JP2011166357A - Image encoding device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image encoding device that reduces an influence of burst noise without performing detailed quantization. <P>SOLUTION: A quantization unit 8 determines whether or not an input image has flat texture by referring to a component of a DCT coefficient output from a conversion unit 6, and quantizes an AC component and a DC component of the DCT coefficient when the input image does not has flat texture, or corrects the DC component of the DCT coefficient and quantizes the AC component of the DCT coefficient and a DC component after the correction when the input image has flat texture. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to an image encoding apparatus that encodes an image signal in units of macroblocks.

FIG. 10 is a block diagram showing a conventional image encoding device disclosed in Non-Patent Document 1. In FIG.
In the image encoding device of FIG. 10, the AVC / H. H.264 (ISO / IEC 14496-10 | ITU-T H.264) is encoded.
Hereinafter, processing contents of the conventional image coding apparatus will be described.

AVC / H. In H.264 (hereinafter, referred to as “AVC”), each frame of an image signal indicating an input image is divided into 16 × 16 macroblocks and input to an image encoding device.
The image signal for each macroblock is composed of a luminance signal 16 × 16 pixels and a color difference signal 8 × 8 pixels corresponding to the luminance signal 16 × 16 pixels.

When the intra prediction unit 101 of the image encoding device receives an image signal in units of macroblocks, the local decoding signal stored in the intra prediction memory 112 is used, and the luminance signal constituting the image signal and An intra prediction process is performed on the color difference signal.
In AVC, as an intra prediction process for a luminance signal, there are 9 types with a 4 × 4 block as a unit, 9 types with a 8 × 8 block as a unit, and 16 × 16 blocks as a unit. Four types are defined as intra prediction processes for color difference signals, and the intra prediction unit 101 determines an intra prediction method to be used from these defined intra prediction methods, and the intra prediction is performed. Intra-prediction processing is performed by the method.
When the intra prediction unit 101 performs intra prediction processing on an input signal to generate a prediction image, the intra prediction unit 101 outputs an intra prediction signal indicating the prediction image to the selection switch 104, and the intra prediction method indicating the determined intra prediction method Information is output to the entropy encoding unit 115.

When an image signal in units of macroblocks is input, the motion detection unit 102 detects a motion between the image signal and the locally decoded image signal stored in the motion compensation prediction frame memory 114, and indicates the motion The vector is output to the motion compensation prediction unit 103 and the entropy encoding unit 115.
When the motion compensation prediction unit 103 receives a motion vector from the motion detection unit 102, the motion compensation prediction unit 103 generates a prediction image using the motion vector and a locally decoded image signal stored in the motion compensation prediction frame memory 114, and the prediction image Is output to the selection switch 104.

When the selection switch 104 receives the intra prediction signal from the intra prediction unit 101 and the motion compensation prediction signal from the motion compensation prediction unit 103, for example, under the instruction of the encoding control unit 107, the selection switch 104 receives the intra prediction signal or the motion compensation prediction signal. Is selected as a prediction signal, and the prediction signal is output to the subtractor 105 and the adder 111.
The selection switch 104 also outputs prediction signal selection information indicating which of the intra prediction signal or the motion compensation prediction signal has been selected to the entropy encoding unit 115.

The subtractor 105 calculates a difference between the image signal in units of macroblocks and the prediction signal output from the selection switch 104, and outputs a prediction difference signal indicating the difference to the conversion unit 106.
Upon receiving the prediction difference signal from the subtractor 105, the conversion unit 106 performs a discrete cosine transform (DCT) on the prediction difference signal, thereby converting the prediction difference signal from a spatial domain signal to a time domain signal. The DCT coefficient indicating the result of the discrete cosine transform is output to the quantization unit 108.
When the quantization unit 108 receives the DCT coefficient from the conversion unit 106, the quantization unit 108 quantizes the DCT coefficient according to the quantization parameter output from the encoding control unit 107, and dequantizes the quantization coefficient indicating the quantized DCT coefficient. Output to the encoding unit 109 and the entropy encoding unit 115.

Upon receiving the quantization coefficient from the quantization unit 108, the inverse quantization unit 109 performs an inverse quantization process corresponding to the quantization process of the quantization unit 108 according to the quantization parameter output from the encoding control unit 107. By performing the quantization, the quantization coefficient is inversely quantized, and the DCT coefficient corresponding to the DCT coefficient output from the conversion unit 106 (dequantization result of the quantization coefficient) is output to the inverse conversion unit 110.
When the inverse transform unit 110 receives the DCT coefficient from the inverse quantization unit 109, the inverse transform unit 110 performs the inverse discrete cosine transform (inverse DCT) on the DCT coefficient, thereby returning the prediction difference signal from the time domain signal to the spatial domain signal. The prediction error signal indicating the inverse discrete cosine transform result (a signal corresponding to the prediction difference signal output from the subtractor 105) is output to the adder 111.

When the adder 111 receives the prediction signal from the selection switch 104 and receives the prediction error signal from the inverse transform unit 110, the adder 111 calculates the local decoded signal by adding the prediction signal and the prediction error signal, and the local decoded signal is calculated. Store in the intra prediction memory 112.
The loop filter 113 performs a deblocking filter process on the local decoded signal stored in the intra prediction memory 112, and uses the local decoded image signal, which is the local decoded signal after the filter process, as a motion compensated prediction frame memory 114. To store.

The entropy coding unit 115 outputs the intra prediction method information output from the intra prediction unit 101, the motion vector output from the motion detection unit 102, the prediction signal selection information output from the selection switch 104, and the output from the coding control unit 107. The quantized parameters output from the quantization unit 108 and the quantized coefficient are entropy encoded, and encoded data indicating the encoding result is output.
Note that the entropy encoding unit 115 outputs information not described above in the encoded data as necessary.

The conventional image coding apparatus is configured as described above. In the conversion unit 106 and the inverse conversion unit 110 of the image coding apparatus, when MPEG-2 or MPEG-4 is applied, discrete cosine transform (DCT) is performed. When inverse discrete cosine transform (inverse DCT) is used and AVC is applied, integer transform obtained by transforming DCT is used so that transform / inverse transform can be performed only by integer calculation.
Hereinafter, the conversion process in the case of using 8 × 8 block size DCT will be described.
FIG. 3 is an explanatory diagram showing indexes of 64 transform coefficients obtained by the transform unit.
In FIG. 3, “DC” is a DC component in the conversion coefficient, and “AC” is an AC component in the conversion coefficient.

