
Irrespective of the distortion, it erases fine patterns etc. that existed in the moving image, edges in the decoded moving image (for example,
(Character) is blurred. This is because a fine pattern or an edge contains a high frequency component, but this component is removed by a low pass filter.

Here, as a method for removing the block distortion, for example, Japanese Patent Laid-Open Nos. H01-055988 and H02-0570.
Japanese Patent No. 67 (Title of Invention: Image Coding Processing System) and the like. In these methods, in order to perform effective filtering, it is determined whether the pixel position currently being processed is a block boundary or not, and adaptive filtering is performed.

However, in order to realize the device, it is necessary to provide a circuit for determining the pixel position, and the circuit scale becomes large. Furthermore, the above is a method for block distortion and cannot be applied to mosquito distortion.

Further, as a method for removing mosquito distortion, there is, for example, Japanese Patent Laid-Open No. H06-319126 (title of the invention: video signal processing device). Circuit (for example, if the signal level passed through the high-pass filter is equal to or higher than the threshold value, it is determined to be an edge, and the determined portion is filtered). Grows larger.

On the other hand, in recent years, the transmission capacity and the storage capacity have tended to increase, and the number of cases of handling high-quality images using a relatively highly efficient coding method such as MPEG-2 has increased. In such a case, the mosquito distortion becomes conspicuous among the above two distortions.

This allows MPEG-2 to send information at a transmission rate four times or more that of MPEG-1. Therefore, MPEG
The quality of the -2 decoded video was better than that of the MPEG-1 decoded video, but on the other hand, "moyamoya" (mosquito distortion) around the edges became noticeable.

To remove this, it is necessary to add a circuit for determining the presence or absence of an edge at each pixel position as described above.

[0017]

SUMMARY OF THE INVENTION As explained above,
In order to remove the mosquito distortion, it is necessary to provide a circuit for determining the presence / absence of an edge at each pixel position, but the circuit scale becomes large by providing this circuit.

An object of the present invention is to improve the image quality by removing the mosquito distortion appearing in the decoded moving image without significantly increasing the circuit scale.

[0019]

FIG. 1 is a diagram illustrating the principle of the first to fourth aspects of the present invention. According to the first aspect of the present invention, when removing noise in a decoded moving image, the decoded moving image is filtered by using a filter whose filter coefficient is updated for each frame.

According to a second aspect of the present invention, the moving picture signal decoding apparatus performs variable length decoding of the input coded moving picture signal, extracts the quantized coefficient, the pixel information and the coding parameter information, and outputs the variable length. A decoding unit, an inverse quantization operation is performed on the image information using the input quantization coefficient, an inverse quantization operation result and quantization information are extracted, and output, and an input inverse quantization unit. An inverse discrete cosine transform is performed on the result of the quantization operation to extract the inverse discrete cosine transform output and the inverse discrete cosine coefficient distribution information, and an inverse discrete cosine transform section that outputs the frame memory section that stores the decoded moving image, Motion compensation is performed using the inverse discrete cosine transform output and the output from the frame memory unit to reproduce a predictively-coded frame and output the frame to the frame memory. Performs filter operation on the decoded moving picture output from the frame memory unit, and the like constituting a filter section for outputting to remove noise.

In a third aspect of the present invention, the filter section uses a filter control section for determining a filter coefficient and a filter coefficient output from the filter control section, and the decoded moving image from the filter memory section is used. It is configured to include a filter calculation unit that performs a filter calculation on the.

According to a fourth aspect of the present invention, the filter control section updates the filter coefficient for each frame using the coding parameter information, the quantization information, and the inverse discrete cosine coefficient distribution information, and outputs the updated filter coefficient.

Here, an ε filter has been proposed as a filter capable of effectively smoothing a small-amplitude noise component superimposed on a signal accompanied by a sharp change (Harashima et al.,
"Ε-separation nonlinear digital filter and its application", IEICE Transactions, 1982/4 Vol. J65-A No.4, p.297 ~
p.304).

Therefore, the present invention uses the ε filter in order to effectively remove the mosquito distortion without impairing the sharpness of the edge itself, and the filter coefficient (hereinafter, referred to as a threshold) of this filter is used for each frame. I updated it to.

