JP2012191500A - Image encoder, image decoder, image encoding method, and image decoding method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve a subjective image quality of a macro block in which a complicated picture and a flat picture are mixed, without causing large increase in the amount of information of a quantization parameter to be transmitted to an image decoder.SOLUTION: An image encoder determines a quantization parameter QP for each macro block. The image encoder specifies, according to an image feature amount of a macro block, a rectangular region within the macro block the quantization parameter QP of which is to be changed, calculates a shift value dQP of the quantization parameter, and calculates from the quantization parameter QP and the shift value dQP of the macro block, the quantization parameter QPt of the orthogonal transformation block included in the rectangular region.

Description

  The present invention relates to an image encoding device and an image encoding method for encoding an image, and an image decoding device and an image decoding method for decoding an image encoded by the image encoding device.

FIG. 13 is an explanatory diagram showing the state of the quantization process described in Non-Patent Document 1 below, for example.
For example, H.264 is an international standard for image coding. In H.264, a frame or field to be encoded is divided into blocks called “macroblocks” composed of pixels of 16 pixels × 16 lines, and a quantization parameter QP that can be changed for each macroblock is used. A transform coefficient that is an orthogonal transform result of a difference image between a macroblock image and a predicted image is quantized.
However, when quantization is performed on an orthogonal transform block of 4 pixels × 4 lines or an orthogonal transform block of 8 pixels × 8 lines inside the macroblock, a common quantization parameter QP is used for all orthogonal transform blocks. To quantize.
FIG. 14 is an explanatory diagram showing the orthogonal transform blocks included in the macroblock, and the numbers in the orthogonal transform blocks indicate the order of encoding.

H. In H.264, quantization is performed using a common quantization parameter QP in all orthogonal transform blocks. However, when a complex pattern and a flat pattern are mixed in a macroblock, orthogonal transform including each pattern is performed. If the quantization parameter of the block can be changed, the subjective image quality of the part can be improved.
For example, when a macroblock is divided into 4 pixels × 4 lines as shown in FIG. 14 and the macroblock is an image as shown in FIG. 15, an orthogonal transform block (with a flat background drawn) ( Orthogonal transform blocks with numbers 0, 1, 2, 4, 5, 8, 10, 11, 14, and 15) and orthogonal transform blocks with detailed patterns (numbers 3, 6, 7, 9, 12 and 13 orthogonal transform blocks), if the quantization parameter of each orthogonal transform block can be changed, the subjective image quality of the portion can be improved.

H. In H.264, the size of a macroblock is 16 pixels × 16 lines, but in the future, an encoding method using a macroblock having a size larger than 16 pixels × 16 lines may be considered.
In such a system, there is a high possibility that a pattern having different characteristics is mixed in a macroblock. Therefore, it becomes more useful if the quantization parameter of each orthogonal transform block can be changed.

Patent Document 1 below discloses a method of changing a quantization parameter in units of orthogonal transform blocks of 4 pixels × 4 lines or 8 pixels × 8 lines.
However, in the method disclosed in Patent Document 1, if the quantization parameter is changed for each orthogonal transform block in the macroblock, it is necessary to transmit the quantization parameter in units of orthogonal transform blocks.

WO2008 / 126135A1

H. is an international standard for image coding. H.264 (MPEG-4AVC) Recommendation (ITU-T Rec. H.264)

  Since the conventional image coding apparatus is configured as described above, by changing the quantization parameter for each orthogonal transform block in the macroblock, a macroblock in which a complex picture and a flat picture are mixed The subjective image quality can be improved. However, since it is necessary to transmit the quantization parameter in units of orthogonal transform blocks, there is a problem that the amount of information of the quantization parameter transmitted to the image decoding apparatus side is greatly increased.

The present invention has been made to solve the above-described problems, and does not cause a significant increase in the amount of quantization parameter information to be transmitted to the image decoding apparatus side. An object of the present invention is to obtain an image encoding device and an image encoding method capable of improving the subjective image quality of a macroblock.
Another object of the present invention is to obtain an image decoding device and an image decoding method applicable to the image encoding device.

  The image encoding device according to the present invention generates a prediction image by performing prediction processing for each macroblock constituting an input image, and generates a difference image between the image of the macroblock and the prediction image Means, a quantization parameter determining means for determining a quantization parameter for each macroblock, a rectangular region in the macroblock whose quantization parameter is changed according to the image feature amount of the macroblock, and the quantization parameter A quantization parameter changing unit that calculates a quantization parameter of the orthogonal transformation block included in the rectangular region from the quantization parameter of the macroblock and the above-described transition value, and a difference image generation unit The generated difference image is orthogonally transformed, and among the transform coefficients that are the orthogonal transformation results of the difference image, The transform coefficient of the orthogonal transform block included in the region is quantized using the quantization parameter of the orthogonal transform block calculated by the quantization parameter changing unit, and the inside of the macro block not included in the rectangular region is quantized. The transform coefficient of the orthogonal transform block is provided with a quantization unit that performs quantization using the macroblock quantization parameter determined by the quantization parameter determination unit, and the entropy encoding unit is quantized by the quantization unit. Transform coefficients, macroblock quantization parameters determined by the quantization parameter determining means, quantization parameter transition values calculated by the quantization parameter changing means, rectangular area specifying information for specifying a rectangular area, and difference image generating means The predicted image generation information used when the predicted image is generated by It is obtained so as to generate a bit stream by Ntoropi coding.

  According to the present invention, the difference image generation means for generating a prediction image by performing prediction processing for each macroblock constituting the input image, and generating a difference image between the image of the macroblock and the prediction image, the macro Quantization parameter determination means for determining a quantization parameter for each block, and a rectangular area in the macroblock whose quantization parameter is changed according to the image feature amount of the macroblock, and a change value of the quantization parameter The difference generated by the difference parameter generating means and the quantization parameter changing means for calculating and calculating the quantization parameter of the orthogonal transform block included in the rectangular area from the quantization parameter of the macro block and the transition value The image is orthogonally transformed, and the transform coefficient that is the result of orthogonal transformation of the difference image is included in the rectangular area. The transform coefficient of the orthogonal transform block is quantized using the quantization parameter of the orthogonal transform block calculated by the quantization parameter changing means, and the transform of the orthogonal transform block in the macroblock not included in the rectangular area is performed. The coefficient is provided with quantization means for quantizing using the quantization parameter of the macroblock determined by the quantization parameter determination means, and the entropy encoding means is a transform coefficient, a quantum quantized by the quantization means. The macroblock quantization parameter determined by the quantization parameter determination means, the transition value of the quantization parameter calculated by the quantization parameter change means, the rectangular area specifying information for specifying the rectangular area, and the prediction image is generated by the difference image generation means The information for generating the predicted image used when Since a bitstream is generated by generating a macroblock that contains a complex pattern and a flat pattern without causing a significant increase in the amount of quantization parameter information transmitted to the image decoding apparatus. There is an effect that the subjective image quality can be improved.

It is a block diagram which shows the image coding apparatus by Embodiment 1 of this invention. It is a flowchart which shows the processing content (image coding method) of the image coding apparatus by Embodiment 1 of this invention. It is a block diagram which shows the image decoding apparatus by Embodiment 1 of this invention. It is a flowchart which shows the processing content (image decoding method) of the image decoding apparatus by Embodiment 1 of this invention. FIG. 16 is an explanatory diagram illustrating an example of a variance value of pixels in units of 4 pixels × 4 lines when the macroblock is an image as illustrated in FIG. 15. It is explanatory drawing which shows the some rectangular area from which the position or magnitude | size in a macroblock differs. It is explanatory drawing which shows the rectangular area information memorize | stored in the rectangular area information storage part. It is explanatory drawing which shows eight rectangular areas in a macroblock. It is explanatory drawing which shows eight rectangular area information memorize | stored in the rectangular area information storage part. It is explanatory drawing which shows the example of transmission of the small number corresponding to a rectangular area number. It is explanatory drawing which shows an example of the macroblock in which a complicated picture and a flat picture are mixed. It is explanatory drawing which shows the example which is covering two complicated pictures using two rectangular areas. It is explanatory drawing which shows the mode of the quantization process described in the nonpatent literature 1. It is explanatory drawing which shows the orthogonal transformation block contained in the macroblock. It is explanatory drawing which shows an example of the macroblock in which a complicated picture and a flat picture are mixed.