FIG. 4 is an explanatory diagram illustrating an example of an 8 × 8 block image signal input to the conversion unit.
In the example of FIG. 4, the value of 63 pixels in the image signal is “100”, but only the value of one pixel is “164” and burst noise is superimposed.
When such a block is subjected to discrete cosine transform (DCT) defined in MPEG-2, DCT coefficients as shown in FIG. 5 are obtained.

Since the DC component takes a value that is eight times the average value of the 64 pixels in the block, when the values of the 64 pixels are all “100”, the value of the DC component is “800”. And 63 AC component values are all zero.
However, when an image signal as shown in FIG. 4 is input, all 63 AC components become non-zero due to burst noise of one pixel. Therefore, the DC component also takes a value of “808” instead of “800”.

In the image encoding device disclosed in the following Patent Document 1, when generating encoded data by quantizing the DCT coefficients as described above, it is determined that the texture is flat and the quantization level is reduced. Like to do.
In this case, since fine quantization is performed, burst noise can also be reproduced with high quality, and a decoded image signal almost the same as the image signal before encoding can be obtained. Since it is necessary to encode the component or a number of AC components close to it, the amount of code increases, and data compression cannot be realized.

Conversely, if coarse quantization is performed, the amount of code can be reduced, but a large amount of coding noise is generated in the decoded image signal.
In particular, considering the case where all 63 AC components are quantized to zero, the value of the DC component is “808”, so the pixel value of the decoded image signal is not “100” but “101”. Therefore, the pixel value of the entire block is increased by 1.
When such a block is included in a wide flat texture, coding noise becomes very conspicuous.

Japanese Patent Laid-Open No. 9-172641

MPEG-4 AVC (ISO / IEC 14496-10) / ITU-TH H.264 standard

  Since the conventional image coding apparatus is configured as described above, when coarse quantization is performed when burst noise is included in the image signal indicating the input image, the image signal indicating the input image is Even if the noise is only one pixel, the entire block is affected by the noise, and there is a problem that significant coding noise is generated.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to provide an image coding apparatus that can reduce the influence of burst noise without performing fine quantization.

  In the image encoding device according to the present invention, the quantization unit refers to the component of the transform coefficient output from the signal conversion unit, determines whether or not the input image has a flat texture, and the input image is flat. If the texture is not a smooth texture, the AC component and DC component of the conversion coefficient are quantized. If the input image is a flat texture, the DC component of the conversion coefficient is corrected, and the AC component and the corrected component of the conversion coefficient are corrected. The DC component is quantized.

  According to this invention, the quantizing unit refers to the component of the transform coefficient output from the signal converting unit to determine whether or not the input image has a flat texture, and the input image must have a flat texture. For example, the AC component and DC component of the conversion coefficient are quantized, and if the input image has a flat texture, the DC component of the conversion coefficient is corrected, and the AC component and the corrected DC component of the conversion coefficient are quantized. Therefore, there is an effect that the influence of burst noise can be reduced without performing fine quantization.

It is a block diagram which shows the image coding apparatus by Embodiment 1 of this invention. It is a block diagram which shows the inside of the quantization part 8 of the image coding apparatus by Embodiment 1 of this invention. It is explanatory drawing which shows the index of 64 conversion coefficients obtained by the conversion part. 6 is an explanatory diagram illustrating an example of an 8 × 8 block image signal input to a conversion unit 6; FIG. It is explanatory drawing which shows an example of the DCT coefficient output from the conversion part. It is a block diagram which shows the image coding apparatus by Embodiment 3 of this invention. It is a block diagram which shows the inside of the quantization part 16 of the image coding apparatus by Embodiment 3 of this invention. It is a block diagram which shows the image coding apparatus by Embodiment 4 of this invention. FIG. 6 is an explanatory diagram showing a locally decoded image signal when inverse quantization and inverse transformation are performed on the DCT coefficient of FIG. 5 output from the transform unit 6 after a slightly coarse quantization. 1 is a configuration diagram illustrating a conventional image encoding device disclosed in Non-Patent Document 1. FIG.

Embodiment 1 FIG.
In the first embodiment, the image encoding apparatus applies AVC, and each frame of the image signal indicating the input image is divided into 16 × 16 macroblocks and input to the image encoding apparatus. To do.
The image signal for each macroblock is composed of a luminance signal 16 × 16 pixels and a color difference signal 8 × 8 pixels corresponding to the luminance signal 16 × 16 pixels.

FIG. 1 is a block diagram showing an image coding apparatus according to Embodiment 1 of the present invention.
In FIG. 1, when an intra prediction unit 1 inputs an image signal in units of macroblocks, a luminance signal and a color difference signal constituting the image signal are used using a local decoded signal stored in the intra prediction memory 12. Intra prediction processing is performed for.
In AVC, as an intra prediction process for a luminance signal, there are 9 types with a 4 × 4 block as a unit, 9 types with a 8 × 8 block as a unit, and 16 × 16 blocks as a unit. Four types are defined as intra prediction processes for color difference signals, and the intra prediction unit 1 determines an intra prediction method to be used from the defined intra prediction methods, and the intra prediction is performed. Intra-prediction processing is performed by the method.
When the intra prediction unit 1 generates a prediction image by performing an intra prediction process on the input signal, the intra prediction unit 1 outputs an intra prediction signal indicating the prediction image to the selection switch 4, and intra prediction method information indicating the determined intra prediction method Is output to the entropy encoding unit 15.

When the motion detection unit 2 inputs an image signal in units of macroblocks, the motion detection unit 2 detects a motion between the image signal and the locally decoded image signal stored in the motion compensation prediction frame memory 14, and a motion vector indicating the motion Is output to the motion compensation prediction unit 3 and the entropy encoding unit 15.
The motion compensation prediction unit 3 generates a prediction image using the motion vector output from the motion detection unit 2 and the local decoded image signal stored in the motion compensation prediction frame memory 14, and motion compensated prediction indicating the prediction image Processing for outputting a signal to the selection switch 4 is performed.