As a result, the optimum threshold value can always be set for the decoded moving image without significantly increasing the circuit scale, so that the mosquito distortion is removed and the image quality of the decoded moving image is improved. To be

Now, the principle of the present invention will be described with reference to FIG. The variable-length decoding unit 11 performs variable-length decoding on the input coded moving image signal, and coding parameter information (including intraframe prediction value / interframe prediction value, quantization coefficient, etc.)
And pixel information is extracted, and the encoding parameter information (excluding
Quantized coefficient) to the filter controller 2, and the quantized coefficient and pixel information to the inverse quantizer 12.

The inverse quantization unit 12 performs an inverse quantization operation on the pixel information using a quantization coefficient to extract an inverse quantization operation result and quantization information, and the former is an inverse discrete cosine transform unit 13
Then, the latter is sent to the filter control unit 2.

The inverse discrete cosine transform unit 13 performs an inverse discrete cosine transform operation on the result of the inverse quantization operation, extracts the inverse discrete cosine transform output and the inverse discrete cosine coefficient distribution information, and obtains the inverse discrete cosine coefficient distribution information. Filter control unit 2
To send to.

The frame memory unit 15 stores the input decoded moving image as an image display frame or a reference frame for inter-frame prediction. The motion compensation unit 14 reproduces a predictively encoded frame by performing motion compensation using the output of the inverse discrete cosine transform and the output from the frame memory unit 15, and decodes the motion-compensated decoded video for each frame. The image is output to the frame memory unit 15.

The filter unit 4 performs a filter operation on the decoded moving image (reproduced moving image) output from the frame memory unit 15, removes distortion, and outputs. Here, the filter unit 4 is composed of the filter control unit 2 and the filter calculation unit 3, and the filter control unit 2 uses the input coding parameter information, quantization information, and inverse discrete cosine coefficient distribution information, The optimum threshold value for removing the mosquito distortion is obtained for each frame and output to the filter calculation unit 3.

The filter calculation section 3 includes a frame memory section 15
The ε filter operation using the input threshold value is performed on the output of 1 to output the decoded moving image without mosquito distortion. This makes it possible to improve the image quality by removing the mosquito distortion appearing in the decoded moving image without significantly increasing the circuit scale.

[0032]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 2 is a block diagram of the essential parts of the first to fourth embodiments of the present invention, and FIG.
Is the decoded moving image f (x) input to the ε filter with mosquito distortion, and (b) is the decoded moving image g (x) output from the ε filter without mosquito distortion.

Further, FIG. 4 is a block diagram of a main part of the ε filter calculation unit in FIG. 2, FIG. 5 is a control explanatory diagram of the ε filter calculation unit of FIG. 4, and FIG. 6 is a calculation flow of the ε filter calculation unit of FIG. FIG. 7 is a diagram showing an example of the threshold setting curve of the ε filter,
(a) is an intra-frame coded frame, and (b) is an inter-frame coded frame.

Here, the same reference numerals denote the same objects throughout the drawings. Further, the reference numerals on the left side in FIG. 5 are diagrams for explaining the operation of the same reference numerals in FIG. The operation of the embodiment of the present invention will be described below with reference to FIGS.

The parts described in detail above will be briefly described, and the parts of the present invention will be described in detail. FIG.
In, when the encoded moving image signal is input to the moving image signal decoding device, the variable length decoding unit 11 in the device encodes the coding parameter information extracted from the encoded moving image signal to the threshold value setting unit 22, The quantization coefficient and pixel information in the quantization parameter information are sent to the inverse quantization unit 12.

The inverse quantization unit 12 performs an inverse quantization operation on the input information, and the obtained inverse quantization operation result is sent to the inverse discrete cosine transform unit 13 and the quantization information is sent to the threshold value setting unit 22. Send out. The inverse discrete cosine transform unit 13 performs an inverse discrete cosine transform operation on the input information, and sends the latter out of the obtained inverse discrete cosine transform output and inverse discrete cosine coefficient distribution information to the threshold setting unit 22. .

