JP4533157B2 - Image decoding method - Google Patents

Description

The present invention relates to an image decoding method.

  Motion compensation is a technique used for video compression. This improves the compression rate by transmitting (or recording) image information of blocks similar between frames as motion vector information and a difference value. This motion compensation is also used in the MPEG system known as a moving image compression system.

  In the MPEG system, an inter-frame encoding I picture and an inter-frame prediction encoding P picture that encodes a difference between a prediction value from another frame that is an I picture or a P picture, and a P picture or an I picture There is a B picture of inter-frame predictive coding that encodes a difference from a predicted value from two frames before and after a certain frame. For motion compensation, a maximum of one picture can be referred to in the P picture, and a maximum of two pictures can be referred to in the B picture. The accuracy of the motion vector is 0.5 pixel units.

In the motion compensation process at the time of decoding, the position on the screen is compensated by a block unit called a macro block (MB) using a motion vector. For a motion vector in units of 0.5 pixels, a pixel value calculated by linear interpolation in the half-pixel creation process is used as a reference block. As shown in FIG. 10, there are three types of positions indicated by a square symbol □, a triangle symbol Δ, and a hexagonal symbol in FIG. 10 as positions that may be interpolated due to the presence of a motion vector in units of 0.5 pixels. .

  FIG. 8 shows a schematic block diagram of the decoding processing apparatus in the first conventional example. The Huffman decoding unit 500 performs Huffman decoding on input code data for each macroblock, the inverse quantization unit 501 performs inverse quantization processing on the output data of the Huffman decoding unit, and the inverse orthogonal transform unit 502 performs inverse quantization. The output data of the means 501 is subjected to inverse orthogonal transform processing. Thereafter, for an intra macroblock (intra MB) that does not require motion compensation processing, the output data of the inverse orthogonal transform unit 502 is output as it is as a decoding result.

  The following motion compensation processing is performed on a non-intra macroblock (non-intra MB) that requires motion compensation processing. The block reference means 510 refers to a desired block using a motion vector. A half-pixel block is created by linear interpolation only when the half-pixel creating means 511 is a 0.5 unit vector. The adding unit 512 adds the output of the inverse orthogonal transform unit 502 and the output of the half pixel creating unit 511. The output data of the adding means 512 is output as a decoding result.

  For the conventional example as described above, many methods for speeding up the decoding process have been proposed, particularly focusing on the half-pixel creation process when implemented by software. For example, a configuration described in Patent Document 1 is known.

  FIG. 9 is a block diagram showing a schematic configuration of a decoding processing apparatus according to Conventional Example 2 described in Patent Document 2.

  The Huffman decoding unit 700 performs Huffman decoding on the input code data for each macroblock. The inverse quantization unit 701 performs an inverse quantization process on the output data of the Huffman decoding unit 700. The inverse orthogonal transform unit 702 performs an inverse orthogonal transform process on the output data of the inverse quantization unit 701. Thereafter, for intra MBs that do not require motion compensation processing, the output data of the inverse orthogonal transform means 702 is output as a decoding result.

Non-intra MBs that require motion compensation processing are processed as follows. For non-intra MB of P picture, block reference means 710 refers to a desired block using a motion vector, and creates a half-pixel block by linear interpolation only when half-pixel creation means 711 is a 0.5 unit vector. and, adding means 712 adds the output of the inverse orthogonal transform unit 7 02 and the output of the half-pixel creation means 7 11. The output of the adding means 712 is output as a decoding result.

  Of the non-intra MBs that require motion compensation processing, for the B picture non-intra MB, the motion vector correction unit 720 performs correction by truncating the decimal part of the motion vector, and the block reference unit 721 corrects the motion. The adder 722 adds the output of the inverse orthogonal transform unit 702 and the output of the block reference unit 721 by referring to the desired block using the vector. The output of the adding means 712 is output as a decoding result.

