Video encoding method and apparatus, and video decoding apparatus

Technical Field

The present invention relates to a technique field of video encoding and video decoding corresponding thereto, and more particularly, to a technique for improving encoding efficiency or visual image quality by allowing the operation unit of a quantization calculation to be more finely controlled.

Background Art

In video encoding methods, an enormous amount of information of an original signal is compressed by omitting redundancies in temporal and spatial directions. Specifically, a technique of motion compensation for taking a difference between preceding and succeeding frames by using a motion vector, and a technique of an orthogonal transform for transforming a plane where pixels are distributed on a screen, namely, in horizontal and vertical directions into frequency components, and a technique of rounding an orthogonal transform coefficient to a representative value with quantization are respectively adopted for temporal and spatial directions. Moreover, variable-length encoding (entropy encoding) is used as a technique of arithmetic information compression.

With conventional video encoding methods adopting, especially, motion vector compensation, encoding is fundamentally performed in processing units of MBs (Macroblocks) of 16x16 pixels. However, encoding in units of blocks of 8x8 pixels is enabled with the encoding methods such as H. 263 and MPEG-4. With the latest video encoding of H. 264/AVC (Advanced Video Coding), the number of divisions further increases to 16x16, 16x8, 8×16 and 8x8, and blocks of 8x8 pixels are further divided into sub-blocks of 8x8, 8x4, 4x8 and 4x4.

Conventionally, not only in information compression using motion compensation in a temporal direction but also, for example, in an orthogonal transform, DCT (Discrete Cosine Transform) only in units of 8x8 pixels is implemented. However, with H.264/AVC, switching can be made between the processing units of 4x4 and 8x8 for each macroblock although this switching is limited to a profile higher than a high profile.

<figref idref="f0001">Fig. 1</figref> is a block diagram illustrating a configuration example of functional blocks of a video encoding apparatus (sometimes referred to as an encoder) for implementing the above described video encoding method.

As illustrated in <figref idref="f0001">Fig. 1</figref>, the functional blocks of the video encoding apparatus includes a frame memory 11, an original image macroblock buffer 12, a reference block buffer 13, a motion vector searching unit 21, a prediction determining unit 22, a subtractor 31, a first switch 32, an orthogonal transform (DCT) unit 33, a quantization (Q) unit 34, a variable-length encoding (ENT) unit 51, an inverse quantization (IQ) unit 44, an inverse orthogonal transform (IDCT) unit 43, a second switch 42, and an adder 41.

The frame memory 11 stores past and future images in order to make motion estimation.

The original image macroblock buffer 12 stores macroblocks of an original frame to be encoded of each frame stored in the frame memory 11, whereas the reference block buffer 13 stores reference blocks for the macroblocks of the original frame.

The motion vector searching unit 21 searches for a motion vector by using the macroblocks of the original frame and their reference blocks.

The prediction determining unit 22 evaluates motion estimation for all of division shapes of a macroblock illustrated in <figref idref="f0003">Fig. 3A</figref> to decide a division shape, and determines whether encoding is to be performed either with inter-frame prediction or with intra-frame prediction.

The subtractor 31 calculates a difference between a macroblock and a predicted macroblock.

Switching is made between the first switch 32 and the second switch 42 depending on whether encoding is performed either with inter-frame prediction or with intra-frame prediction.

The orthogonal transform (DCT) unit 33 obtains an orthogonal transform coefficient by performing an orthogonal transform (such as DCT) for image data the information of which is compressed in a temporal direction, and compresses the information in a spatial direction.

The quantization (Q) unit 34 quantizes the orthogonal transform coefficient, and the variable-length encoding (ENT) unit 51 outputs an encoding output by further performing arithmetic compression for the information.

The inverse quantization (IQ) unit 44 obtains an original orthogonal transform coefficient by performing inverse quantization for the quantized orthogonal transform coefficient. The inverse orthogonal transform (IDCT) unit 43 restores data before being orthogonal-transformed from the orthogonal transform coefficient by performing an inverse orthogonal transform.

The adder 41 restores an original image by adding predicted image data to difference data that is the output of the inverse orthogonal transform (IDCT) unit 43 if encoding is performed with inter-frame prediction.

<figref idref="f0002">Fig. 2</figref> is a flowchart of a macroblock process executed in the conventional example.

The flow of <figref idref="f0002">Fig. 2</figref> illustrates the process according to the order where items of information that are generated by processing a macroblock with the encoder are set and transmitted as encoding information. This order conforms to that laid down as decoding syntax elements in H.264. Table 1 to be provided later is a syntax table of a macroblock layer and its lower-level layers in H.264.

Initially, macroblock type information is set as the initial item of the encoding information in step S21. This information includes information indicating whether encoding is performed either with inter-frame prediction or with intra-frame prediction, and information about the division shape of a macroblock. As the next item of the encoding information, motion vector information is set in step S22. Since the division shape of a macroblock varies depending on the type of the macroblock, the motion vector information is set by the number of divisions as indicated by step S23.