The selection switch 4 selects, for example, either an intra prediction signal output from the intra prediction unit 1 or a motion compensated prediction signal output from the motion compensated prediction unit 3 as a prediction signal under the instruction of the encoding control unit 7. Then, a process of outputting the prediction signal to the subtracter 5 and the adder 11 is performed.
Further, the selection switch 4 performs a process of outputting prediction signal selection information indicating which of the intra prediction signal or the motion compensation prediction signal is selected to the entropy encoding unit 15.
The intra prediction unit 1, the motion detection unit 2, the motion compensation prediction unit 3, and the selection switch 4 constitute a predicted image generation unit.

The subtractor 5 calculates a difference between the image signal in units of macroblocks and the prediction signal output from the selection switch 4 and performs a process of outputting a prediction difference signal indicating the difference to the conversion unit 6. The subtracter 5 constitutes a prediction difference signal calculation unit.
The conversion unit 6 performs discrete cosine transform (DCT) on the prediction difference signal output from the subtractor 5 to convert the prediction difference signal from a spatial domain signal to a time domain signal, and shows the discrete cosine transformation result. A process of outputting the DCT coefficient (transform coefficient) to the quantization unit 8 is performed. The conversion unit 6 constitutes signal conversion means.
In this example, the transform unit 6 performs discrete cosine transform (DCT) on the prediction difference signal to convert the prediction difference signal from a spatial domain signal to a time domain signal. As long as the signal in the space domain is converted into the signal in the time domain, the prediction difference signal is not limited to the one subjected to the discrete cosine transform (DCT).

The encoding control unit 7 outputs a quantization parameter that is referred to when the quantization unit 8 and the inverse quantization unit 9 perform quantization / inverse quantization, and controls the code amount and the encoded image quality of the encoded data. Perform the process.
The quantization unit 8 quantizes the DCT coefficient output from the transform unit 6 according to the quantization parameter output from the encoding control unit 7, and uses the quantization DCT coefficient as a quantization coefficient, and the inverse quantization unit 9 and The process which outputs to the entropy encoding part 15 is implemented.
However, when the quantizing unit 8 quantizes the DCT coefficient output from the converting unit 6, the quantizing unit 8 refers to the component of the DCT coefficient output from the converting unit 6, and inputs the input image (for example, 8 × 8 to be encoded). Block) is a flat texture. If the input image is not a flat texture, the AC and DC components of the DCT coefficient are quantized. If the input image is a flat texture, the DCT The DC component of the coefficient is corrected, and the AC component and the corrected DC component of the DCT coefficient are quantized.
The encoding control unit 7 and the quantization unit 8 constitute quantization means.

The inverse quantization unit 9 inversely quantizes the quantization coefficient output from the quantization unit 8 in accordance with the quantization parameter output from the encoding control unit 7 and converts the DCT coefficient that is the inverse quantization result into an inverse conversion unit. 10 is executed.
The inverse transform unit 10 performs inverse discrete cosine transform (inverse DCT) on the DCT coefficient output from the inverse quantization unit 9, and returns the prediction difference signal from the time domain signal to the spatial domain signal, thereby obtaining the inverse discrete signal. A process of outputting a prediction error signal (a signal corresponding to the prediction difference signal output from the subtractor 5) indicating the cosine transform result to the adder 11 is performed.

The adder 11 calculates a local decoded signal by adding the prediction signal output from the selection switch 4 and the prediction error signal output from the inverse transform unit 10, and outputs the local decoded signal to the intra prediction memory 12. Implement the process.
The intra prediction memory 12 is a recording medium such as a RAM for storing the local decoded signal output from the adder 11.

The loop filter 13 performs deblocking filter processing on the local decoded signal stored in the intra prediction memory 12, and the local decoded image signal, which is the local decoded signal after the filter processing, is stored in the motion compensated prediction frame memory 14. Perform the output process.
The motion compensation prediction frame memory 14 is a recording medium such as a RAM for storing the locally decoded image signal output from the loop filter 13.

  The entropy encoding unit 15 includes the intra prediction method information output from the intra prediction unit 1, the motion vector output from the motion detection unit 2, the prediction signal selection information output from the selection switch 4, and the encoding control unit 7 The entropy encoding is performed on the quantization parameter output from, the quantization coefficient output from the quantization unit 8 and other necessary information, and the encoded data indicating the encoding result is output. The entropy encoding unit 15 constitutes variable length encoding means.

  In FIG. 1, an intra prediction unit 1, a motion detection unit 2, a motion compensation prediction unit 3, a selection switch 4, a subtracter 5, a conversion unit 6, a coding control unit 7, and a quantization unit, which are components of the image coding device. 8, each of the inverse quantization unit 9, the inverse transform unit 10, the adder 11, the loop filter 13, and the entropy encoding unit 15 has dedicated hardware (for example, a semiconductor integrated circuit on which a CPU is mounted, a one-chip microcomputer, etc. However, when the image encoding device is configured by a computer, an intra prediction unit 1, a motion detection unit 2, a motion compensation prediction unit 3, a selection switch 4, a subtractor 5, a program describing processing contents of the transform unit 6, the encoding control unit 7, the quantization unit 8, the inverse quantization unit 9, the inverse transform unit 10, the adder 11, the loop filter 13 and the entropy coding unit 15. The Stored in the memory of the computer, CPU of the computer may execute a program stored in the memory.

FIG. 2 is a block diagram showing the inside of the quantization unit 8 of the image coding apparatus according to Embodiment 1 of the present invention.
In FIG. 2, the signal determination unit 21 counts the number N _{AC of} AC components whose absolute values exceed a predetermined threshold S _th among the AC components in the DCT coefficients output from the conversion unit 6, and the number N _AC is the prescribed number N _{R (predetermined} number N _R is one or more positive numbers is predetermined) determines whether more, if the number N _AC is lower than the specified number N _R, the input image If the texture is recognized as a flat texture and the number _NAC is equal to or greater than the specified number N _R , a process of recognizing that the input image is a complex texture (recognizing that the texture is not a flat texture) is performed.