With the method of Conventional Example 2, it is possible to reduce the amount of computation and increase the speed while suppressing the accumulation of image disturbance due to the absence of half-pixel creation.
JP-A-9-322175

  In the configuration of Conventional Example 1, the greater the number of motion vectors after the decimal point, the higher the probability that the half-pixel creation process is performed on the same pixel. As a result, the calculation amount is large and the processing load is heavy.

  With the method of Conventional Example 2, it is possible to reduce the amount of computation and increase the speed while suppressing the accumulation of image disturbance due to the absence of half-pixel creation. However, with respect to the B picture, image disturbance due to not creating half pixels is conspicuous, and the image quality is degraded.

It is an object of the present invention to provide an image decoding method that eliminates such inconveniences.

An image decoding method according to the present invention is an image decoding method, which is a method for decoding a non-intra macroblock that requires a motion compensation process and refers to a desired block using a motion vector. In addition, a switching step for switching between the first motion compensation step and the second motion compensation step based on the expected value of the number of vectors related to the referenced frame, and a first half pixel generation process in all directions of the motion vector are performed. And a second motion compensation step that refers to a half-pixel frame created in advance by performing half-pixel creation processing in all directions of the motion vector, and the switching step is applied to the referenced frame. When the expected value of the number of vectors is smaller than a predetermined value, the process is switched to the first motion compensation step, and the expected value of the number of vectors for the referenced frame is the predetermined value. Ri in case large, and switches to the second motion compensation process.

  According to the present invention, by calculating the number of motion vectors in the motion compensation process existing in the decoding process and performing the optimal motion compensation process, the amount of calculation related to the half-pixel creation process is greatly reduced, and high-speed decoding is performed. Realize processing.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

  FIG. 1 shows a schematic block diagram of an image decoding apparatus according to the first embodiment of the present invention. The Huffman decoding unit 100 performs Huffman decoding on the input code data for each macroblock. The inverse quantization unit 101 performs an inverse quantization process on the output of the Huffman decoding unit 100. The inverse orthogonal transform unit 102 performs an inverse orthogonal transform process on the output of the inverse quantization unit 101.

  The switching means 103 switches the following four motion compensation processes from the statistics of the composition ratio of P picture and B picture and the number of motion vectors.

  In the first motion compensation process, the output data of the inverse orthogonal transform unit 102 is output as it is as a decoding result for an intra macroblock (intra MB) that does not require the motion compensation process. For non-intra macroblocks (non-intra MB) that require motion compensation processing, block reference means 110 refers to a desired block using motion vectors, and half-pixel creation means 111 is a 0.5 unit vector. Only half-pixel blocks are created by linear interpolation, and the adding means 112 adds the output of the inverse orthogonal transform means 102 and the output of the half-pixel creating means 111. The output data of the adding means 112 is output as a decoding result.

  In the second motion compensation processing, for a non-intra macroblock (non-intra MB) that requires motion compensation processing, the block reference means 120 uses a motion vector in the horizontal direction to convert an integer vector from a conventional frame to a horizontal A non-integer vector in the direction refers to a desired block from a previously created horizontal half-pixel frame, and the half-pixel creating means 121 creates a half-pixel block by linear interpolation only when the vertical direction is a non-integer vector. The adding means 122 adds the output of the inverse orthogonal transform means 102 and the output of the half-pixel creating means 121. Further, for the I and P pictures, the horizontal half-pixel creating means 123 creates a horizontal half-pixel block and stores it in the horizontal half-pixel frame.

In the third motion compensation process, first, with respect to a non-intra macroblock (non-intra MB) that requires a motion compensation process, the block reference unit 130 uses a motion vector to set both integers in the horizontal and vertical directions or both in the horizontal and vertical directions. The integer vector is the desired block from the conventional frame, the horizontal only non-integer vector is from the previously created horizontal half-pixel frame, and the vertical only non-integer vector is from the previously created vertical half-pixel frame. , Half-pixel creating means 131 creates a half-pixel block by linear interpolation only when both non-integer vectors in the horizontal and vertical directions, and adding means 1 3 2 and the output of inverse orthogonal transform means 102 The outputs of the pixel creation means 131 are added. Further, for the I and P pictures, the horizontal / vertical direction half-pixel creating unit 133 creates half-pixel blocks in the horizontal and vertical directions and stores them in the horizontal and vertical half-pixel frames, respectively.