Embodiment 1 FIG.
FIG. 1 is a block diagram showing an image coding apparatus according to Embodiment 1 of the present invention.
In FIG. 1, H.264, which is an international standard for image coding systems. An example of an image encoding device applied to H.264 (MPEG-4AVC) is shown, but an image encoding device applied to another image encoding method may be used.
In FIG. 1, an encoding unit 1 performs an encoding process on an input image in units of macroblocks under the instruction of the encoding control unit 2.
In addition to the process of determining the quantization parameter QP for each macroblock, the encoding control unit 2 specifies a rectangular area in the macroblock whose quantization parameter QP is to be changed, and calculates the transition value dQP of the quantization parameter QP. In addition, processing such as calculating the quantization parameter QPt of the orthogonal transform block included in the rectangular area from the quantization parameter QP of the macroblock and the transition value dQP is performed. The encoding control unit 2 constitutes a quantization parameter determining unit and a quantization parameter changing unit.

The intra prediction unit 11 of the encoding unit 1 selects an optimal intra prediction mode for each macroblock constituting the input image, and the local decoded image that has already been encoded according to the intra prediction mode A prediction image is generated from a macroblock local decoded image around the macroblock, and a process of generating a difference image between the macroblock image and the prediction image is performed.
For each macroblock constituting the input image, the motion search unit 12 calculates a motion vector by performing a motion search by comparing the image of the macroblock with the locally decoded image stored in the frame memory 23. Perform the process.

The motion-compensated prediction unit 13 uses the motion vector calculated by the motion search unit 12 to locally decode the image stored in the frame memory 23 in units of partitions that are the same as the macroblock or smaller in size than the macroblock. A process for generating a predicted image is performed by performing a motion compensated prediction process for.
The differentiator 14 performs a process of generating a difference image for each macroblock constituting the input image by obtaining a difference between the image of the macroblock and the predicted image generated by the motion compensation prediction unit 13.
The intra prediction unit 11, the motion search unit 12, the motion compensation prediction unit 13, and the differentiator 14 constitute a difference image generation unit.

The intra / inter determination unit 15 compares the predicted image generated by the intra prediction unit 11 with the image of the macroblock, compares the predicted image generated by the motion compensation prediction unit 13 with the image of the macroblock, and compares both. A process of determining an optimum predicted image with reference to the result is performed.
In addition, the intra / inter determination unit 15 outputs a determination result indicating an optimal prediction image to the switch 16 and the entropy encoding unit 24, and if the optimal prediction image is a prediction image generated by the intra prediction unit 11, The predicted image is output to the adder 21, and if the optimal predicted image is a predicted image generated by the motion compensation prediction unit 13, a process of outputting the predicted image to the adder 21 is performed.

  If the determination result output from the intra / inter determination unit 15 indicates that the prediction image generated by the intra prediction unit 11 is an optimal prediction image, the switch 16 generates a difference generated by the intra prediction unit 11. If an image is selected and output to the orthogonal transform unit 17 and the prediction image generated by the motion compensation prediction unit 13 indicates that it is an optimal prediction image, the difference image generated by the subtractor 14 is selected. The process of outputting to the orthogonal transform unit 17 is performed.

The orthogonal transform unit 17 performs a process of performing orthogonal transform on the difference image output from the switch 16 and outputting a transform coefficient, which is an orthogonal transform result of the difference image, to the quantization unit 18.
The quantization unit 18 quantizes the transform coefficient output from the orthogonal transform unit 17 by using the quantization parameter output from the quantization control unit 33 of the encoding control unit 2 for each orthogonal transform block in the macroblock. Perform the process.
That is, among the transform coefficients output from the orthogonal transform unit 17, the quantization unit 18 uses the transform coefficients of the orthogonal transform block included in the rectangular area specified by the quantization control unit 33 of the encoding control unit 2. Quantization is performed using the quantization parameter QPt of the orthogonal transform block calculated by the quantization control unit 33, and the transform control unit of the orthogonal transform block in the macroblock not included in the rectangular region is quantized. A process of quantization using the quantization parameter QP of the macroblock determined by 33 is performed.
Note that the orthogonal transform unit 17 and the quantization unit 18 constitute quantization means.

The inverse quantization unit 19 inversely quantizes the transform coefficient quantized by the quantization unit 18 using the same quantization parameter as that of the quantization unit 18 for each orthogonal transform block in the macroblock, and the orthogonal transform unit 17 The process which outputs the transformation coefficient equivalent to the transformation coefficient output from (2) to the inverse orthogonal transformation part 20 is implemented.
The inverse orthogonal transform unit 20 performs inverse orthogonal transform on the transform coefficient output from the inverse quantization unit 19 to output a difference image (inverse orthogonal transform result) corresponding to the difference image output from the switch 16 to the adder 21. Perform the process.

The adder 21 adds the difference image output from the inverse orthogonal transform unit 20 and the prediction image selected by the intra / inter determination unit 15 to generate a locally decoded image.
The deblocking filter unit 22 performs a deblocking filter process on the local decoded image generated by the adder 21, compensates for distortion due to compression, and stores the local decoded image after distortion compensation in the frame memory 23. To do.
The frame memory 23 is a recording unit that stores a locally decoded image after distortion compensation.

The entropy encoding unit 24, the transform coefficient quantized by the quantization unit 18, the rectangular region number (rectangular region specifying information) output from the quantization control unit 33 of the encoding control unit 2, and the macroblock quantization parameter The transition value dQP of the QP and the quantization parameter QP, the determination result output from the intra / inter determination unit 15, and the prediction image generation information used when the prediction image is generated (from the intra / inter determination unit 15) If the output determination result indicates that the prediction image generated by the intra prediction unit 11 is an optimal prediction image, the intra prediction mode selected by the intra prediction unit 11 and the motion compensation prediction unit 13 generate the prediction image. If the predicted image thus obtained indicates that it is the optimum predicted image, the type of partition calculated by the motion search unit 12 and its partition Entropy coding index and a motion vector) of Ishon carries out a process of generating a bitstream. The entropy encoding unit 24 constitutes entropy encoding means.
The transmission buffer 25 is connected to an external transmission means such as a line or a storage medium, for example, and temporarily stores the bit stream generated by the entropy encoding unit 24 and then performs processing to output the bit stream. To do.

The image feature amount calculation unit 31 of the encoding control unit 2 performs a process of calculating the image feature amount of the macro block for each macro block constituting the input image.
The rectangular area information storage unit 32 is used as information about a plurality of rectangular areas having different positions or sizes in the macroblock (hereinafter referred to as “rectangular area information”). Of the rectangular area), the horizontal / vertical size of the plurality of rectangular areas (for example, the number of pixels in the horizontal direction of the rectangular area, the number of pixels in the vertical direction), and the rectangular area number for identifying the plurality of rectangular areas Is remembered.
Here, an example is shown in which the horizontal / vertical direction sizes of a plurality of rectangular areas are stored, but instead of the horizontal / vertical direction sizes, the end point coordinates of the plurality of rectangular areas (for example, rectangles in a macroblock) The coordinates at the lower right of the area) may be stored.

For each macroblock constituting the input image, the quantization control unit 33 stores the bitstream buffer amount stored in the transmission buffer 25, the target code amount for each macroblock, and the bitstream actually generated. Processing for determining the quantization parameter QP of the macroblock according to the code amount or the like is performed.
Further, the quantization control unit 33 groups the orthogonal transform blocks in the macroblock according to the image feature value calculated by the image feature value calculation unit 31 (for example, an orthogonal transform block on which a complicated pattern is drawn). And a group of orthogonal transform blocks on which a flat pattern is drawn), and a rectangular area in the macroblock whose quantization parameter is to be changed (for example, an orthogonal transform block on which a complex pattern is drawn) And a process of selecting a rectangular area closest to the identified rectangular area from the rectangular areas indicated by the rectangular area information stored in the rectangular area information storage unit 32 is performed.