When the signal determination unit 21 recognizes that the input image has a flat texture, the DC component correction unit 22 determines the DCT coefficient according to the sum of the absolute values of the AC components in the DCT coefficient output from the conversion unit 6. A process of correcting the DC component is performed.
When the signal determination unit 21 recognizes that the input image has a complex texture, the quantization processing unit 23 outputs the DCT coefficient output from the conversion unit 6 according to the quantization parameter output from the encoding control unit 7. When the signal determination unit 21 recognizes that the input image has a flat texture, the encoding control unit 7 determines that the AC component and the DC component (DC component not corrected by the DC component correction unit 22) are quantized. In accordance with the output quantization parameter, a process of quantizing the AC component of the DCT coefficient output from the conversion unit 6 and the DC component after correction (the DC component corrected by the DC component correction unit 22) is performed.

Next, the operation will be described.
When the intra prediction unit 1 receives an image signal in units of macro blocks, the intra prediction unit 1 uses the local decoded signal stored in the intra prediction memory 12 to perform intra prediction on the luminance signal and the color difference signal constituting the image signal. Implement the process.
In AVC, as an intra prediction process for a luminance signal, there are 9 types with a 4 × 4 block as a unit, 9 types with a 8 × 8 block as a unit, and 16 × 16 blocks as a unit. Four types are defined as intra prediction processes for color difference signals, and the intra prediction unit 1 determines an intra prediction method to be used from the defined intra prediction methods, and the intra prediction method. Intra prediction processing is performed.
When the intra prediction unit 1 generates a prediction image by performing an intra prediction process on the input signal, the intra prediction unit 1 outputs an intra prediction signal indicating the prediction image to the selection switch 4 and indicates the determined intra prediction method. Information is output to the entropy encoding unit 15.
Since the prediction image generation process itself in the intra prediction unit 1 is a known technique, a detailed description thereof will be omitted.

When an image signal in units of macroblocks is input, the motion detection unit 2 detects a motion between the image signal and the locally decoded image signal stored in the motion compensation prediction frame memory 14, and a motion indicating the motion The vector is output to the motion compensation prediction unit 3 and the entropy encoding unit 15.
Since the motion vector detection process itself in the motion detector 2 is a known technique, a detailed description thereof will be omitted.

When the motion compensation prediction unit 3 receives the motion vector from the motion detection unit 2, the motion compensation prediction unit 3 generates a prediction image using the motion vector and the locally decoded image signal stored in the motion compensation prediction frame memory 14, and the prediction image Is output to the selection switch 4.
Since the prediction image generation process itself in the motion compensation prediction unit 3 is a known technique, a detailed description thereof is omitted.

When the selection switch 4 receives the intra prediction signal from the intra prediction unit 1 and the motion compensation prediction signal from the motion compensation prediction unit 3, for example, under the instruction of the encoding control unit 7, the selection switch 4 receives the intra prediction signal or the motion compensation prediction signal. Is selected as a prediction signal, and the prediction signal is output to the subtracter 5 and the adder 11.
Further, the selection switch 4 outputs prediction signal selection information indicating which of the intra prediction signal or the motion compensation prediction signal is selected to the entropy encoding unit 15.
When the image signal in units of macroblocks is input, the subtracter 5 calculates a difference between the image signal and the prediction signal output from the selection switch 4, and outputs a prediction difference signal indicating the difference to the conversion unit 6.

When receiving the prediction difference signal from the subtractor 5, the conversion unit 6 performs a discrete cosine transform (DCT) on the prediction difference signal and outputs a DCT coefficient indicating the result of the discrete cosine transformation to the quantization unit 8.
Here, FIG. 3 is an explanatory diagram showing indexes of 64 transform coefficients obtained by the transform unit 6.
In FIG. 6, “DC” is a DC component in the conversion coefficient, and “AC” is an AC component in the conversion coefficient.
At this time, when an 8 × 8 block image signal on which burst noise as shown in FIG. 4 is superimposed is input to the conversion unit 6, when the conversion unit 6 performs discrete cosine transform (DCT), FIG. DCT coefficients as shown are obtained.
Therefore, as described in the background art section, the quantizing unit 8 simply performs coarse quantization in order to suppress the increase in the code amount (the DCT coefficient quantization without correcting the DC component described later). ), A large amount of coding noise is generated in the decoded image signal.
Therefore, in the first embodiment, the quantization unit 8 performs the quantization of DCT coefficients as described below.

When the input image (8 × 8 block to be encoded) has a flat texture, among the DCT coefficients output from the conversion unit 6, all the absolute values of the 63 AC components are small values close to “0”. There is a tendency to take.
On the other hand, when the input image has a complicated texture, 63 AC components tend to include a plurality of AC components having large absolute values.
Therefore, when the signal determination unit 21 of the quantization unit 8 receives the DCT coefficient from the conversion unit 6, the number N _{AC of} AC components whose absolute value exceeds a predetermined threshold S _th among the AC components in the DCT coefficient. To determine whether the number N _AC is equal to or greater than a specified number N _R.
The signal determination unit 21 recognizes that the input image has a flat texture if the number N _{AC of} AC components whose absolute value exceeds the threshold S _th is less than the specified number N _R (N _AC <N _R ). .
On the other hand, if the number N _{AC of} AC components whose absolute value exceeds the threshold value S _th is equal to or greater than the specified number N _R (N _AC ≧ N _R ), the input image is recognized as a complex texture.

The DC component correction unit 22 of the quantization unit 8 recognizes that the input image has a flat texture by the signal determination unit 21 (for example, when a DCT coefficient as shown in FIG. 5 is output from the conversion unit 6. ) The value of the DC component is “808”, which is larger by “8” than “800” when no burst noise is included, so the value of the DC component is “800”. Such correction processing is performed.
When burst noise is included, from the experimental results, the sum of the absolute values of the 63 AC components in the DCT coefficient is proportional to the value at which the DC component value changes due to the influence of burst noise. I know.
In the case of an 8 × 8 block, by using a value obtained by dividing the sum of absolute values of AC components by “54.8348”, a DC component from which the influence of burst noise is removed can be obtained.

