JP5732125B2 - Video decoder that uses low resolution data to reduce power at low resolution - Google Patents

Description

  The present invention relates to a video decoder for reducing power.

  H. Existing video coding standards such as H.264 / AVC generally increase coding efficiency relatively at the cost of increased computational complexity. Power consumption is significant due to the relatively increased computational complexity, which is particularly problematic for low power devices such as mobile phones.

  In general, power reduction is achieved using two main technologies. The first technique for reducing power is opportunistic, and video coding systems reduce their processing power when processing sequences that are easy to decode. This reduction in processing power can be achieved by frequency scaling, voltage scaling, on-chip data prefetching (caching), and / or systematic idling methods. In many cases, as a result, the processing of the decoder is compliant with the standard. The second technique for reducing power is to discard the frame or image data during the decoding process.

  In general, this technique saves even more power, but at the cost of visibly degrading image quality. Furthermore, as a result, the processing of the decoder is not compliant with the standard.

  An embodiment of the present invention discloses a video decoder that decodes video, (a) an entropy decoding unit that decodes the bit stream that defines the video, and (b) the decoded bit. An inverse transform unit that transforms the stream; (c) a prediction unit that selectively executes intra prediction and motion compensated prediction based on the decoded bit stream; and (d) a compression used for the motion compensated prediction. A buffer including image data, wherein the compressed image data includes low resolution data and high resolution data, and the prediction unit uses the high resolution data based on the low resolution data. Predict both low and high resolution data sets using high resolution prediction information decoded from the above bitstream A video decoder.

  Another embodiment of the present invention discloses a video decoder for decoding video, wherein (a) an entropy decoding unit that decodes the bit stream that defines the video, and (b) decoded An inverse transform unit that transforms the bitstream; (c) a prediction unit that selectively executes intra prediction and motion compensated prediction on the decoded bitstream; and (d) used for the motion compensated prediction. (E) the compressed image data includes a low-resolution data set and a high-resolution data set, and the low-resolution data set is independent of the high-resolution data set. The low-resolution data set is used for decoding in the low-resolution mode, and the low-resolution data set and the high-resolution data set are used. Data sets are both a video decoder to be used when decoding in the high resolution mode.

  The above and other objects, features and advantages of the present invention will be readily understood in view of the following detailed description of the invention with reference to the accompanying drawings.

FIG. 1 shows a decoder. FIG. 2 shows the prediction at low resolution. FIG. 3A shows a decoder and a data flow in the decoder. FIG. 3B shows the decoder and the data flow in the decoder. FIG. 4 shows a sample structure of the frame buffer. FIG. 5 shows the integration of the frame buffer in the decoder. FIG. 6A shows representative pixel values of two blocks. FIG. 6B shows representative pixel values of two blocks.

  It is desirable to enable significant power savings typically associated with discarding frame data without any visible degradation in image quality or non-compliance with standards. If implemented properly, a system with minimal impact on coding efficiency can be used. In order to promote power savings while minimizing image degradation and loss of coding efficiency, the system should alternatively process low and high resolution data. The combination of low resolution data and high resolution data can be full resolution data. The use of low resolution data is particularly suitable when the resolution of the display is smaller than the resolution of the content being transmitted.

  Power is one factor in designing higher resolution decoders. One of the major contributors to power usage is memory bandwidth. Traditionally, memory bandwidth has increased with resolution and frame rate, and has often become a significant bottleneck and cost factor in designing systems. The second thing that is greatly involved in the use of power is the number of high resolution pixels. The number of high resolution pixels is directly determined by the resolution of the image frame, increasing the amount of processing and computation of the pixels. The amount of power required for processing each pixel is determined by the complexity of the decoding process. Historically, decoding complexity has increased as video coding standards have been “improved”.