Next, a quantization parameter value is set in the encoding information in step S24. This value is set for each macroblock.

Then, a flag indicating whether an orthogonal transform is performed in units of either 8x8 or 4x4 is set as orthogonal transform information in the encoding information in step S25.

Lastly, in step S26, a coefficient after being orthogonal-transformed in units of 8x8 or 4x4 is obtained, and transform coefficient information obtained by quantizing the coefficient with the quantization parameter set in step S24 is generated and transmitted in units of sub-blocks. This process is repeated by the number of divisions as indicated by step S27. At this time, a flag cbp (coded block pattern) indicating validity/invalidity for each sub-block is set after the motion vector information and before the quantization parameter information within the encoding information. Only the valid coefficient information of a sub-block, which is indicated by the flag, is transmitted.

<figref idref="f0003 f0004 f0005">Figs. 3A to 3C</figref> are explanatory views of conventional macroblock divisions in video encoding. <figref idref="f0003">Fig. 3A</figref> is an explanatory view of dividing a macroblock in motion estimation. As illustrated in this figure, the macroblock can be divided into 16×16, 16x8, 8×16 and 8x8, and the divided portions of 8x8 can be further divided into 8x4, 4x8 and 4x4.

<figref idref="f0004">Fig. 3B</figref> is an explanatory view of dividing a macroblock in an orthogonal transform. As illustrated in this figure, the macroblock can be divided into blocks of 8x8 and 4x4.

<figref idref="f0005">Fig. 3C</figref> illustrates the case of quantization. As illustrated in this figure, quantization is performed in units of 16x16.

The encoding process is further described next with reference to <figref idref="f0001">Figs. 1</figref> and <figref idref="f0003 f0004 f0005">3A to 3C</figref>.

Motion estimations in all of divisions of 16×16, 16x8, 8×16 and 8x8 illustrated in <figref idref="f0003">Fig. 3A</figref> are evaluated by the prediction determining unit 22 illustrated in <figref idref="f0001">Fig. 1</figref>, and a prediction mode (macroblock type) is decided by determining the most efficient way of divisions and whichever of inter-frame prediction and intra-frame prediction is to be selected.

Next, the size (orthogonal transform information) of a block to be orthogonal-transformed (DCT) is decided depending on whichever units of 8x8 and 4x4 illustrated in <figref idref="f0004">Fig. 3B</figref> as the units of the orthogonal transform further reduces the number of prediction errors, and the orthogonal transform (DCT) unit 33 performs the orthogonal transform process.

Then, the quantization (Q) unit 34 rounds a transformed coefficient to a representative value by using a quantization parameter value decided from the viewpoint of distributing the amount of information, and transmits a remaining valid coefficient of non-zero. At this time, the flag cbp indicating whether or not a valid coefficient exists among quantization coefficient values in units of sub-blocks is calculated. Then, the flag cbp, and only the quantization coefficient information of a sub-block having a valid coefficient indicated by the flag cbp are transmitted as encoding information.

Patent Documents 1 to 4 related to video encoding technology are introduced next.

Patent Document 1 particularly refers to the prediction encoding technique used within a screen. Patent Document 2 particularly refers to the division of a macroblock with an arbitrary line segment. Patent Document 3 particularly refers to the technique of quick re-encoding when an encoding method is converted. Patent Document 4 refers to the technique of performing an orthogonal transform by again dividing into small blocks.

However, none of the documents refer to quantization performed by dividing a macroblock. <ul id="ul0001" list-style="none" compact="compact"><li>Patent Document 1: Japanese Laid-open Patent Publication No. <patcit id="pcit0001" dnum="JP2005318468A"><text>2005-318468</text></patcit></li><li>Patent Document 2: Japanese Laid-open Patent Publication No. <patcit id="pcit0002" dnum="JP2005277968A"><text>2005-277968</text></patcit></li><li>Patent Document 3: Japanese Laid-open Patent Publication No. <patcit id="pcit0003" dnum="JP2005236584A"><text>2005-236584</text></patcit></li><li>Patent Document 4: Japanese Laid-open Patent Publication No. <patcit id="pcit0004" dnum="JPH879753B"><text>H8-79753</text></patcit></li></ul>

Disclosure of Invention

As described above, encoding using motion compensation or an orthogonal transform is performed by dividing a macroblock into small blocks. However, quantization is still performed with one parameter for each macroblock under the present circumstances. Actually, for quantization control, there is a subjective assessment evaluation improvement technique using, for example, fine and coarse quantization parameters respectively for low and high activities in adaptive quantization. However, the quantization process is executed with one parameter for each macroblock. Therefore, even if both fine and coarse images are included within a macroblock, fine and coarse control according to the images cannot be performed.