Further, the quantization control unit 33 calculates the transition value dQP of the quantization parameter QP of the macroblock using the image feature amount calculated by the image feature amount calculation unit 31, and the quantization parameter QP and the transition of the macroblock. From the value dQP, a process of calculating the quantization parameter QPt of the orthogonal transform block included in the selected rectangular area is performed.
The quantization control unit 33 outputs the quantization parameter QP of the macroblock and the quantization parameter QPt of the orthogonal transform block included in the rectangular region to the quantization unit 18, and identifies the selected rectangular region. The number, the quantization parameter QP of the macroblock, and the transition value dQP of the quantization parameter QP are output to the entropy encoding unit 24.

In the example of FIG. 1, each processing unit (intra prediction unit 11, motion search unit 12, motion compensation prediction unit 13, difference unit 14, and the like of the encoding unit 1 and the encoding control unit 2, which are components of the image encoding device, Intra / inter determination unit 15, switch 16, orthogonal transform unit 17, quantization unit 18, inverse quantization unit 19, inverse orthogonal transform unit 20, adder 21, deblocking filter unit 22, entropy encoding unit 24, image feature It is assumed that each of the quantity calculation unit 31 and the quantization control unit 33) is configured by dedicated hardware (for example, a semiconductor integrated circuit on which a CPU is mounted or a one-chip microcomputer). The image encoding device may be configured by a computer.
In this case, a program describing the processing contents of each processing unit of the encoding unit 1 and the encoding control unit 2 is stored in the memory of the computer, and the CPU of the computer executes the program stored in the memory You just have to do it.
FIG. 2 is a flowchart showing the processing contents (image coding method) of the image coding apparatus according to Embodiment 1 of the present invention.

3 is a block diagram showing an image decoding apparatus according to Embodiment 1 of the present invention.
In FIG. 3, the reception buffer 51 performs a process of receiving the bit stream generated by the image encoding device in FIG. 1 and outputting the bit stream to the entropy decoding unit 52.
The entropy decoding unit 52 entropy-decodes the bit stream output from the reception buffer 51, and dequantizes the transform coefficient quantized by the quantization unit 18 of the image encoding device in FIG. The inverse quantization parameter generation unit 54 outputs the rectangular area number (rectangular area identification information), the macroblock quantization parameter QP, and the quantization parameter QP shift value dQP output from the quantization control unit 33 to the quantization unit 55. And the process of outputting the determination result of the intra / inter determination unit 15 to the switch 59 is performed.
In addition, if the determination result of the intra / inter determination unit 15 indicates that the prediction image generated by the intra prediction unit 11 is an optimal prediction image, the entropy decoding unit 52 is intra information that is prediction image generation information. If the prediction mode is output to the intra prediction image generation unit 57 and indicates that the prediction image generated by the motion compensation prediction unit 13 is an optimal prediction image, the type of partition that is the prediction image generation information, A process of outputting the partition index and the motion vector to the inter prediction image generation unit 58 is performed.
The entropy decoding unit 52 constitutes entropy decoding means.

The rectangular area information storage unit 53 is a recording unit that stores the same rectangular area information as the rectangular area information storage unit 32 in the image encoding device of FIG.
That is, the rectangular area information storage unit 53 uses, as rectangular area information about a plurality of rectangular areas having different positions or sizes in the macroblock, the origin coordinates of the plurality of rectangular areas (for example, the upper left coordinates of the rectangular areas in the macroblock ), The horizontal / vertical size of the plurality of rectangular regions (for example, the number of pixels in the horizontal direction of the rectangular region, the number of pixels in the vertical direction), and a rectangular region number for identifying the plurality of rectangular regions.
Here, an example is shown in which the horizontal / vertical direction sizes of a plurality of rectangular areas are stored, but instead of the horizontal / vertical direction sizes, the end point coordinates of the plurality of rectangular areas (for example, rectangles in a macroblock) The coordinates at the lower right of the area) may be stored.

The inverse quantization parameter generation unit 54 reads out the start point coordinates and horizontal / vertical direction sizes of the rectangular region corresponding to the rectangular region number output from the entropy decoding unit 52 from the rectangular region information storage unit 53, and includes them in the rectangular region The orthogonal transform block is identified, and the process of calculating the quantization parameter QPt of the orthogonal transform block from the quantization parameter QP and the transition value dQP of the macroblock output from the entropy decoding unit 52 is performed.
Further, when the inverse quantization parameter generation unit 54 calculates the quantization parameter QPt of the orthogonal transform block, a process of outputting the quantization parameter QPt of the orthogonal transform block and the quantization parameter QP of the macroblock to the inverse quantization unit 55 is performed. carry out.
The inverse quantization parameter generation unit 54 constitutes a quantization parameter calculation unit.

Of the transform coefficients output from the entropy decoding unit 52, the inverse quantization unit 55 outputs from the inverse quantization parameter generation unit 54 the transform coefficients of the orthogonal transform block included in the rectangular region corresponding to the rectangular region number. The inverse quantization block QPt is used to perform inverse quantization, and the transform coefficient of the orthogonal transform block in the macroblock not included in the rectangular region is output from the inverse quantization parameter generation unit 54. A process of performing inverse quantization using the quantization parameter QP of the macroblock is performed.
The inverse orthogonal transform unit 56 performs inverse orthogonal transform on the transform coefficient after inverse quantization output from the inverse quantization unit 55, so that a difference corresponding to the difference image output from the switch 16 in the image encoding device of FIG. Processing for outputting an image (inverse orthogonal transform result) to the adder 60 is performed.
The inverse quantization unit 55 and the inverse orthogonal transform unit 56 constitute inverse quantization means.

When the intra prediction mode is output from the entropy decoding unit 52, the intra predicted image generation unit 57 selects the image of FIG. 1 from the decoded image of the surrounding macroblock that has already been decoded by the adder 60 according to the intra prediction mode. A process of generating a prediction image corresponding to the prediction image generated by the intra prediction unit 11 in the encoding device is performed.
When the motion vector is output from the entropy decoding unit 52, the inter prediction image generation unit 58 uses the motion vector to store the frame memory 62 in units of partitions that are the same as the macroblock or smaller in size than the macroblock. By performing the motion compensation prediction process on the stored decoded image, a process of generating a prediction image corresponding to the prediction image generated by the motion compensation prediction unit 13 in the image encoding device of FIG. 1 is performed.

If the determination result of the intra / inter determination unit 15 output from the entropy decoding unit 52 indicates that the prediction image generated by the intra prediction unit 11 is the optimal prediction image, the switch 59 generates an intra prediction image. If the prediction image generated by the unit 57 is output to the adder 60 and indicates that the prediction image generated by the motion compensation prediction unit 13 is an optimal prediction image, the prediction image generated by the inter prediction image generation unit 58 is generated. A process of outputting the predicted image to the adder 60 is performed.
The intra predicted image generation unit 57, the inter predicted image generation unit 58, and the switch 59 constitute a predicted image generation unit.

The adder 60 performs a process of adding the predicted image output from the switch 59 and the difference image output from the inverse orthogonal transform unit 56 to generate a decoded image. Note that the adder 60 constitutes a decoded image generating means.
The deblocking filter unit 61 performs a deblocking filter process on the decoded image generated by the adder 60 to compensate for distortion accompanying compression, and stores the decoded image after distortion compensation in the frame memory 62.

The frame memory 62 is a recording unit that stores a decoded image after distortion compensation.
The switch 63 performs a process of selecting the decoded image after distortion compensation by the deblocking filter unit 61 or the decoded image stored in the frame memory 62 in the display order and outputting the selected decoded image.

In the example of FIG. 3, an entropy decoding unit 52, an inverse quantization parameter generation unit 54, an inverse quantization unit 55, an inverse orthogonal transform unit 56, an intra prediction image generation unit 57, and an inter prediction image generation, which are components of the image decoding apparatus. Each of the unit 58, the switch 59, the adder 60, the deblocking filter unit 61, and the switch 63 is configured by dedicated hardware (for example, a semiconductor integrated circuit on which a CPU is mounted, or a one-chip microcomputer). However, the image encoding apparatus may be configured by a computer.
In this case, the entropy decoding unit 52, the inverse quantization parameter generation unit 54, the inverse quantization unit 55, the inverse orthogonal transform unit 56, the intra prediction image generation unit 57, the inter prediction image generation unit 58, the switch 59, the adder 60, A program describing the processing contents of the blocking filter unit 61 and the switch 63 may be stored in the memory of the computer, and the CPU of the computer may execute the program stored in the memory.
FIG. 4 is a flowchart showing the processing contents (image decoding method) of the image decoding apparatus according to Embodiment 1 of the present invention.