Therefore, the DC component correction unit 22 corrects the DC component of the DCT coefficient according to the sum of the absolute values of the AC components in the DCT coefficient output from the conversion unit 6, as shown in the following equation (1). I am doing so.
DC (correction value)
= DC (input value) + α · (sum of absolute values of AC components) ÷ 54.8348 (1)

However, “α” in equation (1) is a constant and takes a value of “+1” or “−1”, but cannot be determined only from the value of the AC component. The value of “α” is determined by comparison with the value of the DC component.
That is, since the flat region tends to be continuous, the value of the DC component also tends to be continuous.
Therefore, if the input DC component to be encoded is larger than the DC components of the surrounding encoded blocks, α = −1 is determined. Conversely, the DC component to be encoded is Is small, α = + 1 is determined.
It should be noted that adjacent left blocks and upper blocks are used as peripheral encoded blocks. In addition, if the DCT coefficient of the block on the left or upper right, or the block far away such as two blocks left or two blocks above is used, the reliability of the value of “α” can be obtained. Can increase the sex.

When the signal determination unit 21 recognizes that the input image has a complex texture, the quantization processing unit 23 of the quantization unit 8 converts the conversion unit 6 according to the quantization parameter output from the encoding control unit 7. Quantize the AC component and DC component (DC component not corrected by the DC component correction unit 22) of the DCT coefficient output from the DCT coefficient and use the quantized DCT coefficient as a quantization coefficient, and the inverse quantization unit 9 and the entropy code To the conversion unit 15.
On the other hand, when the signal determination unit 21 recognizes that the input image has a flat texture, according to the quantization parameter output from the encoding control unit 7, the AC component of the DCT coefficient output from the conversion unit 6 and The corrected DC component (the DC component corrected by the DC component correction unit 22) is quantized, and the quantized DCT coefficient is output to the inverse quantization unit 9 and the entropy encoding unit 15 as a quantization coefficient.

When the inverse quantization unit 9 receives the quantization coefficient from the quantization unit 8, the inverse quantization unit 9 inversely quantizes the quantization coefficient in accordance with the quantization parameter output from the encoding control unit 7, and the inverse quantization result is obtained. The DCT coefficient is output to the inverse transform unit 10.
When the inverse transform unit 10 receives the DCT coefficient from the inverse quantization unit 9, the inverse transform unit 10 performs inverse discrete cosine transform (inverse DCT) on the DCT coefficient and returns the prediction difference signal from the time domain signal to the spatial domain signal. Then, a prediction error signal indicating the inverse discrete cosine transform result (a signal corresponding to the prediction difference signal output from the subtractor 5) is output to the adder 11.

When the adder 11 receives the prediction signal from the selection switch 4 and receives the prediction error signal from the inverse transform unit 10, the adder 11 adds the prediction signal and the prediction error signal in order to prepare for the next encoding process. The signal is calculated, and the locally decoded signal is stored in the intra prediction memory 12.
When the adder 11 stores the local decoded signal in the intra prediction memory 12, the loop filter 13 performs a deblocking filter process on the local decoded signal, and a local decoded image signal which is a local decoded signal after the filter process Are stored in the motion compensation prediction frame memory 14.

  The entropy encoding unit 15 includes intra prediction scheme information output from the intra prediction unit 1, a motion vector output from the motion detection unit 2, prediction signal selection information output from the selection switch 4, and an encoding control unit. 7 is entropy-coded with the quantization parameter output from 7, the quantization coefficient output from the quantization unit 8, and other necessary information, and outputs encoded data indicating the encoding result.

  As apparent from the above, according to the first embodiment, the quantization unit 8 refers to the DCT coefficient component output from the conversion unit 6 to determine whether or not the input image has a flat texture. If the input image is not a flat texture, the AC and DC components of the DCT coefficient are quantized. If the input image is a flat texture, the DC component of the DCT coefficient is corrected and the DCT coefficient is corrected. Since the AC component and the DC component after correction are quantized, there is an effect that the influence of burst noise can be reduced without performing fine quantization.

In the first embodiment, the DC component correction unit 22 determines the value of “α” by comparing with the DC component values of the surrounding encoded blocks. The obtained DCT coefficient is divided by “8”, and DC ′ which is a result of the division and M pixels (M is 1) out of 64 pixels in the prediction difference signal supplied from the subtractor 5 to the conversion unit 6. The value of “α” may be determined by comparing the magnitude relationship of the above positive numbers).
DC ′ is an average value of 64 pixels in the block including burst noise, and the correction of the DC component correction unit 22 is correction to obtain an average value of 63 pixels excluding noise. When the value of DC ′ is larger than the value of the pixel that is not noise, α = −1 is determined. On the contrary, if the value of DC ′ is smaller, α = + 1 is determined. Make a decision.
However, when M = 1, noise is hit with a probability of 1/64. Therefore, in order to obtain “α” with high reliability, it is desirable that M ≧ 3 or more.

Embodiment 2. FIG.
In the first embodiment, the signal determination unit 21 of the quantization unit 8 includes the number N of AC components whose absolute values exceed the predetermined threshold S _th among the AC components in the DCT coefficients output from the conversion unit 6. counts the _AC, it is determined whether the number N or _AC is the specified number N _R above, if the number N _AC is lower than the specified number N _R, an input image is recognized as a flat texture However, in order to identify whether or not the input image has a flat texture, the input image is recognized as a complex texture if the number N _AC is equal to or greater than the specified number N _R. In addition, when the block located in the periphery of the block related to the image signal to be processed has a flat texture, the threshold value S _th or the specified number N _R is set as compared with the case where the block located in the periphery is not a flat texture. Big It is also possible to determine whether or not the block relating to the image signal has a flat texture.

Specifically, it is as follows.
In the first embodiment, one threshold value S _th is prepared in advance. In the second embodiment, a threshold value S _th1 and a threshold value S _th2 (S _th1 <S _th2 ) are prepared as the threshold value S _th. .
Here, for simplification of explanation, an example in which two threshold values _Sth are prepared will be described, but three or more threshold values may be prepared.

The signal determination unit 21 stores a determination result (determination result indicating whether or not the texture is a flat texture) of a block that has already been determined.
When the signal determination unit 21 determines whether the block to be encoded has a flat texture, among the blocks that have already been subjected to the determination process, the signal determination unit 21 selects a block located around the block to be encoded. Refer to the judgment result.
For example, the determination result of the block located right above the block to be encoded or the block located on the left side is referred to.