Namely, motion compensation and orthogonal transform processes can be performed in units of small blocks of a variable size, whereas quantization cannot be finely controlled.

An object of the present invention is therefore to allow a fine quantization control to be performed according to the property of an image within a macroblock.

The present invention realizes finer quantization control by allowing a quantization parameter value to be changed in units of sub-blocks equal to or smaller than a macroblock in a similar manner as in motion compensation and orthogonal transform processes. Namely, even if a plurality of images having different properties coexist in a macroblock, a finer-tuned quantization control is performed, for example, by selecting fine and coarse quantization parameters respectively for corresponding sub-blocks.

According to the present invention, a finer encoding control than conventional technology can be realized by performing quantization in units of sub-blocks equal to or smaller than a macroblock.

Brief Description of Drawings

<ul id="ul0002" list-style="none" compact="compact"><li><figref idref="f0001">Fig. 1</figref> is a block diagram illustrating a configuration example of functional blocks of a conventional video encoding apparatus;</li><li><figref idref="f0002">Fig. 2</figref> is a flowchart of a macroblock process executed in the conventional example;</li><li><figref idref="f0003">Fig. 3A</figref> is an explanatory view of dividing a macroblock in conventional motion estimation in video encoding;</li><li><figref idref="f0004">Fig. 3B</figref> is an explanatory view of dividing a macroblock in a conventional orthogonal transform in video encoding;</li><li><figref idref="f0005">Fig. 3C</figref> illustrates that a macroblock is not divided in conventional quantization in video encoding;</li><li><figref idref="f0006">Fig. 4</figref> is a block diagram illustrating a configuration example of functional blocks of a video encoding apparatus according a first embodiment of the present invention;</li><li><figref idref="f0007">Fig. 5</figref> is a block diagram illustrating a configuration example of functional blocks of a quantization division determining unit illustrated in <figref idref="f0006">Fig. 4</figref>;</li><li><figref idref="f0008">Fig. 6</figref> is a flowchart of a macroblock process according to the first embodiment of the present invention;</li><li><figref idref="f0009">Fig. 7</figref> is a flowchart of a macroblock process according to a second embodiment of the present invention;</li><li><figref idref="f0010">Fig. 8A</figref> is an explanatory view of a third embodiment according to the present invention;</li><li><figref idref="f0011">Fig. 8B</figref> is an explanatory view of the third embodiment according to the present invention;</li><li><figref idref="f0012">Fig. 8C</figref> is an explanatory view of the third embodiment according to the present invention; and</li><li><figref idref="f0013">Fig. 9</figref> is a block diagram illustrating a configuration example of functional blocks of a video decoding apparatus corresponding to the video encoding apparatus according to the present invention.</li></ul>

Best Mode of Carrying Out the Invention

The present invention also allows a quantization parameter value to be changed in units of sub-blocks equal to or smaller than a macroblock in a similar manner as in motion compensation and orthogonal transform processes. To implement this, quantization parameter switching according to the present invention is realized by more finely setting the control units of the quantization (Q) unit 34 and the inverse quantization (IQ) unit 44 among the functional blocks illustrated in <figref idref="f0001">Fig. 1</figref>.

<figref idref="f0006">Fig. 4</figref> is a block diagram illustrating a configuration example of the functional blocks of a video encoding apparatus according to the first embodiment of the present invention. This figure illustrates portions related to the present invention.

A differences from the conventional example illustrated in <figref idref="f0003 f0004 f0005">Fig. 3</figref> is an addition of a quantization division determining unit 35 for determining in which units of sub-blocks a quantization parameter is set. Also the quantization (Q) unit 34a and the inverse quantization (IQ) unit 44a are modified to switch the units of sub-blocks to be processed according to quantization division information that is the output of the quantization division determining unit 35.

To the quantization division determining unit 35 illustrated in <figref idref="f0006">Fig. 4</figref>, pixel information in units of macroblocks is input via the first switch 32. For intra-frame prediction, an image itself of a macroblock is input from the original image macroblock buffer 12. In contrast, for inter-frame prediction, a difference image is input via the subtractor 31.

The quantization division determining unit 35 executes mode processes for the pixels of the input macroblock in predetermined units of divisions in a similar manner as in, for example, prediction evaluations that the prediction determining unit 22 makes in all the motion vector division modes, makes a prediction evaluation expressing which of the division modes is good, and outputs the result of the evaluation to the quantization (Q) unit 34a and the inverse quantization (IQ) unit 44a.

<figref idref="f0007">Fig. 5</figref> is a block diagram illustrating a configuration example of functional blocks of the quantization division determining unit 35 illustrated in <figref idref="f0006">Fig. 4</figref>.