The frame memory unit 15 stores the decoded moving image as an image display frame or a reference frame for inter-frame prediction. The motion compensating unit 14 performs motion compensation using the inverse discrete cosine transform output from the inverse discrete cosine transform unit and the output from the frame memory unit 15 to reproduce a predictively encoded frame and perform motion compensation. The decoded moving image for each frame is output to the frame memory unit 15.

Therefore, the filter unit is the frame memory unit.
The decoded moving image (reproduced moving image) output from 15 is filtered to remove distortion and then output. here,
The filter unit includes a threshold value setting unit 22 and an ε filter calculation unit 31.
The decoder control unit 21 controls the operation of each unit in the decoding device, and makes a threshold setting request for each frame to the threshold setting unit 22.

As a result, the threshold value setting unit 22 generates a threshold value ε ₀ for each frame using the input coding parameter information, quantization information, and inverse discrete cosine coefficient distribution information, and the ε filter calculation unit Send to 31.

Now, the ε filter, particularly the two-dimensional ε filter in the present invention, will be described. The image f (x,
image g (x, y) output from this filter when y) is input
Is defined as follows.

[0041]

(Equation 1)

However, F (x) is the following function.

[0043]

(Equation 2)

That is, F (x) is a function whose output is limited to −ε ₀ ≦ F (x) ≦ ε ₀ , and ε ₀ is called the threshold value of this function.

The ε filter averages the pixels whose absolute difference value is equal to or smaller than the threshold ε ₀ among the peripheral pixels of the pixel currently being processed. For example, FIG.
By applying the ε filter operation to the small fluctuations (mosquito distortion) at the bottom and top of the edge shown in (a),
As shown in FIG. 3B, the edge having a large signal difference is maintained as it is, and only the minute fluctuations around the edge are averaged to remove the mosquito distortion.

Here, the above-mentioned ε-filter computing section has, for example, the configuration shown in FIG. 4, and the function of each section is as follows. That is, the threshold setting unit 22 of FIG. 2 sets the threshold ε ₀ updated for each frame together with the register setting signal as shown in FIG.
It outputs to the threshold value holding register 311 of. Therefore, the threshold value holding register 311 holds the threshold value input using the register setting signal.

The subtractor 312 is the frame memory unit 15 of FIG.
The difference between the processed pixel value output from the pixel and the peripheral pixel value of this processed pixel is calculated, and the difference value is output to the comparator 313 and the selector 314. Note that this part is f (x, y) − f (x
+ I, y + j).

The comparator 313 has a threshold value holding register 311.
The threshold value output from the subtractor 312 is compared with the threshold value output by the subtracter 312, and the absolute value of the output from the subtractor 312 is compared with the threshold value holding register.
When it is smaller than the output of 311, the select signal "1" is output, and when it is larger, the select signal "0" is output.

The selector 314 outputs the output of the subtractor 312 when the select signal from the comparator 313 is "1", and the output when the select signal is "0".
Output "0". In addition, the comparator 313 and the selector 314
It realizes the function of equation (2).

The divider 316 divides the output from the selector 314 by a fixed value and outputs the division result. The fixed value is (2M + 1) ² , and M is the number of peripheral pixels in the horizontal or vertical direction to be calculated. In addition, this part corresponds to the division operation of equation (1).

The adder 315 adds the output from the divider 316 and the output from the register 319 and outputs the result. Control unit 317
Controls the filter operation. That is, the selector 318 outputs select signals corresponding to the two input signals to this selector, and the register 319 outputs the select signal to the register 31.
Output 9 load signal.

As a result, the selector 318 has the control section 317.
Either the output from the adder 315 or the processed pixel value input from the outside is selected and output according to the select signal from.

The register 319 registers the output from the selector 318 according to the load signal from the control unit 317.
Save it to 319. (1) by adder 315 and register 319
It realizes the multiply-accumulate operation of expressions.