Next, the operation will be described.
First, the processing content of the image coding apparatus will be described.
The intra prediction unit 11 of the encoding unit 1 selects an optimal intra prediction mode for each macroblock constituting the input image, and surrounding macroblocks that have already been encoded according to the intra prediction mode. A predicted image is generated from the locally decoded image (step ST1 in FIG. 2). However, since a technique for generating a predicted image by selecting an optimal intra prediction mode is a known technique, a detailed description thereof is omitted (see, for example, H.264).
When the intra prediction unit 11 generates a prediction image from the locally decoded image, the intra prediction unit 11 generates a difference image between the macroblock image forming the input image and the prediction image, and outputs the difference image to the switch 16 (step ST2). . In addition, the optimal intra prediction mode is output to the entropy encoding unit 24.

For each macroblock constituting the input image, the motion search unit 12 performs a motion search by comparing the image of the macroblock and the locally decoded image stored in the frame memory 23, and obtains a motion vector. calculate. Since the motion vector calculation process is also a known technique, a detailed description thereof will be omitted (see, for example, H.264).
When the motion search unit 12 calculates a motion vector, the motion compensation prediction unit 13 stores the motion vector in the frame memory 23 in units of partitions that are the same as the macroblock or smaller in size than the macroblock. A predicted image is generated by performing motion compensation prediction processing on a local decoded image (step ST3). Since the motion compensation prediction process is also a known technique, a detailed description thereof is omitted (for example, see H.264).
When the motion compensation prediction unit 13 generates a prediction image, the difference unit 14 generates a difference image by obtaining a difference between the image of the macroblock and the prediction image, and outputs the difference image to the switch 16 (step ST4). .

When the intra prediction unit 11 and the motion compensation prediction unit 13 generate a prediction image, the intra / inter determination unit 15 compares the prediction image generated by the intra prediction unit 11 with the image of the macroblock and the motion compensation prediction unit 13. Is compared with the image of the macroblock, and the optimum prediction image is determined with reference to the comparison result of both.
As a method for determining the optimum predicted image, a generally widely used method is to evaluate a value obtained by accumulating the absolute difference value or the square difference value of the pixel at the same pixel position of the macroblock image and the predicted image for each block. Used as a value. In this case, it is determined that the predicted image with the smaller evaluation value is the optimal predicted image.
When the intra / inter determination unit 15 determines an optimal predicted image, the intra / inter determination unit 15 outputs the predicted image to the adder 21. In addition, the determination result indicating the optimum predicted image is output to the switch 16 and the entropy encoding unit 24.

If the determination result output from the intra / inter determination unit 15 indicates that the prediction image generated by the intra prediction unit 11 is the optimal prediction image (step ST5), the switch 16 determines that the intra prediction unit 11 Is selected and output to the orthogonal transform unit 17 (step ST6).
On the other hand, if the determination result indicates that the prediction image generated by the motion compensation prediction unit 13 is an optimal prediction image (step ST5), the difference image generated by the differentiator 14 is selected and orthogonally transformed. It outputs to the part 17 (step ST7).

  Upon receiving the difference image from the switch 16, the orthogonal transform unit 17 orthogonally transforms the difference image in units of orthogonal transform blocks (4 × 4 blocks or 8 × 8 blocks) as shown in FIG. The transform coefficient that is the orthogonal transform result of the difference image is output to the quantization unit 18 (step ST8).

The image feature amount calculation unit 31 of the encoding control unit 2 calculates the image feature amount of the macro block for each macro block constituting the input image.
For example, as the image feature amount of the macroblock, for each orthogonal transform block constituting the macroblock, a variance value of pixels of the orthogonal transform block is calculated.
An orthogonal transform block having a small variance value represents a block in which a flat pattern is drawn, and an orthogonal transform block having a large variance value represents a block in which a fine pattern is drawn.
For example, when the macroblock is divided into 4 pixels × 4 lines as shown in FIG. 14, if the macroblock is an image as shown in FIG. 15, the numbers are 0, 1, 2, 4, 5, The variance values of the orthogonal transform blocks of 8, 10, 11, 14, and 15 are decreased, and the variance values of the orthogonal transform blocks of numbers 3, 6, 7, 9, 12, and 13 are increased.
FIG. 5 is an explanatory diagram illustrating an example of a dispersion value of pixels in units of 4 pixels × 4 lines when the macroblock is an image as illustrated in FIG. 15.

For each macroblock constituting the input image, the quantization control unit 33 stores the bitstream buffer amount stored in the transmission buffer 25, the target code amount for each macroblock, and the bitstream actually generated. The quantization parameter QP of the macroblock is determined according to the code amount of (step ST9).
As a process for determining the quantization parameter QP of the macroblock, for example, there is a method called TM5 used in the MPEG-2 verification test (for example, the Journal of Television Society, April 1995, Vol 49, No. 4). See).

When the image feature amount calculation unit 31 calculates the image feature amount of the macro block, the quantization control unit 33 groups the orthogonal transform blocks in the macro block according to the image feature amount, and changes the quantization parameter. A rectangular area in the macroblock is specified.
For example, an area including an orthogonal transformation block in which a complicated pattern is drawn is divided into a group of orthogonal transformation blocks in which a complicated picture is drawn and a group of orthogonal transformation blocks in which a flat picture is drawn. Then, it is specified as a rectangular area in the macroblock whose quantization parameter is to be changed.
As described above, since the variance value is small in a flat background and the variance value is large in a fine pattern portion, for example, the variance values are separated as 0 to 9, 10 to 19,. Grouping is performed according to which classification the variance value of belongs to.
Alternatively, the average of the variance values of the 16 orthogonal transform blocks is obtained, and the magnitudes of the average value and the variance value of each orthogonal transform block are compared to classify each orthogonal transform block into two groups.

Here, as shown in FIG. 6, the rectangular area information storage unit 32 stores rectangular area information related to a plurality of rectangular areas having different positions or sizes in the macroblock. In the example of FIG. 6, three rectangular areas are represented.
That is, in the rectangular area information storage unit 32, as shown in FIG. 7, the origin coordinates of a plurality of rectangular areas (for example, the upper left coordinates of the rectangular area in the macroblock) and the horizontal sizes of the plurality of rectangular areas are displayed. In addition, the vertical size (for example, the number of pixels in the horizontal direction of the rectangular region, the number of pixels in the vertical direction) and a rectangular region number for identifying a plurality of rectangular regions are stored.

When the quantization control unit 33 groups the orthogonal transform blocks in the macroblock and specifies a rectangular region in the macroblock whose quantization parameter is to be changed, the quantization control unit 33 stores the rectangle stored in the rectangular region information storage unit 32. A rectangular area closest to the specified rectangular area is selected from the rectangular areas indicated by the area information.
For example, as shown in FIGS. 8 and 9, eight rectangular areas in the macroblock are defined in advance, and rectangular area numbers for identifying the eight rectangular areas are stored in the rectangular area information storage unit 32. If the area indicated by the bold frame in FIG. 5 is specified as a rectangular area by the quantization control unit 33, the rectangle whose rectangular area number is “4” is the rectangular area closest to the rectangular area. A region is selected.
Note that the quantization control unit 33 stores the rectangular area information of the rectangular area selected for each macroblock, and ranks the rectangular areas in descending order of the number of rectangular areas selected in the frame. . According to the order, the rectangular area information of the rectangular area to be held in the rectangular area information storage unit 32 in the next frame is determined.

The quantization control unit 33 identifies a rectangular region in the macroblock whose quantization parameter is to be changed, and among the rectangular regions indicated by the rectangular region information stored in the rectangular region information storage unit 32, the identified rectangular region Is selected, the transition value dQP of the quantization parameter QP of the macroblock is calculated using the image feature value calculated by the image feature value calculation unit 31.
For example, an average value of variance values of orthogonal transform blocks included in the selected rectangular area is calculated, and a larger transition value dQP is obtained as the average value is larger.