  Referring to FIG. 1, the system includes an entropy decoding unit 10, an inverse transform unit 20 (such as inverse transform using IDCT), an intra prediction unit 30, a motion compensation prediction unit 40, an adder 80, A blocking filter 50 and a power adaptive loop filter 60 are provided, and a memory compression / decompression module associated with the buffer 70 is provided. The arrangement and selection of the various modules of the video system may be varied as needed. In one aspect, the system preferably reduces power requirements for both the high resolution pixel count of the buffer and the memory bandwidth. Using frame buffer compression techniques in video coder design reduces memory bandwidth. The purpose of the frame buffer compression technique is to reduce the memory bandwidth (and power) required to access the data in the reference picture buffer. Assuming that the reference picture buffer itself is a compressed version of the original image data, the reference frame can be compressed in many applications without significant coding loss.

  In order to deal with the high resolution pixel count, the video codec should support a low resolution processing mode with no drift. This means that the decoder may switch between processing points between low resolution and full resolution and comply with the standard. This can be realized by predicting both low resolution data and high resolution data using full resolution prediction information but using only low resolution image data. Furthermore, this can be improved by using a deblocking process that determines block distortion removal using only low resolution data. Deblocking can be applied to low resolution data, but can also be applied to high resolution data as needed. Low resolution pixel deblocking does not depend on high resolution pixels. Low resolution deblocking and high resolution deblocking may be performed sequentially and / or in parallel. However, deblocking of high resolution pixels may depend on low resolution pixels. In this method, since the low resolution processing is independent of the high resolution processing, the power saving mode is effective. On the other hand, since the high resolution processing depends on the low resolution processing, the image quality can be improved when necessary.

  Referring to FIG. 2, when processing in the low resolution mode (S10), the decoder can greatly reduce the number of pixels to be processed using the low resolution prediction and improved deblocking properties. This can be realized by predicting only the low resolution data (S12). Then, after predicting low resolution data, residual data is calculated only for low resolution pixels (ie, pixel positions), and residual data is not calculated for high resolution pixels (ie, pixel positions) (S14). Residual data is generally transmitted in the form of a bit stream. The residual data calculated for the low resolution pixel value has the same pixel value as the full resolution residual data at the position of the low resolution pixel. The main difference is that the residual data only needs to be calculated at the position of the low resolution pixels. Following the calculation of the residual, the low resolution residual is added to the low resolution prediction to determine a low resolution pixel value (S16). The obtained signal is deblocked. In order to reduce power consumption, it is desirable to perform deblocking only at low-resolution pixel positions (S18). Finally, the result is stored in the frame buffer of the reference picture for future prediction. Optionally, the result may be processed using an adaptive loop filter. The adaptive loop filter may relate to an adaptive loop filter for full resolution data, may be signaled independently, or may be omitted.

  A typical diagram of a system operating in the low resolution mode is shown in FIGS. 3A and 3B. The system may also have a mode that operates in full resolution mode. As shown in FIGS. 3A and 3B, entropy decoding 100 may be performed at full resolution. On the other hand, it is desirable that the inverse quantization / inverse DCT (inverse quantization IDCT) 200 and the prediction (intra prediction 300, motion compensated prediction (MCP) 400) are executed at a low resolution. Deblocking 500 is preferably performed in stages so that the deblocking of low resolution pixels does not depend on additional high resolution data. Finally, a frame buffer that includes memory compression stores low resolution data that is used for future predictions.

  The residual data (101) of the full resolution pixel is entropy decoded by the entropy decoding 100 shown in FIG. 3A. Pixels with shadows in the residual 101 represent low-resolution positions. On the other hand, pixels not shaded represent high-resolution positions. The inverse quantization / inverse DCT 200 inversely transforms only the low-resolution pixel data in the residual 101 to generate a residual-after-dequant-and-IDCT 201 after inverse quantization and inverse DCT.

  In the case of an intra picture, intra prediction 300 generates prediction information 301 only for low resolution positions (drawn by shaded pixels). The adder 800 adds the low-resolution pixel data of the residual 201 after inverse quantization and inverse DCT to the low-resolution pixel data in the prediction information 301, thereby reducing the low-resolution (drawn by the shaded pixels). Reconstruction data 801 is generated only for the position.