FIG. 5 shows the relationship between the input / output of the ε-filter operation unit in FIG. 4 and the generation timing of the control signal output by the internal control unit 317. FIG. 5 will be described with reference to FIG. Here, f0 in FIG. 5 is a pixel value to be processed, f1, f2, f3 ...
fn is the pixel value of the surrounding pixels used for filtering, s1, s2,
・ ・ Sn is the difference between f0 and f1, f2, ・ ・ fn, g is the processing result, and R is the register load signal. In addition, a part with a cross in the square indicates that no pixel value is input. Further, on the left-hand side are inputs and outputs of the parts with the same reference numerals in FIG.

The processed pixel value f0 input at time t ₁₁ in FIG. 5 is applied to the register 319 through the selector 318. At this time, since the control signal from the control unit 317 is applied, the processed pixel value f0 is stored in the register 319.

The processed image value f0 and the peripheral pixel value f1 input at time t ₁₂ are subtracted by the subtracter 312, and (f1-f0) = s1 is applied to the adder 315 via the selector and the divider. Adder 315
Since f0 is also applied stored in the register 319 at time t ₁₁ to where it is cumulatively added f0 + s1 is stored in the register.

At time t ₁₃ , the same calculation as above is performed.
f0 + s1 + s2 is stored. Note that s2 = (f2-f0). By repeating the cumulative addition in this way, at the time t _1n , the cumulative addition value (f0 + s1 + s2 ... + sn) up to the time t _1n is stored in the register 319 as the final processing result g.

Further, FIG. 6 shows a calculation flow of the ε filter calculation section shown in FIG. 4, and the calculation procedure will be described with reference to this figure. Note that FIG. 6 shows one-dimensional ε for the sake of simplicity.
A filter, M ₁ is the total number of peripheral pixels used for filtering, and f (n) is the nth peripheral pixel value.

First, step 1 (hereinafter abbreviated as S1)
Is the initialization step, f (0) is the pixel to be processed,
Insert f (x, y) into f (0). Further, when performing the filter calculation, peripheral pixels are sequentially numbered, but the peripheral pixel numbers are added to n.

Then, the difference value between the pixel set in S1 and the peripheral pixel is calculated, the absolute value of the difference value is compared with the preset threshold value ε _0, and if the absolute value is larger, End the operation (see N in S2, S3).

However, if the absolute value is smaller than the threshold value ε ₀ , the difference value is cumulatively added from sum = sum− (diff / M ₁ ). Then, if n = M _{1 is} not satisfied, the value of n is increased by 1 and the above calculation is repeated from S2. This iteration ends the calculation when n = M (see S4 to S6).

The sum at the end of the calculation is a value after being processed by the ε filter, and a decoded moving image from which mosquito distortion has been removed can be obtained. The threshold setting curve of the ε filter shown in FIG. 7 is obtained by comparing the coding parameter information (interframe coding frame / intraframe coding frame, etc.) output by the variable length decoding unit 11 of FIG. It is obtained experimentally using the quantized coefficient information output by 12 and the inverse discrete cosine coefficient distribution information (highest frequency) output by the inverse discrete cosine transform unit 13.

In FIG. 7, the horizontal axis represents the average value of the quantized coefficients of each block in one frame output by the inverse quantization unit in FIG. 2, and the vertical axis represents the output threshold value. Here, when the coding parameter information output from the variable length decoding unit 11 in FIG. 2 indicates an “intra-frame coded frame”, the curve in FIG. 7A is used.

The value of Fmax attached to this curve is determined by zigzag scanning the highest non-zero frequency component of the discrete cosine coefficient of each block in one frame output from the inverse discrete cosine transform unit 13 of FIG. It is represented by the average of the positions when performing.

For example, in each block, Fm
The value of ax is "0" when it contains only the DC component, and 63 when the highest frequency component is non-zero (when the discrete cosine transform is performed at 8x8). In other words, the larger the value of Fmax,
It shows that the image of the high frequency component is included.

As described above, the threshold setting curve of FIG. 7A is obtained by using the average quantization coefficient output from the inverse quantizer 12 and the highest frequency output from the inverse discrete cosine transform. Determine the threshold value.

On the other hand, when the coding parameter information output from the variable length decoding unit indicates "interframe coding frame", FIG. 7B is used. In this case, the threshold value is determined from the average quantization coefficient output from the inverse quantization unit 12.