If the determination result of the blocks located in the vicinity indicates “flat texture”, there is a high possibility that the flat blocks are continuous.
Therefore, the signal determination unit 21 selects the larger threshold S _th2 of the two thresholds prepared in advance in order to increase the possibility of determining that the block to be encoded is “flat texture”. To do.
Then, the signal determination unit 21 counts the number N _{AC of} AC components whose absolute value exceeds the threshold value S _th2 and determines whether or not the number N _AC is equal to or greater than a specified number N _R.

On the other hand, if the determination result of the blocks located in the vicinity indicates “complex texture”, there is a high possibility that the complex blocks are continuous.
Therefore, the signal determination unit 21 selects the smaller threshold S _th1 of the two thresholds prepared in advance in order to increase the possibility of determining that the block to be encoded is a “complex texture”. To do.
Then, the signal determination unit 21 counts the number N _{AC of} AC components whose absolute value exceeds the threshold value S _th1 and determines whether or not the number N _AC is equal to or greater than a specified number N _R.

Here, an example in which the threshold value S _th1 and the threshold value S _th2 (S _th1 <S _th2 ) are prepared as the threshold value S _th is shown, but as the specified number N _R , the specified number N _R1 and the specified number N _R2 (N _R1 < N _R2 ) may be prepared.
If the determination result of the blocks located in the vicinity indicates that the block is a “flat texture”, the signal determination unit 21 may determine that the block to be encoded is a “flat texture”. In order to increase it, the larger prescribed number N _R2 is selected from the two prescribed numbers prepared in advance.
Then, the signal determination unit 21 counts the number N _{AC of} AC components whose absolute value exceeds the threshold value S _th and determines whether or not the number N _AC is equal to or greater than a specified number N _R2 .

If the determination result of the blocks located in the vicinity indicates that the block is a “complex texture”, the signal determination unit 21 may determine that the block to be encoded is a “complex texture”. In order to increase it, the smaller prescribed number N _R1 is selected from the two prescribed numbers prepared in advance.
Then, the signal determination unit 21 counts the number N _{AC of} AC components whose absolute value exceeds the threshold value S _th and determines whether or not the number N _AC is equal to or greater than the specified number N _R1 .

As is apparent from the above, according to the second embodiment, when the block located around the encoding target block is a flat texture, the block located around is a complex texture. In some cases, the threshold value S _th or the specified number N _R is set to a large value and it is determined whether or not the block to be encoded is a flat texture. Therefore, compared to the first embodiment, There exists an effect which can improve the determination precision of whether it is a flat texture.

In the second embodiment, the reference is made to the determination result of the block that is spatially close to the encoding target block. You may make it refer to the determination result of the block which is present.
For example, refer to the block determination result in the previous picture that has already been encoded (for example, the determination result of the block in the same position as the block to be encoded in the picture). May be.
In this case, since the determination result becomes continuous in the screen, the reliability of the determination result can be improved.

Embodiment 3 FIG.
6 is a block diagram showing an image coding apparatus according to Embodiment 3 of the present invention. In the figure, the same reference numerals as those in FIG.
The quantization unit 16 quantizes the DCT coefficient output from the transform unit 6 according to the quantization parameter output from the encoding control unit 7, and uses the quantization DCT coefficient as the quantization coefficient, and the inverse quantization unit 9 and The process which outputs to the entropy encoding part 15 is implemented.
However, the quantization unit 16 refers to the quantized DCT coefficient component to determine whether or not the input image (eg, 8 × 8 block to be encoded) has a flat texture, and the input image is If the texture is not flat, the quantized DCT coefficient is output to the variable length coding unit 15, and if the input image is flat, the DC component of the quantized DCT coefficient is corrected and quantized. The AC component of the later DCT coefficient and the corrected DC component are output to the variable length coding unit 15.
Note that the encoding control unit 7 and the quantization unit 16 constitute quantization means.

FIG. 7 is a block diagram showing the inside of the quantization unit 16 of the image coding apparatus according to Embodiment 3 of the present invention.
In FIG. 7, the quantization processing unit 31 performs a process of quantizing the DCT coefficient output from the conversion unit 6 according to the quantization parameter output from the encoding control unit 7.
Further, the quantization processing unit 31 counts the number N _AC of non-zero AC components among the AC components in the quantized DCT coefficient, and the number _AC is a predetermined number N _R (the predetermined number N _R is predetermined). If the number N _AC is less than the specified number N _R , the input image is recognized as a flat texture, and the number N _AC is If it is equal to or greater than the prescribed number N _R , a process of recognizing that the input image is a complex texture (recognizing that it is not a flat texture) is performed.

  When the quantization processing unit 31 recognizes that the input image has a complex texture, the DC component correction unit 32 uses the DCT coefficient quantized by the quantization processing unit 31 as a quantization coefficient, and the inverse quantization unit 9 and the entropy. When output to the encoding unit 15 and the quantization processing unit 31 recognizes that the input image has a flat texture, according to the sum of the absolute values of the AC components in the DCT coefficients quantized by the quantization processing unit 31 Then, the DC component of the DCT coefficient is corrected, and the AC component and the corrected DC component in the DCT coefficient are output to the inverse quantization unit 9 and the entropy encoding unit 15.

Next, the operation will be described.
Compared with the image encoding device in FIG. 1, the quantization unit 8 is the same except that the quantization unit 16 is replaced. Therefore, only the processing content of the quantization unit 16 will be described.

When the quantization processing unit 31 of the quantization unit 16 receives the DCT coefficient from the conversion unit 6, the quantization processing unit 31 quantizes the DCT coefficient according to the quantization parameter output from the encoding control unit 7.
Further, the quantization processing unit 31 counts the number N _AC of non-zero AC components among the AC components in the quantized DCT coefficient, and the number is a predetermined number N _R (the predetermined number N _R is predetermined). It is determined whether or not it is greater than or equal to one or more positive numbers.
If the number N _AC of non-zero AC components is less than the specified number N _R (N _AC <N _R ), the quantization processing unit 31 has a flat texture for the input image (8 × 8 block to be encoded). Recognize that there is.
On the other hand, if the number N _AC of non-zero AC components is equal to or greater than the specified number N _R (N _AC ≧ N _R ), the input image is recognized as a complex texture.