  In the case of an inter picture, the MCP 400 shown in FIG. 3B reads out the low-resolution pixel data of the reference picture (drawn with the shaded pixels in the reference picture data 702) from the memory 700 and has already removed it. The high-resolution pixel data thus generated is generated by interpolation. For example, as indicated by the interpolation 401, the MCP 400 generates high resolution pixel data C from the low resolution pixel data of the adjacent pixel group by interpolation. a) Average value of low-resolution pixel data of pixel groups located above and below C, b) Average value of low-resolution pixel data of pixel groups located on the left and right sides of C, c) Above and below C The average values of the average values of the low-resolution pixel data of the pixel groups located on the left side, the left side, and the right side may be used as interpolation.

Deblocking 500 is performed in stages. Deblocking 500 first filters the low resolution data (501). Next, the deblocking 500 filters the high resolution data (502). More specifically, the deblocking 500 is performed by the following method.
Step 1) (501)
Deblocking 500 is applied only to low resolution data using low resolution data and high resolution data by interpolation.
Step 2) (502)
Deblocking 500 is applied only to high resolution data using low resolution data and high resolution data by interpolation.

  The picture after deblocking 500 is stored in memory 700. The following describes the pictures (701, 702, 703) stored in the memory 700 after the deblocking 500 and read out for the MCP 400. The picture 502 after the deblocking 500 is a full resolution picture, but may be referred to as a picture 701. The picture 701 after the deblocking 500 is thinned out with a checkerboard pattern, so that only the low-resolution position is left and stored in the memory 700. When used for prediction, the thinned out high resolution pixel data (drawn with 702 unshaded pixels) is interpolated and the interpolated picture is used to generate the predicted picture The

  The frame buffer compression technique is preferably a component of a low resolution function. The frame buffer compression technique desirably divides the pixel data into a plurality of sets such that the first set of pixel data does not depend on other sets. In one form, the system uses a checkerboard as shown in FIG. In FIG. 4, the positions of the pixel groups with shadows belong to the first group, and the pixel groups without shadows belong to the second group. Other sample structures may be used as needed. For example, pixel groups along the column direction may be assigned to the first group every other column. Alternatively, pixel groups along the row direction may be assigned to the first group every other row. Similarly, pixel groups along the column direction may be assigned to the first group every other column, and pixel groups along the row direction may be assigned to the first group every other row. Any method suitable for dividing into groups of pixel groups can be used.

  For memory compression / decompression, in the frame buffer compression technique, it is desirable that the second set of pixels for pixels be linearly predicted from the first set of pixels for pixels. The prediction may be determined in advance. Alternatively, the prediction may vary spatially or may be determined using any other suitable technique.

In one embodiment, the first set of pixels for a pixel is encoded. This encoding may use any suitable technique. Examples of such techniques include block truncation coding (BTC) described in the following documents,
・ "Digital Video Bandwidth Compression Using Block Truncation Coding" co-authored by Healy, D. and Mitchell, O., IEEE Transatcions on Communications [ legacy, pre-1988], Vol. 29, No. 12 (pages 1809-1817), December 1981,
Absolute moment block truncation coding (AMBTC) described in the following documents,
・ "Absolute Moment Block Truncation Coding and Its Application to Color Images", co-authored by Lema, M. and Mitchell, O., IEEE Transatcions on Communications [legacy, pre-1988], Volume 32, No. 10 (pages 1148 to 1157), October 1984,
Another example is scalar quantization. Similarly, the second set of pixels for a pixel may be encoded and predicted using any suitable technique. For example, the prediction may be made using a linear process known for frame buffer compression encoders and frame buffer compression decoders. Then, the difference between the predicted value and the pixel value may be calculated. Finally, the calculated difference may be compressed. In one embodiment, the system may compress the second set of pixels using block truncation coding (BTC). In another embodiment, the system may compress the second set of pixels using absolute moment block truncation coding (AMBTC). In another embodiment, the system may compress the second set of pixels using quantization. In yet another embodiment, the system may predict pixel values for the second set of pixels using bilinear interpolation. In yet another embodiment, the system may predict the pixel values of the second set of pixels using bicubic interpolation. In another embodiment, the system predicts the pixel values of the second set of pixels using bilinear interpolation and uses absolute moment block truncation coding (AMBTC) for the second set of pixel values. The residual between the predicted pixel value and the second set of pixel value groups may be compressed.