Input pixel information in units of macroblocks is stored in the macroblock memory 110, and read into the buffers of, for example, one small block 1 (16×16) 121, two small blocks 2 (16×8) 122, two small blocks 3 (8×16) 123, and four small blocks 4 (8x8) 124 according to the division modes, and the activities of the sub-blocks are respectively calculated by their corresponding activity calculating units 131 to 134, and evaluation values of the division modes are output.

Calculation results obtained by the activity calculating units 131 to 134 are input to a quantization division deciding unit 150 via a division mode switch 140.

The quantization division deciding unit 150 decides a division mode based on the calculation results, and outputs quantization division information corresponding to the decided division mode.

The present invention is intended to perform quantization for sub-blocks, the activities of which are different, by respectively using suitable quantization parameters. If the number of divisions is simply increased, the amount of processing and the amount of encoding increase. Therefore, it is necessary to consider a balance between adaptive quantization and an increase in the amount of encoding, etc.

From an adaptive quantization viewpoint, it is desirable to divide portions having a difference in an activity as sub-blocks. An activity is associated with the degree of fineness of an image pattern. Therefore, the activity of a region of large variance of pixels is high, whereas that of a region of small variance of pixels is low. Namely, if one image region is divided into sub-blocks and if they include a sub-block of large variance of pixels and a sub-block of small variance of pixels, such a division is preferable. For example, for an image where the upper half and the lower half are respectively the blue sky and the sea with a complex wave pattern like the rough sea on the sunny day, the division of 16x8 illustrated in <figref idref="f0003">Fig. 3A</figref> is preferable.

Accordingly, if the variance of pixels within a sub-block is defined as the activity of the sub-block, a portion having a difference in the activity is divided as a sub-block if the division shape has a large variance of the activity of the sub-block. As a result, the evaluation of the division mode is considered to increase.

In the meantime, to achieve the balance with the amount of encoding increased by the division, it is considered to set a condition that a difference between the variance of the activity of a sub-block resultant from the division and that of the activity of a sub-block before being divided is larger than a predetermined threshold value.

In the above described image example of the rough sea on the sunny day, whether or not to further divide into sub-blocks of 8×8 depends on the balance between the advantage achieved by an increase in the variance of the activity of a sub-block and an increase in the amount of encoding as a result of the division.

Accordingly, the number of divisions is increased within a range of a defined division shape by recognizing a macroblock as the zeroth division shape, and an evaluation is made. Various methods can be possibly adopted as an evaluation method based on the above concept.

Assuming that the variance of the activity of a sub-block, which is resultant from an ith (i≥0) division, is B(i), the ith division shape can be adopted if B (i+1) -B (i) &lt;C is satisfied for the threshold value C. If there are plurality of (i+1)th division shapes, it may be conditioned that the above inequality is satisfied for all the division shapes. Moreover, the threshold value C may be set as C(i) and varied with the value of i.

Additionally, it may be also conditioned that B (j) -B (i) &lt;C is satisfied for all of j that satisfy j&gt;i, not limited to the comparison with the division in the next step.

Also the following method may be adopted.

Here, define the activity A(s) of a sub-block s as <maths id="math0001" num=""><math display="block"><mrow><mi mathvariant="normal">A</mi><mfenced><mi mathvariant="normal">s</mi></mfenced><mo>=</mo><mo>∑</mo><msup><mfenced separators=""><mi mathvariant="normal">each pixel value</mi><mo>−</mo><mi mathvariant="normal">average value of pixels</mi></mfenced><mn mathvariant="normal">2</mn></msup><mo>−</mo><msup><mfenced separators=""><mo>∑</mo><mfenced separators=""><mi mathvariant="normal">each pixel value</mi><mo>−</mo><mi mathvariant="normal">average value of pixels</mi></mfenced></mfenced><mn mathvariant="normal">2</mn></msup></mrow></math><img file="EP3107293A1_D0001.tif" /></maths> The sum is assumed to be taken for the pixels within the sub-block.

Then, the variance BD (k) of the activity of the sub-block in a division shape of D(k) within a macroblock is defined as <maths id="math0002" num=""><math display="block"><mrow><mi mathvariant="normal">BD</mi><mfenced><mi mathvariant="normal">k</mi></mfenced><mo>=</mo><mo>∑</mo><msup><mfenced separators=""><mi mathvariant="normal">A</mi><mfenced><mi mathvariant="normal">s</mi></mfenced><mo>−</mo><mi mathvariant="normal">average value of A</mi><mfenced><mi mathvariant="normal">s</mi></mfenced></mfenced><mn mathvariant="normal">2</mn></msup><mo>−</mo><msup><mfenced separators=""><mo>∑</mo><mfenced separators=""><mi mathvariant="normal">A</mi><mfenced><mi mathvariant="normal">s</mi></mfenced><mo>−</mo><mi mathvariant="normal">average value of A</mi><mfenced><mi mathvariant="normal">s</mi></mfenced></mfenced></mfenced><mn mathvariant="normal">2</mn></msup></mrow></math><img file="EP3107293A1_D0002.tif" /></maths> The sum is assumed to be taken for the sub-block in the division shape of D (k) .