When the quantization processing unit 31 recognizes that the input image has a complex texture, the DC component correction unit 32 of the quantization unit 16 reverses the DCT coefficient quantized by the quantization processing unit 31 as a quantization coefficient. It outputs to the quantization part 9 and the entropy encoding part 15.
On the other hand, when the input image is recognized as a flat texture by the quantization processing unit 31, the absolute value of the AC component in the DCT coefficient quantized by the quantization processing unit 31 as shown in the following equation (2). The DC component of the DCT coefficient is corrected according to the sum of the values, and the AC component and the corrected DC component in the DCT coefficient are output to the inverse quantization unit 9 and the entropy encoding unit 15.
DC (correction value)
= DC (input value) + α · (sum of absolute values of AC components) ÷ 54.8348 (2)

However, “α” in equation (2) is a constant and takes a value of “+1” or “−1”, but cannot be determined only from the value of the AC component. The value of “α” is determined by comparison with the value of the DC component.
That is, since the flat region tends to be continuous, the value of the DC component also tends to be continuous.
Therefore, if the input DC component to be encoded is larger than the DC components of the surrounding encoded blocks, α = −1 is determined. Conversely, the DC component to be encoded is Is small, α = + 1 is determined.
It should be noted that adjacent left blocks and upper blocks are used as peripheral encoded blocks. In addition, if the DCT coefficient of the block on the left or upper right, or the block far away such as two blocks left or two blocks above is used, the reliability of the value of “α” can be obtained. Can increase the sex.

As is apparent from the above, according to the third embodiment, the quantization unit 16 determines whether or not the input image has a flat texture by referring to the component of the DCT coefficient after quantization. If the image is a complex texture, the quantized DCT coefficient is output to the variable length coding unit 15, and if the input image is a flat texture, the DC component of the quantized DCT coefficient is corrected, Since the configuration is such that the AC component of the DCT coefficient after quantization and the DC component after correction are output to the variable-length encoding unit 15, the burst can be performed without performing detailed quantization as in the first embodiment. There is an effect that the influence of noise can be reduced.
In the third embodiment, since the quantization processing unit 31 determines whether or not the input image has a flat texture, it is necessary to provide a signal determination unit 21 as shown in FIG. There is an effect that the configuration can be simplified.

Embodiment 4 FIG.
FIG. 8 is a block diagram showing an image coding apparatus according to Embodiment 4 of the present invention. In the figure, the same reference numerals as those in FIG.
The quantization unit 17 quantizes the DCT coefficient output from the transform unit 6 according to the quantization parameter output from the encoding control unit 7, and uses the quantization DCT coefficient as the quantization coefficient, and the inverse quantization unit 9 and The process which outputs to the entropy encoding part 15 is implemented.
However, the quantization unit 17 refers to the DCT coefficient component output from the conversion unit 6 to determine whether or not the input image (for example, the 8 × 8 block to be encoded) has a flat texture. If the input image is not a flat texture, the AC component and DC component of the DCT coefficient are quantized. If the input image is a flat texture, the DC component of the DCT coefficient is quantized and all the DCT coefficients in the DCT coefficient are quantized. Is quantized to zero.
The encoding control unit 7 and the quantization unit 17 constitute quantization means.

Next, the operation will be described.
Compared with the image encoding device in FIG. 1, the quantization unit 8 is the same except that it is replaced with the quantization unit 17, so only the processing content of the quantization unit 17 will be described.
FIG. 9 is an explanatory diagram showing a locally decoded image signal when inverse quantization and inverse transformation are performed on the DCT coefficient of FIG. 5 output from the transformation unit 6 after a slightly coarse quantization. is there.
FIG. 9 shows an example in which encoding processing is performed using 26 non-zero AC components. However, although a large amount of code is required, most pixels in the block are original. It can be seen that this is different from the pixel value of.

Therefore, in the fourth embodiment, all AC components that can be regarded as obtained by converting burst noise as shown in FIG. 5 are quantized to zero.
That is, when the quantizing unit 17 receives the DCT coefficient from the converting unit 6, the quantizing unit 17 refers to the component of the DCT coefficient and determines whether or not the input image (8 × 8 block to be encoded) has a flat texture. judge.
For example, as in the first embodiment, among the AC components in the DCT coefficient output from the conversion unit 6, the number N _{AC of} AC components whose absolute value exceeds a predetermined threshold S _th is less than the specified number N _R. If so, the input image is recognized as a flat texture.
On the other hand, if the number N _{AC of} AC components whose absolute value exceeds the predetermined threshold S _th is equal to or greater than the specified number N _R , the input image is recognized as a complex texture.

When the quantization unit 17 recognizes that the input image has a complex texture, the quantization unit 17 quantizes the AC component and the DC component of the DCT coefficient output from the conversion unit 6, and performs inverse quantization using the quantized DCT coefficient as a quantization coefficient. To the encoding unit 9 and the entropy encoding unit 15.
When the quantizing unit 17 recognizes that the input image has a flat texture, the quantizing unit 17 quantizes the DC component of the DCT coefficient output from the converting unit 6 and quantizes all the AC components in the DCT coefficient to zero. The quantized DCT coefficient is output to the inverse quantization unit 9 and the entropy coding unit 15 as a quantization coefficient.

As a result, all AC components become zero and are decoded from only the DC component, so that the decoded pixel values do not vary as shown in FIG. 9, and a flat decoded image signal is obtained. Thus, the influence of coding noise can be reduced.
However, when performing fine quantization that can encode all AC components, high-quality images can be obtained using a large amount of code by performing normal encoding processing without using this method. What is necessary is just to encode.

  As apparent from the above, according to the fourth embodiment, the quantization unit 17 refers to the DCT coefficient component output from the conversion unit 6 to determine whether or not the input image has a flat texture. If the input image is not a flat texture, the AC and DC components of the DCT coefficient are quantized. If the input image is a flat texture, the DC component of the DCT coefficient is quantized and the DCT Since all the AC components in the coefficients are quantized to zero, the effect of reducing the influence of burst noise can be achieved without performing fine quantization as in the first embodiment.

In Embodiments 1 to 4, it is assumed that the block size for the conversion process is 8 × 8, but the present invention is not limited to this. For example, even if the block size is 4 × 4 or 16 × 16 Good.
Also, the block size may be a rectangular block such as 8 × 16 or 16 × 8, or may be a conversion process using a three-dimensional block extending in the time direction instead of two-dimensional. Often, similar effects can be obtained by similar processing.