In the fourth motion compensation processing, first, regarding a non-intra macroblock (non-intra MB) that requires motion compensation processing, the block reference means 140 uses the motion vector to convert both integer vectors in the horizontal and vertical directions as in the conventional case. from the frame, from the non-integer vector only the horizontal direction horizontal half-pixel frame created in advance, the vector of vertical but only non-integer from vertical half pixel frame previously prepared, a non-integer both in horizontal and vertical direction vector Refers to a desired block from an oblique half pixel frame created in advance, and the adding means 141 adds the output of the inverse orthogonal transform means 102 and the output of the block reference means 140. Moreover, I, with respect to P picture, all positioned half-pixel creation unit 14 2, the horizontal direction, to create a half-pixel block in the vertical direction and the oblique direction, respectively stored in the horizontal, vertical, and oblique directions and a half pixel frame.

The operation of the switching unit 103 will be described with reference to FIGS. The switching unit 103 receives information on the number of vectors per macroblock from each of the block reference units 110, 120, 130, and 140, and for the P picture and the B picture, the expected number of vectors Vp and Vb per macroblock, respectively. Is calculated. However, 0 ≦ Vp ≦ 1 and 0 ≦ Vb ≦ 2. Using this result and the appearance period M of the P picture indicating the composition ratio of the P picture and the B picture, the switching unit 103 obtains the expected value E of the number of vectors related to the referenced frame using the formula shown in FIG. From the result, the above-described four motion compensation processes are switched according to the determination condition shown in FIG.

  As described above, in this embodiment, it is possible to measure the overlapping degree of the half-pixel creation process based on the statistics of the composition ratio of the P picture and the B picture and the number of motion vectors. Applicable.

  FIG. 4 is a schematic block diagram of the second embodiment of the present invention. The Huffman decoding unit 300 performs Huffman decoding on the input code data for each macroblock, the inverse quantization unit 301 performs inverse quantization processing on the output of the Huffman decoding unit, and the inverse orthogonal transform unit 302 performs inverse quantization. The output of the means 301 is subjected to inverse orthogonal transform processing. Thereafter, the switching unit 303 switches the following four motion compensation processes from the statistics of the composition ratio of P picture and B picture and the number of motion vectors.

  In the first motion compensation process, with respect to an intra macroblock (intra MB) that does not require a motion compensation process, the output data of the inverse quantization means 302 is output as it is as a decoding result. For a non-intra macroblock (non-intra MB) that requires motion compensation processing, the block reference unit 310 refers to a desired block using a motion vector, and the half-pixel generation unit 311 uses a 0.5 unit vector. Only in some cases, a half-pixel block is created by linear interpolation, and the adding means 312 adds the output of the inverse orthogonal transform means 302 and the output of the half-pixel creating means 311. The output of the adding means 312 is output as a decoding result.

In the second motion compensation process, first, for a non-intra macroblock (non-intra MB) that requires a motion compensation process, the block reference unit 320 uses a motion vector to specify the direction (horizontal or vertical). ) Is an integer from a conventional frame, and a vector whose direction (horizontal or vertical) specified by the switching means 303 is a non-integer is to create a half pixel by referring to a desired block from a half pixel frame created in advance. A half-pixel block is created by linear interpolation only when the means 321 is a non-integer vector in the direction (vertical or horizontal) not specified by the switching means 303, and the adding means 322 outputs the half-pixel and the output of the inverse orthogonal transform means 302. The outputs of the creating means 321 are added. The output of the adding means 3 2 2 is output as a decoding result. Further, for the I picture and P picture, the unidirectional half-pixel creating unit 323 creates a half-pixel block in the direction (horizontal or vertical) designated by the switching unit 303 and stores it in the half-pixel frame.