When the quantization control unit 33 calculates the transition value dQP of the quantization parameter QP of the macroblock, it is included in the selected rectangular area from the quantization parameter QP and the transition value dQP of the macroblock as shown below. The quantization parameter QPt of the orthogonal transform block being calculated is calculated (step ST10).
QPt = QP + dQP
Here, dQP can take a positive or negative value. When the quantization parameter QPt of the orthogonal transform block, which is the calculation result, exceeds the range defined by the standard, limit processing is performed so as to be within the range.

When the quantization control unit 33 calculates the quantization parameter QPt of the orthogonal transform block, the quantization control unit 33 outputs the quantization parameter QPt of the orthogonal transform block and the quantization parameter QP of the macroblock to the quantization unit 18.
Further, the rectangular area number for identifying the selected rectangular area, the quantization parameter QP of the macroblock, and the transition value dQP of the quantization parameter QP are output to the entropy encoding unit 24.

The quantization unit 18 quantizes the transform coefficient output from the orthogonal transform unit 17 using the quantization parameter output from the quantization control unit 33 of the encoding control unit 2 for each orthogonal transform block in the macroblock. (Step ST11).
That is, the quantization unit 18 includes orthogonal transform blocks (for example, numbers 3 and 6 in FIG. 14) included in the rectangular region selected by the quantization control unit 33 among the transform coefficients output from the orthogonal transform unit 17. 7, 7, 9, 12, and 13) are quantized using the orthogonal transformation block quantization parameter QPt calculated by the quantization control unit 33.
On the other hand, transform of orthogonal transform blocks (for example, orthogonal transform blocks of 0, 1, 2, 4, 5, 8, 10, 11, 14, 15 in FIG. 14) in the macroblock not included in the rectangular area. The coefficients are quantized using the macroblock quantization parameter QP determined by the quantization control unit 33.

When the quantization unit 18 quantizes the transform coefficient of the difference image, the inverse quantization unit 19 uses the same quantization parameter as the quantization unit 18 for each orthogonal transform block in the macroblock. By dequantizing the transform coefficient quantized by the above, a transform coefficient corresponding to the transform coefficient output from the orthogonal transform unit 17 is output to the inverse orthogonal transform unit 20.
When receiving the transform coefficient from the inverse quantization unit 19, the inverse orthogonal transform unit 20 performs an inverse orthogonal transform on the transform coefficient, thereby obtaining a difference image (inverse orthogonal transform result) corresponding to the difference image output from the switch 16. Output to the adder 21.

The adder 21 adds the difference image output from the inverse orthogonal transform unit 20 and the predicted image selected by the intra / inter determination unit 15 to generate a local decoded image. The generated local decoded image is output to the deblocking filter unit 22 and is also output to the intra prediction unit 11 for generating a predicted image in intra prediction.
When the adder 21 generates a local decoded image, the deblocking filter unit 22 performs a deblocking filter process on the local decoded image to compensate for distortion due to compression, and the distortion-compensated local decoded image is stored in the frame memory 23. To store.

The entropy encoding unit 24 entropy encodes the following information to generate a bit stream (step ST12).
The transform coefficient quantized by the quantization unit 18 The macroblock quantization parameter QP
・ Rectangle area number ・ Transition value dQP of quantization parameter QP
The determination result of the intra / inter determination unit 15 The prediction image generation information used for generating the prediction image (the determination result output from the intra / inter determination unit 15 is the prediction image generated by the intra prediction unit 11) If it indicates that the prediction image is an optimal prediction image, the intra prediction mode selected by the intra prediction unit 11 and the prediction image generated by the motion compensation prediction unit 13 indicate that the prediction image is optimal. (Partition type calculated by the motion search unit 12, index and motion vector of the partition)

  When the entropy encoding unit 24 generates a bit stream, the transmission buffer 25 temporarily holds the bit stream, and then transmits the bit stream to the image decoding device via an external transmission unit such as a line. Send.

As apparent from the above, according to the first embodiment, the quantization parameter QP is determined for each macroblock, while the quantization parameter QP is changed according to the image feature amount of the macroblock. The region is specified, and the transition value dQP of the quantization parameter is calculated, and the quantization parameter QPt of the orthogonal transform block included in the rectangular region is calculated from the quantization parameter QP and the transition value dQP of the macroblock. The quantization control unit 33 is provided, and the quantization unit 18 converts the transform coefficient of the orthogonal transform block included in the rectangular area among the transform coefficients output from the orthogonal transform unit 17. Quantized using the quantization parameter QPt of the orthogonal transform block calculated by, and is not included in the rectangular area Since the transform coefficient of the orthogonal transform block in the macroblock is quantized using the macroblock quantization parameter QP determined by the quantization control unit 33, the quantization coefficient transmitted to the image decoding device side is configured. There is an effect that the subjective image quality of a macroblock in which a complex picture and a flat picture are mixed can be improved without causing a significant increase in the information amount of the conversion parameter.
That is, there is an effect that the image quality of the orthogonal transform block in which the image quality degradation is visually noticeable can be improved in the macro block.

Next, processing contents of the image decoding apparatus will be described.
The reception buffer 51 receives the bit stream transmitted from the image encoding device in FIG. 1 and outputs the bit stream to the entropy decoding unit 52.
When receiving the bit stream from the reception buffer 51, the entropy decoding unit 52 performs entropy decoding on the bit stream (step ST21 in FIG. 4).

The entropy decoding unit 52 outputs the transform coefficient quantized by the quantization unit 18 of the image encoding device in FIG. 1 among the decoded data of the bitstream to the inverse quantization unit 55 and outputs from the quantization control unit 33. The rectangular region number, the macroblock quantization parameter QP, and the quantized parameter QP transition value dQP are output to the inverse quantization parameter generation unit 54, and the determination result of the intra / inter determination unit 15 is output to the switch 59.
The entropy decoding unit 52 is the prediction image generation information if the determination result of the intra / inter determination unit 15 indicates that the prediction image generated by the intra prediction unit 11 is the optimal prediction image. If the intra prediction mode is output to the intra predicted image generation unit 57 and indicates that the predicted image generated by the motion compensation prediction unit 13 is an optimal predicted image, the type of partition that is the predicted image generation information, The partition index and motion vector are output to the inter prediction image generation unit 58.

When the inverse quantization parameter generation unit 54 receives the rectangular region number, the macroblock quantization parameter QP, and the quantized parameter QP transition value dQP from the entropy decoding unit 52, the inverse quantization parameter generation unit 54 corresponds to the rectangular region number from the rectangular region information storage unit 53. The starting point coordinates, the horizontal size and the vertical size of the rectangular area to be read are read out, and the orthogonal transformation block included in the rectangular area is specified.
Then, the inverse quantization parameter generation unit 54 calculates the quantization parameter QPt of the orthogonal transform block from the quantization parameter QP and the transition value dQP of the macroblock output from the entropy decoding unit 52 (step ST22).
QPt = QP + dQP
After calculating the quantization parameter QPt of the orthogonal transform block, the inverse quantization parameter generation unit 54 outputs the quantization parameter QPt of the orthogonal transform block and the quantization parameter QP of the macroblock to the inverse quantization unit 55.

When the inverse quantization unit 55 receives the quantization parameter QPt of the orthogonal transform block and the quantization parameter QP of the macro block from the inverse quantization parameter generation unit 54, the inverse quantization unit 55 performs the quantization of the orthogonal transform block for each orthogonal transform block in the macro block. The transform coefficient output from the entropy decoding unit 52 is inversely quantized using the quantization parameter QPt or the macroblock quantization parameter QP (step ST23).
In other words, the inverse quantization unit 55 uses the inverse quantization parameter generation unit for the transform coefficient of the orthogonal transform block included in the rectangular region corresponding to the rectangular region number among the transform coefficients output from the entropy decoding unit 52. Inverse quantization is performed using the quantization parameter QPt of the orthogonal transform block output from 54.
On the other hand, the transform coefficient of the orthogonal transform block in the macroblock not included in the rectangular area is subjected to inverse quantization using the macroblock quantization parameter QP output from the inverse quantization parameter generation unit 54. To implement.