This difference is that in the case of "intra-frame coded frame", the difference information is not sent and all the discrete cosine coefficients of the edge part in the image are sent, so it is necessary to change the threshold value based on the distribution of these coefficients. is there.

However, in the case of the “inter-frame coded frame”, since only the difference information such as the minute luminance change is sent by the motion prediction, the discrete cosine coefficient does not show the high frequency component representing the edge portion. Therefore, the threshold is determined only by the average quantization coefficient.

In other words, the present description is intended to remove the mosquito distortion appearing in the decoded moving image while decoding the encoded moving image, and it is possible to effectively remove the mosquito distortion using the ε filter. There is.

Further, by the configuration in which the threshold value of the ε filter is adaptively updated on a frame-by-frame basis, the optimum threshold value is always set for an arbitrary moving image without significantly increasing the circuit scale, and decoding is performed. There is an advantage that the quality of the captured moving image can be improved.

[0072]

As described above in detail, it is possible to suppress the mosquito distortion appearing in the decoded moving image and improve the quality of the decoded moving image without significantly increasing the circuit scale. Is obtained.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating the principle of the first to fourth aspects of the present invention.

FIG. 2 is a main part configuration diagram of an embodiment of the first to fourth aspects of the present invention.

FIG. 3 is a diagram for explaining the operation of the ε filter, where (a) is the decoded moving image f (x) input to the ε filter, and when mosquito distortion is present,
(b) is the decoded moving image g (x) output from the ε filter,
This is the case without mosquito distortion.

FIG. 4 is a configuration diagram of a main part of an ε filter calculation unit in FIG. 2.

5 is a control explanatory diagram of an ε filter calculation unit in FIG. 4;

FIG. 6 is a calculation flowchart of the ε filter calculation unit in FIG.

FIG. 7 is a diagram showing an example of a threshold setting curve of an ε filter, where (a) is a case of an intra-frame coded frame and (b) is a case of an inter-frame coded frame.

[Explanation of symbols]

2 Filter control unit 3 Filter calculation unit 11 Variable length decoding unit 12 Inverse quantization unit 13 Inverse discrete cosine transform unit 14 Motion compensation 15 Frame memory unit

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁶ Identification code Agency reference number FI Technical display location H04N 7/30 H04N 7/133 Z

Claims (4)

[Claims] 1. A moving picture signal decoding apparatus for decoding a predictively coded moving picture signal, wherein a filter whose filter coefficient is updated for each frame is used when noise in the decoded moving picture is removed. Then, the moving picture signal decoding apparatus is characterized in that the decoded moving picture is filtered. 2. A variable length decoding unit, wherein the moving picture signal decoding device performs variable length decoding of an input coded moving picture signal, extracts a quantized coefficient, pixel information and coding parameter information, and outputs the quantized coefficient, pixel information and coding parameter information. , An inverse quantization unit that performs an inverse quantization operation on image information using the input quantization coefficient, extracts an inverse quantization operation result and quantization information, and outputs the inverse quantization unit, and an input inverse quantization operation result Inverse discrete cosine transform output and inverse discrete cosine coefficient distribution information are extracted and output, and an inverse discrete cosine transform unit for outputting the decoded moving image, a frame memory unit for storing the decoded moving image, and the inverse discrete cosine A motion compensating unit that reproduces a predictively encoded frame and outputs the frame to the frame memory by performing motion compensation using the converted output and the output from the frame memory unit, and the frame memory Performs filter operation on the decoded moving picture output from the moving picture signal decoding apparatus according to claim 1, characterized in that it is constituted by a filter unit to output the removed noise. 3. The filter unit uses a filter control unit that determines a filter coefficient and a filter coefficient output from the filter control unit, and performs a filter operation on the decoded moving image from the filter memory unit. The moving picture signal decoding apparatus according to claim 2, wherein the moving picture signal decoding apparatus is configured by a filter calculation unit that performs 4. The filter control unit is configured to update and output a filter coefficient for each frame using the coding parameter information, the quantization information, and the inverse discrete cosine coefficient distribution information. The moving picture signal decoding apparatus according to claim 2 or 3.