  A characteristic of the frame buffer compression technique is control using a flag for indicating that low-resolution processing is possible. In one form, if this flag does not indicate that low resolution processing is possible, the frame buffer decoder may optionally include a first set of pixel value groups (ie, low resolution pixel data). And a second set of pixel values (ie, high resolution pixel data) predicted from the first set of pixel values and refined by optional residual data. . In another form, if the flag indicates that low resolution processing is possible, the frame buffer decoder may include a first set of pixel values, optionally compressed, and a first set. A second set of pixel value groups is generated that are predicted from the pixel value groups, but not refined by the optional residual data. Therefore, this flag indicates whether or not to use optional residual data. The residual data may indicate the difference between the predicted pixel value group and the actual pixel value group.

  For a frame buffer compression encoder, if the flag indicates that low resolution processing is not possible, the encoder stores the first set of pixel values, possibly in a compressed format. The encoder predicts a second group of pixel values from the first group of pixel values. In some embodiments, the encoder identifies the residual between the predicted pixel value and the actual pixel value and stores this residual, possibly in a compressed form. In some embodiments, the encoder selects a preferred prediction method from among multiple prediction methods for the second set of pixels. Then, the encoder stores the selected prediction method in the frame buffer. In one embodiment, the multiple prediction methods consist of multiple linear filters, and the encoder calculates a predicted pixel value for each linear filter and selects a linear filter that calculates the predicted pixel value closest to the actual pixel value. By doing so, the prediction method is selected. In one embodiment, the plurality of prediction methods are composed of a plurality of linear filters, and the encoder calculates a group of predicted pixel values in a block composed of a plurality of pixel positions for each linear filter, and the group of predicted pixel values is actually A prediction method is selected by selecting a linear filter that calculates those predicted pixel values in the block closest to the group of pixel values. A pixel group block is a set of pixels inside an image. The block whose predicted pixel value group is closest to the actual pixel value group is determined by selecting the block having the smallest sum of absolute values of differences between the actual pixel value group and the predicted pixel value group. Also good. Alternatively, the sum of square differences may be used to select a block. In other embodiments, the residual is compressed by block truncation coding (BTC). In one embodiment, the residual is compressed by absolute moment block truncation coding (AMBTC). In one embodiment, the parameter group used to compress the second set of pixel groups is determined from the parameter group used to compress the first set of pixel groups. In one embodiment, the first set of pixels and the second set of pixels use AMBTC, and the first parameter used in the AMBTC scheme for the first set of pixels is the second set. This is related to the first parameter used in the AMBTC system for the pixel group. In one embodiment, the first parameter used for the second set of pixels is the same as the first parameter used for the first set of pixels and is not stored. In another embodiment, the first parameter used for the second set of pixels is related to the first parameter used for the first set of pixels. In one embodiment, the relationship may be defined as a scale factor, and the scale factor may be stored instead of the first parameter used for the second set of pixels. In other embodiments, the relationship may be defined as an index into a scale factor look-up table, where the index is stored instead of the first parameter used for the second set of pixels. It may be. In another embodiment, the relationship may be predetermined. In another embodiment, the encoder combines the selected prediction method and the residual determination step. In comparison, if the flag indicates that low resolution processing is possible, the encoder stores the first set of pixels, possibly in a compressed format. However, the encoder does not store residual information. In the embodiment described above that identifies the selected prediction method, the encoder does not calculate the selected prediction method from the reconstructed data. Instead, any selected prediction method is communicated from the encoding to the decoder.

  By transmitting the flag, the low resolution decoding function becomes effective. The decoder need not decode the low resolution sequence even if the flag indicates that low resolution decoding is possible. Alternatively, the decoder may decode either a full resolution sequence or a low resolution sequence. In these sequences, the pixel values after decoding at the pixel positions on the low resolution grid are the same. These sequences may or may not have the same pixel values after decoding at pixel positions on the high resolution grid. The flag transmission may be performed for each frame, may be performed for each sequence, or may be performed based on other criteria.