Furthermore, an evaluation expression H(k) of the division shape D (k) is defined as follows on the condition that α is a positive parameter. <maths id="math0003" num=""><math display="block"><mrow><mi mathvariant="normal">H</mi><mfenced><mi mathvariant="normal">k</mi></mfenced><mo>=</mo><mo>−</mo><mo>∑</mo><mi mathvariant="normal">A</mi><mfenced><mi mathvariant="normal">s</mi></mfenced><mo>−</mo><mi mathvariant="normal">α</mi><mo>*</mo><mi mathvariant="normal">BD</mi><mfenced><mi mathvariant="normal">k</mi></mfenced></mrow></math><img file="EP3107293A1_D0003.tif" /></maths> The sum is assumed to be taken for the sub-block in the division shape of D (k) .

The above described evaluation expression H(k) is intended to finally select the lowest one among the activities (the values of the evaluation expression H(k)) of the division modes as a division mode of quantization by further decreasing the value of the evaluation expression H(k) if there is a big difference (namely, a sub-block having a fine drawing pattern and a sub-block having a coarse drawing pattern exist among blocks) among the activity values of divided sub-blocks.

As the number of divisions increases, the first term and the second term of H (k) increases and decreases, respectively. The value of α is a variable for adjusting the weights of these two terms, and is assumed to be a suitably adjustable value (the activity of pixels is the variance of 256 pixels of 16×16, and the activity of a sub-block is the variances of two to four sub-blocks in the example of the division shape illustrated in <figref idref="f0006">Fig. 4</figref>. The value of α is adjusted in consideration of these conditions, specific pixel values, etc.)

As described above, a division mode is selected in consideration of the advantages and the disadvantages provided by the divisions.

Quantization division information generated by the quantization division deciding unit 150 is passed to the quantization (Q) unit 34a as described above, and used to quantize a transform coefficient after being orthogonal-transformed in specified division units.

<figref idref="f0008">Fig. 6</figref> is a flowchart of a macroblock process according to the first embodiment of the present invention.

Compared with the flow illustrated in <figref idref="f0002">Fig. 2</figref>, a process for obtaining motion vector information and for setting the obtained motion vector information in encoding information in steps S61 to S63 of <figref idref="f0008">Fig. 6</figref> is similar to that executed in steps S21 to S23 illustrated in <figref idref="f0002">Fig. 2</figref>. As the initial item of encoding information, macroblock type information as motion compensation division information is set in step S61. This information includes the information indicating whether encoding is performed either with inter-frame prediction or with intra-frame prediction, and information about the division shape of a macroblock as described above. Next, motion vector information is set in step S62. Since the division shape of a macroblock varies depending on the type of the macroblock, the motion vector information is set by the number of divisions as indicated by step S63.

Next, in step S64, quantization division information generated by the quantization division determining unit 35 is set.

A process in the subsequent steps S65 and S66 is intended to obtain quantization parameters by the number of divisions in quantization, which is indicated by the quantization division information, and to set the obtained parameters in the encoding information.

In the configuration of the quantization division determining unit 35 illustrated in <figref idref="f0007">Fig. 5</figref>, quantization division information of any of one small block of 16x16 that is not substantially divided into small blocks, two small blocks of 16x8, two small blocks of 8×16, and four small blocks of 8x8 is selected. These division shapes are selected so that a selected division shape becomes that obtained by being further divided as the division shape in the orthogonal transform illustrated in <figref idref="f0004">Fig. 3B</figref>.

In the first embodiment, quantization parameters can be switched in units of sub-blocks according to the division shape of a macroblock, which is selected by the quantization division determining unit 35. Unlike the flow illustrated in <figref idref="f0002">Fig. 2</figref>, quantization parameters are inserted in units of sub-blocks.

Next, the flag indicating whether the orthogonal transform is performed in units of either 8x8 or 4x4 is set as orthogonal transform information in the encoding information in step S67.

Lastly, in step S68, the coefficient after being orthogonal-transformed in units of 8x8 or 4x4 is obtained, and transform coefficient information obtained by quantizing the coefficient with the quantization parameter set in step S65 is generated in units of sub-blocks, and transmitted. This process is repeated by the number of divisions as indicated by step S69. At this time, the flag cbp (coded block pattern) indicating validity/invalidity for each sub-block is set before the quantization parameter information and after the motion vector information within the encoding information. In a similar manner as in the conventional example illustrated in <figref idref="f0002">Fig. 2</figref>, only the valid coefficient information of a sub-block, which is indicated by the flag, is transmitted.