In the international standardized MPEG-2 and MPEG-4, the intra prediction unit 1, the intra prediction memory 12, and the loop filter 13 are not provided, and a part of the processing is different from that of the AVC. By applying to the unit 6 and the inverse transform unit 10, it is possible to obtain the same effect as in the case of AVC.
In addition, in the international standardized JPEG, there is no motion detection unit 2, motion compensation prediction unit 3, selection switch 4, subtractor 5, inverse quantization unit 9, inverse transform unit 10, motion compensation prediction frame memory 14, and the like. By applying the present invention, it is possible to obtain the same effect as the above AVC.

  DESCRIPTION OF SYMBOLS 1 Intra prediction part (prediction image generation means), 2 motion detection part (prediction image generation means), 3 motion compensation prediction part (prediction image generation means), 4 selection switch (prediction image generation means), 5 subtractor (prediction difference) Signal calculation means), 6 conversion section (signal conversion means), 7 encoding control section (quantization means), 8, 16, 17 quantization section (quantization means), 9 inverse quantization section, 10 inverse conversion section, DESCRIPTION OF SYMBOLS 11 Adder, 12 Intra prediction memory, 13 Loop filter, 14 Motion compensation prediction frame memory, 15 Entropy encoding part (variable length encoding means), 21 Signal determination part, 22, 32 DC component correction part, 23, 31 quantization processing unit, 101 intra prediction unit, 102 motion detection unit, 103 motion compensation prediction unit, 104 selection switch, 105 subtractor, 106 conversion unit, 107 encoding control , 108 quantization unit 109 inverse quantization unit 110 inverse transform unit, 111 an adder, 112 intra-prediction memory, 113 loop filter, 114 a motion compensated prediction frame memory, 115 entropy encoding unit.

Claims (8)

  A prediction image generating unit that generates a prediction image from an image signal indicating an input image and outputs a prediction signal indicating the prediction image; a difference between the prediction signal output from the prediction image generation unit and the image signal indicating the input image And a prediction difference signal calculation means for outputting a prediction difference signal indicating the difference, and a conversion process for converting the prediction difference signal output from the prediction difference signal calculation means from a spatial domain signal to a time domain signal. And a signal conversion unit that outputs a conversion coefficient indicating the conversion result, a quantization unit that quantizes the conversion coefficient output from the signal conversion unit, and a variable conversion coefficient quantized by the quantization unit In the image coding apparatus comprising variable length coding means for performing long coding, the quantization means refers to the component of the transform coefficient output from the signal transform means, and If the input image is not a flat texture, the AC component and the DC component of the conversion coefficient are quantized. If the input image is a flat texture, the conversion is performed. An image coding apparatus characterized by correcting a DC component of a coefficient and quantizing the AC component of the transform coefficient and the corrected DC component.   The quantization means determines that the input image has a flat texture if the number of AC components whose absolute value exceeds the threshold among the AC components in the conversion coefficient output from the signal conversion means is less than a specified number. 2. The image encoding apparatus according to claim 1, wherein if the number of AC components whose absolute value exceeds a threshold value is equal to or greater than a predetermined number, the input image is determined not to be a flat texture.   The quantization means corrects the DC component of the conversion coefficient according to the sum of the absolute values of the AC components in the conversion coefficient output from the signal conversion means if the input image is a flat texture. The image encoding device according to claim 1 or 2.   The quantization means increases the threshold value or the prescribed number when the block located around the block related to the image signal to be processed has a flat texture than when the block located around the block is not a flat texture. 3. The image coding apparatus according to claim 2, wherein a value is set to determine whether or not a block related to the image signal has a flat texture.   A prediction image generating unit that generates a prediction image from an image signal indicating an input image and outputs a prediction signal indicating the prediction image; a difference between the prediction signal output from the prediction image generation unit and the image signal indicating the input image And a prediction difference signal calculation means for outputting a prediction difference signal indicating the difference, and a conversion process for converting the prediction difference signal output from the prediction difference signal calculation means from a spatial domain signal to a time domain signal. And a signal conversion unit that outputs a conversion coefficient indicating the conversion result, a quantization unit that quantizes the conversion coefficient output from the signal conversion unit, and a variable conversion coefficient quantized by the quantization unit In the image coding apparatus comprising variable length coding means for performing long coding, the quantization means refers to a component of the transform coefficient after quantization, and the input image has a flat texture. If the input image is not a flat texture, the quantized transform coefficient is output to the variable length coding means. If the input image is a flat texture, the quantized An image coding apparatus characterized in that the DC component of the transform coefficient is corrected, and the quantized transform coefficient AC component and the corrected DC component are output to the variable length coding means.   The quantization means determines that the input image has a flat texture if the number of non-zero AC components in the AC component in the transformed transform coefficient after quantization is less than a specified number, and determines the non-zero AC component. 6. The image encoding apparatus according to claim 5, wherein if the number is equal to or greater than a predetermined number, the input image is determined not to be a flat texture.   The quantization means corrects the DC component of the transform coefficient according to the sum of absolute values of the AC components in the transform coefficient after quantization if the input image is a flat texture. The image encoding device according to claim 6.   A prediction image generating unit that generates a prediction image from an image signal indicating an input image and outputs a prediction signal indicating the prediction image; a difference between the prediction signal output from the prediction image generation unit and the image signal indicating the input image And a prediction difference signal calculation means for outputting a prediction difference signal indicating the difference, and a conversion process for converting the prediction difference signal output from the prediction difference signal calculation means from a spatial domain signal to a time domain signal. And a signal conversion unit that outputs a conversion coefficient indicating the conversion result, a quantization unit that quantizes the conversion coefficient output from the signal conversion unit, and a variable conversion coefficient quantized by the quantization unit In the image coding apparatus comprising variable length coding means for performing long coding, the quantization means refers to the component of the transform coefficient output from the signal transform means, and If the input image is not a flat texture, the AC component and the DC component of the conversion coefficient are quantized. If the input image is a flat texture, the conversion is performed. An image coding apparatus characterized by quantizing a DC component of a coefficient and quantizing all AC components in the transform coefficient to zero.

Images