The decoder preferably performs the following steps when a flag appears in the bitstream.
(A) Invalidate the residual calculation by the frame buffer compression technique. As shown in FIG. 5, this step includes the step of invalidating the calculation of residual data while reading the reference frame in addition to the step of invalidating the calculation of residual data while saving the reference frame. ing.
(B) As described above, low resolution deblocking is performed using a low resolution pixel value group. As described above, another deblocking process is used for sample positions at higher resolution positions.
(C) Save the reference frame before applying the adaptive loop filter.

  With these changes, the decoder can continue processing in the full resolution mode. Specifically, for future frames, the decoder retrieves the full resolution frame from the compressed reference buffer, performs motion compensation, adds the residual, performs deblocking processing, and performs filtering by the loop filter. It can be carried out. A full resolution frame is then obtained. This frame still contains frequency components that occupy the entire range of the grid of full resolution pixels.

  However, instead, the decoder may choose to process only low resolution data. This is possible because the low resolution grid is independent of the higher resolution grid in the buffer compression structure. For motion compensation, the interpolation process may be modified to take advantage of the fact that high resolution pixels are in a linear relationship with low resolution data. Therefore, the motion compensation process may be executed at a low resolution using the changed interpolation filter. Similarly, for residual computation, the system may take advantage of the fact that low resolution data does not depend on high resolution data in subsequent steps of the decoder. Therefore, the system uses a simple inverse transform process that only calculates low resolution pixels from the full resolution transform coefficients. Finally, the system uses a deblocking filter that deblocks low resolution data that is independent of the high resolution pixels (high resolution depends on the low resolution). This is because there is a linear relationship between high resolution data and low resolution data.

  For the block size of 8 × 8, the existing deblocking filter in the JCT-VC under consideration test model JCTVC-A119 is desirable. In the luminance deblocking filter, the process starts by determining whether the block boundary is deblocked. This can be realized by calculating the following equation.

Here, d is a variable, and pi _j and qi _j are pixel values. The position of the pixel value is depicted in FIG. In FIG. 6, two 4 × 4 encoding units are depicted. However, the pixel value may be determined from an arbitrary block size by considering the position of the pixel related to the block boundary.

  Next, the value obtained as d is compared with a threshold value. If the value d is smaller than the threshold value, the deblocking filter is activated. If the value d is greater than or equal to the threshold value, filtering is not applied and the value of the pixel subjected to the deblocking process becomes the same value as the input pixel value. Note that the threshold may be a function of a quantization parameter, and may be described as beta (QP). Deblocking is determined independently at the horizontal and vertical boundaries.

  If the boundary d value is such a value that determines deblocking, the process then proceeds to determine the type of filter to apply. The type of filter used in the deblocking process may be a strong filter or a weak filter. The selection of the strength of the filtering is based on the already calculated d, beta (QP), as well as additional local differences. This is calculated for each line (row or column) of the deblocking boundary. For example, the following equation is calculated for the first row of pixel positions shown in FIG.

Here, t _C is a threshold and is generally a function of the quantization parameter QP.

  For luminance samples, if the above process determines that the boundary is deblocked and then the line (row or column) is deblocked with a weak filter, the filtering process is as follows: Also good. Here, a description will be given by filtering processing for the boundary between block A and block B in FIG. Processing is

It becomes. Here, Δ is an offset, and Clip _0-255 () is an operator that maps the input value to the range [0,255]. In another embodiment, the operator may map the input value to another range ([16,235], [0,1023], or other range).

  For luminance samples, if the above process determines that the boundary is deblocked and then the line (row or column) is deblocked with a strong filter, the filtering process is as follows: Also good. Here, a description will be given by filtering processing for the boundary between block A and block B in FIG. Processing is

It becomes. Here, Clip _0-255 () is an operator that maps the input value to the range [0,255]. In another embodiment, the operator may map the input value to another range ([16,235], [0,1023], or other range).