Additionally, since the coefficient set as the orthogonal transform information is that after being quantized, the divided quantization parameter information and the orthogonal transform information are inserted in the encoding information in this order in a similar manner as in the conventional example illustrated in <figref idref="f0002">Fig. 2</figref>.

The second embodiment according to the present invention is described next. The second embodiment eliminates the need for encoding quantization division information by making the division shape of a macroblock in the quantization process identical to a division shape in another encoding process. Accordingly, the amount of encoding can be prevented from increasing by the amount of encoding of the quantization division information.

<figref idref="f0009">Fig. 7</figref> is a flowchart of a macroblock process in the case where the division shape of quantization is made identical to that of the orthogonal transform. Compared with the flow of the macroblock process according to the first embodiment illustrated in <figref idref="f0008">Fig. 6</figref>, the process for setting the quantization division information in step S65 is eliminated.

Accordingly, in the second embodiment, the quantization division determining unit 35 illustrated in <figref idref="f0006">Fig. 4</figref> is not necessary. Alternatively, information about the division shape of the orthogonal transform is output from the prediction determining unit to the quantization (Q) unit 34a and the inverse quantization (IQ) unit 44a.

For example, if the size of a sub-block obtained by dividing a macroblock in the orthogonal transform is 8x8, also the size of a sub-block obtained by dividing a macroblock in the quantization process is 8x8.

A process for obtaining motion vector information and for setting the obtained information in the encoding information in steps S71 to S73 is similar to the process in steps S21 to S23 of <figref idref="f0002">Fig. 2</figref>, and the process in steps S61 to S63 of <figref idref="f0008">Fig. 6</figref>.

Next, the flag indicating whether the orthogonal transform is performed in units of either 8x8 or 4x4 is set as orthogonal transform information in the encoding information in step S75.

Then, quantization parameter information is set for each sub-block according to the division shape set in step S75 in step S76, and the coefficient information of the orthogonal transform, which is quantized with the set quantization parameter information, is obtained and set in the video encoding information in step S77.

As indicated by steps S78, steps S76 and S77 are repeated by the number of divisions of sub-blocks in the orthogonal transform.

The process illustrated in <figref idref="f0009">Fig. 7</figref> is described by assuming that the division shape in the quantization is made identical to that in the orthogonal transform. However, a division shape in motion compensation and that in the quantization may be made identical as an alternative to the division shape in the orthogonal transform. In this case, the quantization parameter information is set for each sub-block according to the division shape set in step S71 in step S76. In step S77, the coefficient information in the orthogonal transform, which is quantized with the set quantization parameter information, is obtained and set in the video encoding information. However, the division shape in the motion compensation, namely, the division shape in the quantization must be divisions into sub-blocks of a size equal to or larger than that of a sub-block in the orthogonal transform in this case. This is because if the division shape in the quantization is fine, also the orthogonal transform process must be executed according to this division shape.

A third embodiment is described next. The third embodiment relates to the encoding of a quantization parameter, and can be implemented simultaneously with the above described first or second embodiment.

Conventional quantization is performed in units of macroblocks, and an encoding method of a quantization parameter is intended to encode a difference between the quantization parameter values of the current macroblock and a macroblock positioned at the left of the current macroblock.

The present invention allows quantization to be performed in units of sub-blocks. Therefore, in the third embodiment, a difference between quantization parameter values is encoded with reference to a plurality of neighboring sub-blocks, and the amount of encoding of quantization parameter information is further reduced.

<figref idref="f0010 f0011 f0012">Figs. 8A to 8C</figref> are explanatory views of the third embodiment. These figures illustrate a sub-block currently being encoded and neighboring sub-blocks to be referenced. In <figref idref="f0010 f0011 f0012">Figs. 8A to 8C</figref>, a macroblock or a sub-block D, which is depicted with solid lines, is the block currently being encoded, and macroblocks or sub-blocks B, C and D, which are depicted with dotted lines, are the neighboring blocks to be referenced.

For example, for a motion vector, the intermediate value of motion vectors of the left macroblock (A), the upper macroblock (B) and the upper right macroblock (C), which are illustrated in <figref idref="f0010">Fig. 8A</figref>, is recognized as a predicted vector of the current macroblock, and a difference from the predicted vector is encoded. Similarly, even when quantization is performed in units of sub-blocks, the amount of information of the quantization parameter value of the macroblock D can be reduced by calculating the intermediate value of the quantization parameter values of the left macroblock (A), the upper macroblock (B) and the upper right macroblock (C), and by encoding a difference from the intermediate value.

<figref idref="f0011">Fig. 8B</figref> illustrates that the current macroblock D is encoded by referencing the quantization parameter values of a sub-block A of the left macroblock, a sub-block B of the upper macroblock, and a sub-block C of the upper right macroblock. <figref idref="f0012">Fig. 8C</figref> illustrates that the current sub-block D of the current macroblock is encoded by referencing the quantization parameter values of the left macroblock A, an upper sub-block B of the current macroblock and an upper right sub-block C of the current macroblock.