  When the inverse orthogonal transform unit 56 receives the transform coefficient after the inverse quantization from the inverse quantization unit 55, the inverse orthogonal transform is performed on the transform coefficient so that the difference output from the switch 16 in the image coding apparatus in FIG. A difference image corresponding to the image (inverse orthogonal transform result) is output to the adder 60 (step ST24).

When the intra prediction image generation unit 57 receives the intra prediction mode from the entropy decoding unit 52, the intra prediction image generation unit 57 uses the image code of FIG. 1 according to the intra prediction mode from the decoded images of the surrounding macroblocks that have already been decoded by the adder 60. A prediction image corresponding to the prediction image generated by the intra prediction unit 11 in the conversion apparatus is generated (step ST25).
When receiving the motion vector from the entropy decoding unit 52, the inter prediction image generation unit 58 stores the motion vector in the frame memory 62 in units of partitions that are the same as the macroblock or smaller in size than the macroblock. By performing the motion compensation prediction process on the decoded image, a prediction image corresponding to the prediction image generated by the motion compensation prediction unit 13 in the image encoding device of FIG. 1 is generated (step ST26).

If the determination result of the intra / inter determination unit 15 output from the entropy decoding unit 52 indicates that the prediction image generated by the intra prediction unit 11 is the optimal prediction image, the switch 59 (step ST27). The predicted image generated by the intra predicted image generation unit 57 is output to the adder 60 (step ST28).
On the other hand, if the determination result of the intra / inter determination unit 15 indicates that the prediction image generated by the motion compensation prediction unit 13 is an optimal prediction image (step ST27), the determination result is generated by the inter prediction image generation unit 58. The predicted image is output to the adder 60 (step ST29).

The adder 60 adds the predicted image output from the switch 59 and the difference image output from the inverse orthogonal transform unit 56 to generate a decoded image, and outputs the decoded image to the deblocking filter unit 61 ( Step ST30).
When receiving the decoded image from the adder 60, the deblocking filter unit 61 performs a deblocking filter process on the decoded image, compensates for distortion due to compression, and stores the decoded image after distortion compensation in the frame memory 62. .
The switch 63 selects the decoded image generated by the adder 60 or the decoded image stored in the frame memory 62 in the display order, and outputs the selected decoded image.

As is apparent from the above, according to the first embodiment, the starting point coordinates and the horizontal / vertical direction sizes of the rectangular region corresponding to the rectangular region number output from the entropy decoding unit 52 are obtained from the rectangular region information storage unit 53. Read, specify the orthogonal transform block included in the rectangular area, and calculate the quantization parameter QPt of the orthogonal transform block from the quantization parameter QP and the transition value dQP of the macroblock output from the entropy decoding unit 52 The inverse quantization parameter generation unit 54 is provided, and the inverse quantization unit 55 converts the transform coefficient of the orthogonal transform block included in the rectangular region corresponding to the rectangular region number among the transform coefficients output from the entropy decoding unit 52. For the quantization parameter QPt of the orthogonal transform block output from the inverse quantization parameter generation unit 54, The transform coefficient of the orthogonal transform block in the macroblock that is quantized and not included in the rectangular area is inversely quantized using the macroblock quantization parameter QP output from the inverse quantization parameter generation unit 54. Since it comprised as mentioned above, there exists an effect which can improve the subjective image quality of the macroblock in which a complicated picture and a flat picture are mixed.
That is, there is an effect that the image quality of the orthogonal transform block in which the image quality degradation is visually noticeable can be improved in the macro block.

In this Embodiment 1, H.264 is used. An image encoding apparatus to which H.264 is applied has been shown. In H.264, the orthogonal transform block in the macroblock is fixed to one of 4 × 4 block and 8 × 8 block.
However, in future encoding schemes, a plurality of types of orthogonal transform blocks may be mixed in a macro block including orthogonal transform blocks of other sizes (for example, 16 × 16).

  In addition, when the macroblock size is expanded to 32 × 32 blocks that are larger than the current 16 × 16 blocks, it is assumed that fine patterns and flat patterns are often mixed in the same macroblock. Therefore, the ability to change the quantization parameter in units of orthogonal transform blocks is more effective in improving image quality.

In the first embodiment, for each macroblock constituting the input image, a transition value dQP for converting the quantization parameter QP of the macroblock into the quantization parameter QPt of the orthogonal transform block is included in the bitstream. Although what is transmitted to the image decoding apparatus side is shown, the transition value dQP may be transmitted for each picture or slice.
In this case, the same transition value dQP is applied to the rectangular areas in all the macroblocks in the picture or slice regardless of the rectangular area number, but the transition value dQP is set for each macroblock. Since transmission is not performed, the bit transmission amount is reduced accordingly.
Alternatively, a set of rectangular area number and transition value dQP may be transmitted for each picture or slice.
In this case, since the transition value dQP is not transmitted for each macroblock, the bit transmission amount is reduced correspondingly, and different quantization parameters can be applied depending on the type of the rectangular area.

In the first embodiment, the rectangular area information is stored in the rectangular area information storage unit 32. For example, the rectangular area information is obtained based on the statistics for each picture or slice obtained from the input image. May be updated (including new additions). The update of the rectangular area information is performed at a picture or slice cycle.
When the rectangular area information stored in the rectangular area information storage unit 32 is updated, the entropy encoding unit 24 performs entropy encoding on the updated rectangular area information.
The entropy encoding unit 24 multiplexes the updated rectangular area information with the header information of a picture or a slice and transmits the multiplexed information to the image decoding apparatus side.
Thereby, the rectangular area information storage unit 53 of the image decoding apparatus stores the updated rectangular area information.

In the first embodiment, the number of pieces of rectangular area information stored in the rectangular area information storage unit 32 is not particularly limited. However, when the number of pieces of rectangular area information increases, rectangular area numbers for identifying those rectangular areas are identified. The amount of bits used to represent is also increased.
Therefore, in advance, one or more rectangular area numbers used in a picture or slice are multiplexed with the header information of the picture or slice and transmitted to the image decoding apparatus side, and a small number corresponding to the rectangular area number is prepared.
In macroblock units, instead of transmitting a rectangular area number for identifying a rectangular area in a macroblock whose quantization parameter is to be changed, a small number corresponding to the rectangular area number is transmitted to the image decoding device side, thereby The amount of bits for representing the area number is reduced.
Specifically, it is as follows.

For example, as shown in FIG. 10, it is assumed that there are 32 rectangular area numbers from 0 to 31 stored in the rectangular area information storage unit 32.
In this case, 5 bits are required to represent 32 rectangular area numbers, and if an attempt is made to transmit a rectangular area number for each macroblock, it is necessary to transmit an information amount of 5 bits for each macroblock.
However, there are cases where the number of rectangular area numbers used in a picture or slice is not 32, but is limited to, for example, “1”, “5”, “13”, and “24”.
In such a case, four rectangular area numbers are multiplexed in advance with the header information of a picture or slice and transmitted to the image decoding apparatus side.

Also, small numbers corresponding to the four rectangular area numbers are prepared. In the example of FIG. 10, the small number “0” corresponding to the rectangular area number “1”, the small number “1” corresponding to the rectangular area number “5”, and the small number “2” corresponding to the rectangular area number “13”. A small number “3” corresponding to the rectangular area number “24” is prepared.
When the entropy encoding unit 24 transmits a rectangular area number for identifying a rectangular area in a macroblock whose quantization parameter is changed in units of macroblocks, the entropy encoding unit 24 corresponds to the rectangular area number instead of transmitting the rectangular area number. The small number to be multiplexed is multiplexed with the bit stream and transmitted to the image decoding apparatus side.
For example, if the rectangular area number for identifying the rectangular area whose quantization parameter is to be changed is “1”, the small number “0” is transmitted, and if the rectangular area number is “5”, the small number “1” is transmitted. If the rectangular area number is “13”, the small number “2” is transmitted. If the rectangular area number is “24”, the small number “3” is transmitted.
Since the small numbers “0” to “3” can be expressed by 2 bits, the transmission bit amount can be reduced.
Here, an example in which small numbers “0” to “3” are prepared is shown, but if the number is smaller than the rectangular area numbers “1”, “5”, “13”, and “24”, It is not limited to the small numbers “0” to “3”.