For chrominance samples, if the above process determines that the boundary is deblocked , all lines (rows or columns ) of the chrominance component are processed by weak filtering. Here, a description will be given by filtering processing for the boundary between block A and block B in FIG. In FIG. 6, it is assumed that these blocks include color difference pixel values. Processing is

It becomes. Here, Δ, also referred to as delta in this specification, is an offset, and Clip _0-255 () is an operator that maps input values to the range [0,255]. In another embodiment, the operator may map the input value to another range ([16,235], [0,1023], or other range).

  The pixel positions in the image frame may be divided into two or more sets. If the flag is transmitted in the bitstream or communicated in any way, the system will exclude the pixel values for the second set of pixel locations and enable processing for the first set of pixel locations. To do. FIG. 4 shows an example of this division. In FIG. 4, the block is divided into two sets of pixels. The first set corresponds to a position with a shadow, and the second set corresponds to a position with no shadow.

  When this other mode is enabled, the system changes the previous deblocking process as follows.

  First, in determining whether a boundary should be deblocked, the system uses the above equation or other suitable equation. However, for pixel values corresponding to pixel positions that do not belong to the first set of pixel groups, pixel values derived from the first set of pixel positions may be used. 6A and 6B, p01, p03, p05, p07, q00, q02, q04, and q06 are the first set of pixels calculated by entropy decoding, inverse transformation, and prediction. p00, p02, p04, p06, q01, q03, q05, q07 are calculated by the equations as shown in FIG. 3B or FIG.

Equation 1, Equation 2, Equation 3, Equation 4, and Equation 5 are calculated using these pixel values.

  In one embodiment, the system derives the pixel value as a linear weighting of adjacent pixel values of the first set of pixels. In the second embodiment, the system uses bilinear interpolation of a plurality of pixel values of the first set of pixels. In a preferred embodiment, the system includes a pixel value of the first set of pixel groups that is above the current pixel position, and a pixel value of the first set of pixel groups that is the current pixel. A linear average of pixel values below the position is calculated. The above explanation is based on the premise that the system processes a vertical block boundary (applies a horizontal deblocking process). When the system processes horizontal block boundaries (applying vertical deblocking), it calculates the average of the pixels to the left and right of the current position. In another embodiment, the system may limit the average calculation of pixel values within the same block. For example, if the pixel value located above the current pixel does not belong to the same block but the pixel value located below the current pixel belongs to the same block, the pixel value of the current pixel is below the current pixel. It is set to a value equal to the pixel value located on the side.

  Second, in calculating whether to use a filter with a strong boundary or a weak filter, the system may use the same method as described above. That is, pixel values that do not correspond to the first set of pixels are derived from the first set of pixels. After making the aforementioned determination by computation, the system may process the first set of pixels based on the determination. The decoder processing the subsequent set of pixels processes the subsequent set of pixels based on the same determination.

  If the above process determines that the boundary is deblocked and then the line (row or column) is deblocked with a weak filter, the system may use the weak filtering process described above. However, when calculating the value of Δ, the system does not use the pixel values corresponding to the group of pixels following the first group. Alternatively, the system may derive pixel values as described above. By way of example, the value of Δ is applied to a first set of actual pixel values, and the delta value (value of Δ) is applied to a second set of actual pixel values.

  If the above process determines that the boundary is deblocked and then the line (row or column) is deblocked with a strong filter, the system may perform the following process.

  That is, in one embodiment, the system may use the strong luminance filter equation described above. However, for pixel values that are not in the first set of pixel positions, the system may derive the pixel values from the first set of pixel positions, as described above. Then, the system stores the calculation result of the filter processing for the first set of pixel positions. Subsequently, for a decoder that generates the subsequent pixel position as an output, the system performs the calculation of the strong filter previously obtained for the first pixel position and the reconstruction process of the subsequent pixel position (in the filter process). No) For the calculation result, the above-described strong luminance filter formula is used. The system then applies this filter only to subsequent pixel locations. The output will be the filtered first pixel location corresponding to the initial filtering and the filtered subsequent pixel location corresponding to the additional filter path.