The operation of the switching unit 303 will be described with reference to FIGS. The switching unit 303 receives vector information per macroblock from the block reference units 310 and 320 , and for the P picture and B picture, the expected value HPp of the number of non-integer vectors in the horizontal or vertical direction per macroblock, respectively. (0 ≦ HPp ≦ 1), HPb (0 ≦ HPb ≦ 2), expected value HPhp (0 ≦ HPhp ≦ 1), HPhb (0 ≦ HPhb ≦ 2) in the horizontal direction, and vertical direction Thus, the expected values HPvp (0 ≦ HPvp ≦ 1) and HPvb (0 ≦ HPvb ≦ 2) of the non-integer number of vectors are calculated. Of these results, the expected values HPp and HPb of the number of non-integer vectors in the horizontal or vertical direction and the appearance period M of the P picture indicating the composition ratio of the P picture and the B picture are used in the equation shown in FIG. An expected value E of the number of non-integer vectors related to the reference frame is obtained. From this result, the two motion compensation processes described above are switched according to the determination condition shown in FIG.

  When the second motion compensation process is selected, the appearance period M of the P picture, the expected values HPhp and HPhb of the non-integer number of vectors in the horizontal direction, and the expected value HPvp of the non-integer number of vectors in the vertical direction , HPvb, the expected values Eh, Ev of the number of non-integer vectors in the horizontal direction and the vertical direction applied to the frame to be referred to are obtained by the equation shown in FIG. The direction of the half-pixel frame to be created (horizontal or vertical) is notified, and the result is notified to the related block reference means 320, half-pixel creation means 321 and unidirectional half-pixel creation means 323.

  In this way, the effectiveness of the half-pixel creation process created in advance can be measured by the statistics of the composition ratio of P picture and B picture and the number of non-integer motion vectors. By using this measurement result, an optimal motion compensation process can be applied.

It is a schematic block diagram of the first embodiment of the present invention. It is a calculation formula used with the switching means of 1st Example. This is a switching condition used by the switching means of the first embodiment. It is a schematic block diagram of the second embodiment of the present invention. It is the calculation formula 1 used with the switching means of 2nd Example. This is a switching condition used by the switching means of the second embodiment. It is the calculation formula 2 used with the switching means of 2nd Example. It is a schematic block diagram of a first conventional example. It is a schematic block diagram of a 2nd prior art example. It is operation | movement explanatory drawing regarding a motion compensation process.

Explanation of symbols

100: Huffman decoding means 101: Inverse quantization means 102: Inverse orthogonal transformation means 103: Switching means 110: Block reference means 111, 121, 131: Half pixel creation means 112, 122, 132: Addition means 123: Horizontal half pixel Creation means 133: Horizontal vertical half-pixel creation means 142: Omni-directional half-pixel creation means 300: Huffman decoding means 301: Inverse quantization means 302: Inverse orthogonal transform means 303: Switching means 310, 320: Block reference means 311 and 321 : Half pixel creation means 312, 322: addition means 323: unidirectional half pixel creation means 500: Huffman decoding means 501: inverse quantization means 502: inverse orthogonal transform means 510: block reference means 511: half pixel creation means 512: addition Means 700: Huffman decoding means 701: Inverse quantization means 702: Inverse orthogonal transform means 10,721: block reference means 711 is a half pixel creation means 712, 722: adder means 720: the motion vector correction unit

Claims (1)

An image decoding method,
When referring to a desired block using a motion vector for a non-intra macroblock that requires motion compensation processing , the first motion compensation step is performed based on the expected value of the number of vectors for the referenced frame. A switching step for switching to the second motion compensation step;
A first motion compensation step for performing half-pixel creation processing in all directions of the motion vector;
A second motion compensation step that refers to a half-pixel frame created in advance by performing half-pixel creation processing in all directions of the motion vector
And
The switching step switches to the first motion compensation step when the expected value of the number of vectors applied to the referenced frame is smaller than a predetermined value, and the expected value of the number of vectors applied to the referenced frame is less than the predetermined value. If larger , the image decoding method is switched to the second motion compensation step .

Images