As illustrated in <figref idref="f0011">Figs. 8B</figref> and <figref idref="f0012">8C</figref>, a plurality of sub-blocks sometimes contact at the positions of the left macroblock, the upper macroblock, and the upper right macroblock. In such a case, it is predefined, for example, that a further upper sub-block, a further left sub-block, and a further lower left sub-block are selected respectively as the left sub-block A, the upper sub-block B and the upper right sub-block C in a similar manner as in the prediction of a motion vector.

As described above in detail, according to the present invention, favorable video encoding can be realized. Moreover, the present invention was devised to minimize the amount of encoding, which is required for a quantization information transmission and can be possibly increased by changing quantization parameters in finer units.

Video decoding corresponding to the video encoding method according to the present invention is described last.

<figref idref="f0013">Fig. 9</figref> is a block diagram illustrating a configuration example of functional blocks of a video decoding apparatus corresponding to the video encoding apparatus according to the present invention. This figure illustrates only portions related to the present invention. A process of the video decoding apparatus is almost similar to the image restoration process in and after the inverse quantization in the video encoding apparatus.

For example, an encoding output from the video encoding apparatus illustrated in <figref idref="f0006">Fig. 4</figref> is input to a variable-length decoder (ENT) 151 via a storage medium or a communication medium. The variable-length decoder (ENT) 151 corresponds to the variable-length encoding (ENT) unit 51 illustrated in <figref idref="f0001">Figs. 1</figref> and <figref idref="f0006">4</figref>. This decoder performs an inverse transform for arithmetic information compression performed by the variable-length encoding (ENT) unit 51, and decodes data before the arithmetic information compression, namely, data after being quantized, etc.

A quantization division information obtaining unit 135 obtains quantization division information set by the quantization division determining unit 35 illustrated in <figref idref="f0006">Fig. 4</figref> from the data after being decoded, and controls the inverse quantization process of the inverse quantization (IQ) unit 144. As stated earlier, the quantization division information indicates the division shape of a macroblock, in which quantization parameters can be switched in units of sub-blocks. Therefore, the inverse quantization (IQ) unit 144 performs inverse quantization for an orthogonal transform coefficient that is quantized in units of sub-blocks according to the instructed division shape in a macroblock to be processed, and obtains the original orthogonal transform coefficient.

The inverse orthogonal transform (IDCT) unit 143 restores data before being orthogonal-transformed from the orthogonal transform coefficient with an inverse orthogonal transform.

In the meantime, decoded data of the variable-length decoder (ENT) 151 is provided to a prediction mode obtaining unit 122.

The prediction mode obtaining unit 122 obtains, from the decoded data, information indicating whether the prediction mode selected by the prediction determining unit 22 illustrated in <figref idref="f0006">Fig. 4</figref> is either inter-frame prediction or intra-frame prediction, and switches a switch 142 to an intra side or an inter side according to the information.

In the case of the intra-frame prediction mode, the output of the inverse orthogonal transform (IDCT) unit 143 is output as a restored image, and stored in the frame memory 111 in order to be used to restore an image in the inter-frame prediction mode.

In the case of the inter-frame prediction mode, the output of the inverse orthogonal transform (IDCT) unit 143 is difference data. Base image data of the difference data is read from the frame memory 111, and motion vector information obtained by the motion vector obtaining unit 121 is added to the image data, which is then provided from the prediction mode obtaining unit 122 to an adder 141. The adder 141 generates a restored image by adding the difference data that is the output of the inverse orthogonal transform (IDCT) unit 143 to the image data from the prediction mode obtaining unit 122.

The video decoder illustrated in <figref idref="f0013">Fig. 9</figref> is described by being assumed to correspond to the video encoding apparatus according to the first embodiment. However, it is evident that this decoder can also correspond to the video encoding apparatus according to the second embodiment by obtaining macroblock type information or orthogonal transform information in the quantization division information obtaining unit 135. Accordingly, it is also evident that the configuration example of the functional blocks of the video decoder illustrated in <figref idref="f0013">Fig. 9</figref> is not limited to that according to the first embodiment. <tables id="tabl0001" num="0001"><img file="EP3107293A1_D0004.tif" /></tables><tables id="tabl0002" num="0002"><img file="EP3107293A1_D0005.tif" /></tables><tables id="tabl0003" num="0003"><img file="EP3107293A1_D0006.tif" /></tables><tables id="tabl0004" num="0004"><img file="EP3107293A1_D0007.tif" /></tables><tables id="tabl0005" num="0005"><img file="EP3107293A1_D0008.tif" /></tables><tables id="tabl0006" num="0006"><img file="EP3107293A1_D0009.tif" /></tables><tables id="tabl0007" num="0007"><img file="EP3107293A1_D0010.tif" /></tables><tables id="tabl0008" num="0008"><img file="EP3107293A1_D0011.tif" /></tables><tables id="tabl0009" num="0009"><img file="EP3107293A1_D0012.tif" /></tables>