  In summary, as described above, the system uses the first pixel value to interpolate the missing pixel value and calculate the result of the strong filtering process for the first pixel value. The system then updates the missing pixel value to the actual reconstruction value and calculates the result of a strong filtering process for the missing pixel position.

  In the second embodiment, the system uses the strong luminance filter equation described above. The system derives pixel values that are not in the first set of pixel positions from the first set of pixel positions, as described above. The system then uses the derived value to calculate a strong filtering result for both the first set of pixel positions and the subsequent set of pixel positions. Finally, the system calculates a weighted average of the reconstructed pixel value at the subsequent position and the output of the strong filtering process for the subsequent position. In one embodiment, the weight is transmitted from the encoder to the decoder. In another embodiment, the weight is a fixed value.

  If the above process determines to deblock the boundary in the color difference, the system uses a weak filtering process as described above. However, when calculating the value of Δ, the system does not use the pixel values corresponding to the group of pixels following the first group. Instead, it is desirable for the system to derive pixel values as described above. By way of example, the value of Δ is applied to a first set of actual pixel values, and the delta value (value of Δ) is applied to a second set of actual pixel values.

  The terms and expressions used up to now in the specification are used as terms for description rather than limitation, and the use of such terms and expressions excludes equivalents of the features shown and described or portions thereof. There is no intention to do. It should be recognized that the scope of the invention is defined and limited only by the claims that follow.

Claims (10)

A video decoder for decoding video,
(A) an entropy decoding unit that decodes a bitstream that defines the video;
(B) an inverse quantization and inverse transform unit that inversely quantizes and inversely transforms the decoded bitstream;
(C) a prediction unit that selectively executes intra prediction and motion compensation prediction based on the decoded bitstream;
(D) a buffer including image data used for the intra prediction and motion compensation prediction, the image data including low resolution data and high resolution data;
(E) an adder that generates reconstructed image data based on the prediction data generated by the prediction unit and the bitstream inversely quantized and inversely transformed by the inverse quantization inverse transform unit;
(F) a deblocking unit that selectively performs deblocking processing on the reconstructed image data,
The entropy decoding unit receives, from the bitstream, a flag indicating a low resolution mode in which low resolution decoding is possible or a high resolution mode in which low resolution decoding is not possible ,
A video decoder , wherein when the flag indicates a low resolution mode, low resolution image data is stored in the buffer, and high resolution image data is not stored .   The video decoder according to claim 1, wherein when the flag indicates the low resolution mode, the prediction unit performs processing on a low resolution pixel position using only low resolution data.   The prediction unit performs processing on a high-resolution pixel position using low-resolution data and high-resolution data interpolated from the low-resolution data when the flag indicates the low-resolution mode. Item 3. The video decoder according to Item 2.   The video decoder according to claim 1, wherein when the flag indicates the high resolution mode, the prediction unit performs processing on a high resolution pixel position using high resolution data. The low resolution data is independent of the high resolution data,
When the flag indicates the low resolution mode, only the low resolution data is used,
The video decoder according to claim 1, wherein when the flag indicates the high resolution mode, the low resolution data and the high resolution data are used together.   The deblocking unit uses only the reconstructed image data at the low resolution pixel position for deblocking processing of the reconstructed image data at the low resolution pixel position when the flag indicates the low resolution mode. The video decoder according to claim 1.   The deblocking unit also uses the interpolated image data at the high resolution pixel position for deblocking processing of the reconstructed image data at the high resolution pixel position when the flag indicates the low resolution mode. Item 7. The video decoder according to Item 6.   The deblocking unit uses the reconstructed image data at the high resolution pixel position for deblocking processing of the reconstructed image data at the high resolution pixel position when the flag indicates the high resolution mode. Item 4. The video decoder according to Item 1.   4. The video decoder according to claim 3, wherein the interpolated high resolution data is linearly predicted from the low resolution data.   The video decoder according to claim 7, wherein the reconstructed image data at the high resolution pixel position is linearly predicted from the reconstructed image data at the low resolution pixel position.

Images