  In the first embodiment, the image feature quantity computing unit 31 computes the variance value of the pixels of the orthogonal transform block for each orthogonal transform block constituting the macro block as the image feature quantity of the macro block. However, the present invention is not limited to the one that calculates the variance value. For example, the distribution of the transform coefficient of the orthogonal transform block output from the orthogonal transform unit 17 is computed as the image feature amount of the macro block. Also good.

Embodiment 2. FIG.
In the first embodiment, the quantization control unit 33 identifies one rectangular region in the macroblock whose quantization parameter is to be changed, and the rectangular region indicated by the rectangular region information stored in the rectangular region information storage unit 32 In this example, the rectangular area closest to the identified one rectangular area is selected, but a plurality of rectangular areas in a macroblock whose quantization parameter is to be changed are specified, and a plurality of rectangular areas are specified. The rectangular areas closest to each other may be selected.
In this case, a quantization parameter transition value dQP is calculated for each of the selected plurality of rectangular regions, and the orthogonality included in the plurality of rectangular regions is calculated from the macroblock quantization parameter QP and the transition value dQP. A quantization parameter QPt for each transform block is calculated.
Specifically, it is as follows.

When the image feature amount calculation unit 31 calculates the image feature amount of the macro block, the quantization control unit 33 groups the orthogonal transform blocks in the macro block according to the image feature amount.
For example, the macroblock size is 64 pixels × 64 lines, etc. In image coding in which a macroblock having a size larger than H.264 can be selected, it is considered that the picture in the macroblock becomes more complicated.

Here, FIG. 11 is an explanatory diagram showing an example of a macro block in which a complex pattern and a flat pattern are mixed. In the example of FIG. 11, two complex patterns are drawn.
When a plurality of complicated patterns are drawn, it is difficult to efficiently cover with one rectangular area.
In such a case, it can be efficiently covered by using a plurality of rectangular areas. FIG. 12 shows an example in which two complicated areas are covered using two rectangular areas.

  When the quantization control unit 33 selects two rectangular regions (a rectangular region with a rectangular region number M and a rectangular region with a rectangular region number N), the quantization control unit 33 calculates a transition value dQP for the rectangular region with the rectangular region number M, The orthogonal transformation block included in the rectangular area is calculated by calculating the quantization parameter QPt of the orthogonal transformation block included in the rectangular area, and calculating the transition value dQP for the rectangular area having the rectangular area number N. The quantization parameter QPt is calculated.

The calculation process of the transition value dQP and the quantization parameter QPt is the same as that of the first embodiment. However, as shown in FIG. 12, a part of the rectangular area with the rectangular area number M and the rectangular area with the rectangular area number N Are overlapped, the transition value dQP of the overlapping part is selected from either the transition value dQP for the rectangular area with the rectangular area number M or the transition value dQP for the rectangular area with the rectangular area number N, The quantization parameter QPt of the overlapping portion is calculated.
For example, if there is an agreement between the image encoding device and the image decoding device to select the transition value dQP for the rectangular region with the larger rectangular region number, the transition value dQP for the rectangular region with the larger rectangular region number. If there is an agreement to select the transition value dQP for the rectangular area with the smaller rectangular area number, the transition value dQP for the rectangular area with the smaller rectangular area number is selected.

When the quantization control unit 33 selects a plurality of rectangular regions and calculates a transition value dQP of a plurality of quantization parameters, the entropy encoding unit 24 calculates a plurality of rectangular region numbers and a plurality of quantizations. The parameter transition value dQP is entropy-coded and transmitted to the image decoding apparatus.
In the example of FIG. 12, the rectangular area numbers M and N and the transition value dQP for the rectangular areas having the rectangular area numbers M and N are entropy-encoded and transmitted to the image decoding apparatus side.

The entropy decoding unit 52 of the image decoding apparatus performs entropy decoding on the bitstream, and among the decoded data of the bitstream, a plurality of rectangular area numbers, a plurality of quantization parameter transition values dQP, and a macroblock quantum The quantization parameter QP is output to the inverse quantization parameter generation unit 54.
In the example of FIG. 12, the rectangular region numbers M and N, the transition value dQP for the rectangular regions with the rectangular region numbers M and N, and the quantization parameter QP of the macroblock are output to the inverse quantization parameter generation unit 54.

When receiving the plurality of rectangular region numbers, the plurality of quantization parameter transition values dQP, and the macroblock quantization parameter QP from the entropy decoding unit 52, the inverse quantization parameter generation unit 54 receives the quantum block quantization parameter QP. Based on the quantization parameter QP and the transition values dQP of the plurality of quantization parameters, the quantization parameter QPt of the orthogonal transform block included in the rectangular region having the plurality of rectangular region numbers is calculated.
In the example of FIG. 12, the quantization parameter QPt of the orthogonal transform block included in the rectangular area with the rectangular area number M is calculated from the quantization parameter QP of the macroblock and the transition value dQP for the rectangular area with the rectangular area number M. Is calculated.
Further, the quantization parameter QPt of the orthogonal transform block included in the rectangular area with the rectangular area number N is calculated from the quantization parameter QP of the macroblock and the transition value dQP for the rectangular area with the rectangular area number N.

In the example of FIG. 12, the inverse quantization unit 55 performs inverse quantization on the transform coefficients of the orthogonal transform block included in the rectangular region with the rectangular region number M among the transform coefficients output from the entropy decoding unit 52. Inverse quantization is performed using the quantization parameter QPt of the orthogonal transform block in the rectangular area with the rectangular area number M output from the parameter generation unit 54.
For the transform coefficient of the orthogonal transform block included in the rectangular region with the rectangular region number N, the quantization parameter QPt of the orthogonal transform block in the rectangular region with the rectangular region number N output from the inverse quantization parameter generation unit 54 is used. Use inverse quantization.
The transform coefficient of the orthogonal transform block in the macroblock not included in any of the rectangular area numbers M and N is inverted using the macroblock quantization parameter QP output from the inverse quantization parameter generation unit 54. Quantize.

By transmitting a set of a plurality of rectangular area numbers and quantization parameter transition values dQP, the amount of bits to be transmitted increases, but with a large macroblock size such as 64 pixels × 64 lines, 16 pixels × 16 lines In this case, the number of macroblocks is equal to 16, so the overall increase is not so large.
Furthermore, the bit amount can be greatly reduced as compared with the case where the quantization parameter is transmitted in units of orthogonal transform blocks of 4 pixels × 4 lines.

  In the present invention, within the scope of the invention, any combination of the embodiments, or any modification of any component in each embodiment, or omission of any component in each embodiment is possible. .

  DESCRIPTION OF SYMBOLS 1 encoding part, 2 encoding control part (quantization parameter determination means, quantization parameter change means), 11 intra prediction part (difference image generation means), 12 motion search part (difference image generation means), 13 motion compensation prediction (Difference image generation means), 14 differencer (difference image generation means), 15 intra / inter determination section, 16 switch, 17 orthogonal transform section (quantization means), 18 quantization section (quantization means), 19 inverse Quantization unit, 20 inverse orthogonal transform unit, 21 adder, 22 deblocking filter unit, 23 frame memory, 24 entropy coding unit (entropy coding means), 25 transmission buffer, 31 image feature value computing unit, 32 rectangular area Information storage unit, 33 quantization control unit, 51 reception buffer, 52 entropy decoding unit (entropy decoding means), 53 rectangle Area information storage unit, 54 Inverse quantization parameter generation unit (quantization parameter calculation unit), 55 Inverse quantization unit (inverse quantization unit), 56 Inverse orthogonal transform unit (inverse quantization unit), 57 Intra prediction image generation unit (Predicted image generating means), 58 inter predicted image generating section (predicted image generating means), 59 switch (predicted image generating means), 60 adder (decoded image generating means), 61 deblocking filter section, 62 frame memory, 63 switch.