Further embodiments are as follows: <ul id="ul0003" list-style="none"><li>E1. A video encoding method for dividing each frame of a video into macroblocks of a predetermined size, and for encoding each of the macroblocks by combining temporal compression using difference data and motion compensation, an orthogonal transform for transforming spatial information into frequency information, and quantization of an orthogonal transform coefficient, comprising: <ul id="ul0004" list-style="none" compact="compact"><li>further dividing the macroblocks into small blocks;</li><li>deciding a quantization parameter for each of the small blocks; and</li><li>quantizing the orthogonal transform coefficient with the quantization parameter for each of the small blocks.</li></ul></li><li>E2. The video encoding method according to E1, wherein the orthogonal transform is performed in units of small blocks into which the macroblocks are further divided, and a division shape of the small blocks for the quantization parameter is made identical to a division shape of small blocks in the orthogonal transform.</li><li>E3. The video encoding method according to E1, wherein the motion compensation is performed in units of small blocks into which the macroblocks are further divided, and a division shape of the small blocks for the quantization parameter is made identical to a division shape of small blocks for the motion compensation.</li><li>E4. The video encoding method according to any one of E1 to E3, wherein a difference of a quantization parameter from a more suitable quantization parameter among quantization parameters of encoded neighboring macroblocks or small blocks is encoded.</li><li>E5. The video encoding method according to E4, wherein the more suitable quantization parameter is an intermediate value of quantization parameters of the encoded neighboring macroblocks or small blocks.</li><li>E6. A video encoding apparatus, which has an orthogonal transform unit for compressing information of image data in a spatial direction and a quantization unit for quantizing an orthogonal transform coefficient of the image data that is orthogonal-transformed by the orthogonal transform unit, for generating video encoding information by dividing each frame of a video into macroblocks of a predetermined size and by encoding the macroblocks, wherein the quantization unit divides the macroblocks into small blocks based on quantization division information for dividing the macroblocks into the small blocks, each of which is quantized, and quantizes the orthogonal transform coefficient for each of the small blocks.</li><li>E7. The video encoding apparatus according to E6, further comprising a quantization division determining unit for obtaining quantization division information for dividing the macroblocks into small blocks, each of which is quantized, based on input pixel values in units of the macroblocks, wherein the quantization division information is set in the video encoding information.</li><li>E8. The video encoding apparatus according to E7, wherein the quantization division determining unit comprises <ul id="ul0005" list-style="none" compact="compact"><li>a division mode evaluation value calculating unit for calculating an evaluation value of a division shape for each division shape in which a macroblock is divided into small blocks, and</li></ul> a quantization division deciding unit for deciding a division shape of the macroblock based on the division mode evaluation value, and for outputting quantization division information corresponding to the division shape.</li><li>E9. The video encoding apparatus according to E6, wherein the orthogonal transform unit performs an orthogonal transform in units of small blocks, into which the macroblocks are further divided, based on macroblock division information of the orthogonal transform, which is set in the video encoding information, and the quantization unit divides the small blocks for quantization based on the macroblock division information of the orthogonal transform, and quantizes the orthogonal transform coefficient for each of the small blocks.</li><li>E10. The video encoding apparatus according to any one of E6 to E9, wherein the quantization unit quantizes a difference of a quantization parameter, which is used to quantize the orthogonal transform coefficient, from a more suitable quantization parameter among quantization parameters of encoded neighboring macroblocks or small blocks.</li><li>E11. The video encoding apparatus according to E10, wherein the more suitable quantization parameter is an intermediate value of the quantization parameters of the encoded neighboring macroblocks or small blocks.</li><li>E12. A video decoding apparatus for decoding a video encoded by the video encoding apparatus according to E7, comprising: <ul id="ul0006" list-style="none" compact="compact"><li>a quantization division obtaining unit for obtaining the quantization division information set in the video encoding information; and</li><li>an inverse quantization unit, wherein</li></ul> the inverse quantization unit performs inverse quantization for an orthogonal transform coefficient quantized for each of small blocks according to the obtained quantization division information in a macroblock to be processed.</li><li>E13. A video decoding apparatus for decoding a video encoded by the video encoding apparatus according to E9, comprising: <ul id="ul0007" list-style="none" compact="compact"><li>a quantization division obtaining unit for obtaining macroblock division information of the orthogonal transform, which is set in the video encoding information, and</li><li>an inverse quantization unit; wherein</li></ul> the inverse quantization unit performs inverse quantization for an orthogonal transform coefficient quantized for each of small blocks according to the obtained macroblock division information of the orthogonal transform in a macroblock to be processed.</li></ul>