Claims (14)

  Prediction processing is performed for each macroblock constituting the input image to generate a prediction image, and a difference image generation means for generating a difference image between the image of the macroblock and the prediction image, and quantization for each macroblock A quantization parameter determining means for determining a parameter, a rectangular region in the macroblock whose quantization parameter is changed according to an image feature amount of the macroblock, a transition value of the quantization parameter is calculated, and the macro The quantization parameter changing means for calculating the quantization parameter of the orthogonal transformation block included in the rectangular area from the quantization parameter of the block and the transition value, and the difference image generated by the difference image generation means are orthogonally transformed. Among the transform coefficients that are the results of orthogonal transform of the difference image, the orthogonality included in the rectangular area The transform coefficient of the transform block is quantized using the quantization parameter of the orthogonal transform block calculated by the quantization parameter changing unit, and the transform coefficient of the orthogonal transform block in the macroblock not included in the rectangular area. Is determined by the quantization means for quantizing using the quantization parameter of the macroblock determined by the quantization parameter determination means, the transform coefficient quantized by the quantization means, and determined by the quantization parameter determination means When the prediction parameter is generated by the quantization parameter of the macroblock, the transition value of the quantization parameter calculated by the quantization parameter changing unit, the rectangular region specifying information for specifying the rectangular region, and the difference image generating unit Entropy-encode the prediction image generation information used for The image coding apparatus and an entropy encoding means for generating a stream.   The quantization parameter changing means stores a rectangular area number for identifying a plurality of rectangular areas having different positions or sizes in the macroblock, and a rectangular area for identifying the rectangular area in the macroblock whose quantization parameter is to be changed 2. The image encoding apparatus according to claim 1, wherein the number is output to the entropy encoding means as rectangular area specifying information.   The quantization parameter changing means includes, as information on a plurality of rectangular areas having different positions or sizes in the macroblock, the origin coordinates of the plurality of rectangular areas, the horizontal / vertical direction sizes of the plurality of rectangular areas, or the plurality of rectangular areas. The end point coordinates and the rectangular area numbers for identifying a plurality of rectangular areas are stored, and the rectangular area number for identifying the rectangular area in the macroblock whose quantization parameter is changed is stored in the entropy encoding means as the rectangular area specifying information. The image encoding apparatus according to claim 1, wherein the image encoding apparatus outputs the image.   4. The entropy encoding means, when the information about the rectangular area stored in the quantization parameter changing means is updated, entropy-encodes the information about the updated rectangular area and includes it in the bitstream. The image encoding device described.   5. The image encoding apparatus according to claim 4, wherein the entropy encoding means multiplexes information on the updated rectangular area with header information of a picture or a slice.   4. The image encoding apparatus according to claim 2, wherein the entropy encoding means entropy encodes the transition value of the quantization parameter and the rectangular area number in units of macroblocks.   4. The image encoding apparatus according to claim 2, wherein the entropy encoding means entropy encodes the transition value of the quantization parameter in units of pictures or slices.   4. The image encoding device according to claim 2, wherein the entropy encoding means entropy encodes the transition value of the quantization parameter and the rectangular area number in units of pictures or slices.   The entropy encoding means prepares a small number corresponding to the rectangular area number in the case where one or more rectangular area numbers used in a picture or slice are previously multiplexed with the header information and transmitted. 4. The image coding apparatus according to claim 2, wherein a small number corresponding to a rectangular area number for identifying a rectangular area in a macroblock whose quantization parameter is changed is entropy-coded in units.   When there are a plurality of rectangular areas in the macroblock for changing the quantization parameter, the quantization parameter changing means calculates a transition value of the quantization parameter for the plurality of rectangular areas, and the quantization parameter of the macroblock and the above-described quantization parameter 10. The image encoding device according to claim 1, wherein quantization parameters of orthogonal transform blocks included in a plurality of rectangular areas are calculated from the transition value. 11.   A bitstream is entropy decoded, and a quantized transform coefficient that is decoded data of the bitstream, a macroblock quantization parameter, a transition value of the quantization parameter, rectangular area specifying information for specifying a rectangular area, and a prediction image An entropy decoding unit that outputs prediction image generation information used at the time of generation, and an orthogonality included in the rectangular region from the quantization parameter and the transition value of the macroblock output from the entropy decoding unit Of the transform parameters output from the entropy decoding means, the transform parameter of the orthogonal transform block included in the rectangular area indicated by the rectangular area specifying information, among the transform coefficients output from the entropy decoding means. Is the orthogonal transform block calculated by the quantization parameter calculating means. For the transform coefficient of the orthogonal transform block in the macroblock that is not included in the rectangular area, the inverse quantization is performed using the quantization parameter of the macroblock, and the inverse quantization is performed. Inverse quantization means for inverse orthogonal transform of the quantized transform coefficient, prediction image generation means for generating a prediction image using the prediction image generation information output from the entropy decoding means, and the prediction image generation means An image decoding apparatus comprising: a decoded image generating unit that generates a decoded image by adding the prediction image generated by the above and a difference image that is a result of the inverse orthogonal transform of the inverse quantization unit.   When the quantization parameter transition values for the plurality of rectangular regions are output from the entropy decoding unit, the quantization parameter calculation unit is included in the plurality of rectangular regions from the quantization parameter of the macroblock and the transition value. 12. The image decoding apparatus according to claim 11, wherein a quantization parameter of the orthogonal transform block is calculated.   A difference image generation processing step, wherein the difference image generation means performs a prediction process for each macroblock constituting the input image to generate a prediction image, and generates a difference image between the macroblock image and the prediction image; A quantization parameter determination processing step in which the quantization parameter determination means determines a quantization parameter for each macroblock; and the macroblock in which the quantization parameter change means changes the quantization parameter according to the image feature amount of the macroblock A rectangular area within the area, and the transition value of the quantization parameter is calculated, and the quantization parameter of the orthogonal transform block included in the rectangular area is calculated from the quantization parameter of the macroblock and the transition value. A quantization parameter changing process step and a quantization means for performing the difference image generating process step. The difference image generated in step 2 is orthogonally transformed, and among the transform coefficients that are the result of the orthogonal transform of the difference image, the transform parameter of the orthogonal transform block included in the rectangular area is the quantization parameter changing process. Quantization is performed using the quantization parameter of the orthogonal transform block calculated in step, and the transform coefficient of the orthogonal transform block in the macroblock not included in the rectangular region is determined in the quantization parameter determination processing step. A quantization processing step for quantizing using the quantization parameter of the macroblock, and an entropy encoding means, the transform coefficient quantized in the quantization processing step, the macro determined in the quantization parameter determination processing step The quantization parameter of the block, the quantum calculated in the above quantization parameter change processing step Entropy for generating a bitstream by entropy encoding the parameter transition value, the rectangular area specifying information for specifying the rectangular area, and the prediction image generation information used when the prediction image is generated in the difference image generation processing step An image encoding method comprising: an encoding process step.   Entropy decoding means entropy decodes the bitstream, and the quantized transform coefficient that is the decoded data of the bitstream, the quantization parameter of the macroblock, the transition value of the quantization parameter, and the rectangular area specification that specifies the rectangular area An entropy decoding processing step for outputting prediction image generation information used when the information and the prediction image are generated, and a quantization parameter calculating means, the quantization parameter of the macroblock output in the entropy decoding processing step, The quantization parameter calculation processing step for calculating the quantization parameter of the orthogonal transform block included in the rectangular area from the transition value, and the inverse quantization means, among the transform coefficients output in the entropy decoding processing step Included in the rectangular area indicated by the rectangular area specifying information The transform coefficient of the orthogonal transform block is inversely quantized using the quantization parameter of the orthogonal transform block calculated in the quantization parameter calculation processing step, and the orthogonal transform in the macroblock not included in the rectangular area is performed. For the transform coefficient of the block, the inverse quantization process step of performing inverse quantization using the quantization parameter of the macroblock and performing the inverse orthogonal transform on the transform coefficient after the inverse quantization, and the predicted image generation unit include the entropy decoding process. A prediction image generation processing step for generating a prediction image using the prediction image generation information output in the step, and a decoded image generation means, the prediction image generated in the prediction image generation processing step and the inverse quantization processing A decoded image generation processing step of generating a decoded image by adding a difference image which is a result of inverse orthogonal transformation in step; Image decoding method comprising.

Images