JP2933561B2 - Video transcoder - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a moving image code conversion device, and more particularly to a moving image code conversion device for converting a moving image code into a videophone code.

[0002]

2. Description of the Related Art Recently, a transcoding technique for converting a bit rate of a digital signal obtained by encoding a moving image into a moving image has become important. Video coding standards used in such audio-visual communication systems are standardized for bit rate divisions. The transcoding technique is a technique for converting a data stream having a predetermined bit rate into another bit stream having a different bit rate. However, a problem occurs when one signal form interfaces with another signal form.

As an example of image transmission using such a transcoding technique, a "transcoding method and apparatus" described in Japanese Patent Application Laid-Open No. 8-51631 is known.

This publication discloses a transcoding technique in which a variable length decoding channel section is provided at the subsequent stage with a variable length coding channel section, and a prediction function section having an image memory and a motion vector compensation circuit is connected between these two channels. Are listed.

[0005]

The above-described conventional moving image transcoder described above writes data into an image memory when converting a bit stream, reads data from the same image memory, and transcodes the data as it is. Since the input is input to an encoder that performs H.264 encoding, a bit stream of an image having a different spatial resolution, for example, an MPEG2 bit stream (luminance 720 × 576 pixels) is converted to an H.264 image. 263
It has the disadvantage that it cannot be converted to a bit stream (brightness 360 × 288 pixels).

It is an object of the present invention to provide a moving image code conversion apparatus for performing bit stream conversion of moving images having different spatial resolutions.

[0007]

Video code conversion apparatus of the present invention, in order to solve the problems] decodes the moving image data obtained by image compression, the decoded image
Image is intra-frame coded image (I-picture), frame
Inter prediction image (P picture), bi-directional prediction
Any picture type of encoded image (B picture)
Image identification data and motion vector data
Image decoding means for outputting compressed pixel data and outputting image data reproduced from the compressed pixel data; and outputting corrected image data which has been subjected to the thinning processing of the image data, and selecting any picture type based on the image identification data.
It is determined whether the image is an image.
Decimation processing and motion vector interpolation for vector data
Image conversion means for outputting corrected vector data calculated formally ; image coding means for compressing the corrected image data and coding the compressed image together with the corrected vector data into images having different resolutions; It is characterized by that.

[0008] The image decoding means decodes the moving image data, and the decoded image is an I picture, a P picture,
Indicating the picture type of the B picture
Image identification data and the motion vector data and the pressure
A variable length decoding unit that outputs reduced pixel data ; an inverse quantization unit that outputs an inversely quantized pixel of the compressed pixel data; a compression unit that decompresses the inversely quantized pixel and outputs the reproduced image data And an inverse orthogonal transform unit.

An image memory unit for storing the reproduced image data; a thinning process of the image data read from the image memory unit being performed by activating a first control signal; a pixel sampling unit for outputting; which pixels by the image identification data
It is determined whether the image is a true type image. If the image is of a specific type, the processing is not performed. If the image is of another type, the first control signal and the second control signal are output. An image determining unit that performs a thinning process on the motion vector data by the second control signal.
A vector correction unit that outputs the correction vector data calculated by the motion vector correction formula .

An orthogonal transform unit for performing image compression processing on the corrected image data and outputting compressed pixel data; a quantization unit for quantizing the compressed pixel data and outputting quantized pixel data; A variable-length encoding unit that encodes the quantized pixel data together with the correction vector data into images having different resolutions.

[0011] The inverse orthogonal transform unit is an inverse discrete cosine transform.

[0012] The orthogonal transform unit is a discrete cosine transform.

[0013]

[0014]

Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing one embodiment of a moving picture transcoder according to the present invention.

In the embodiment shown in FIG. 1, a decoder 1 for decoding an MPEG2 bit stream 21 which is compressed moving image data, a transcoder 2 for converting a data rate of the decoded pixel data 23, and The corrected image data 29 is coded and the videophone code H.264 is used. 26
And an encoder 3 that outputs a 3-bit stream 30.

The decoder 1 decodes the data of the input MPEG2 bit stream 21 and separates the pixel data 14, the motion vector data 22, and the picture type data 24 in the frequency domain into a variable length decoding unit 4; An inverse quantization unit 5 for inversely quantizing the pixel data 14;
An inverse DCT unit 6 for converting the inversely quantized frequency domain pixel data 15 into spatial domain pixel data 23;

Here, MPEG means Moving Pi.
This is an abbreviation of “cure Experts Group”, which is a standard adopted by the committee for transmitting and storing moving images. The image data encoded by MPEG is
MPEG2 bit stream and MPEG
It is classified into one bit stream. A transcoder is data having a predetermined bit rate (here, M
The function of converting a PEG2 bit stream into data having different bit rates (here, an H.263 bit stream) is shown. DCT is Discr.
et Cosine Transformation
This is an abbreviation of an orthogonal compression method. The image data before the DCT processing is referred to as frequency domain data, and the image data after the DCT processing is referred to as spatial domain data. Generally, when orthogonal transform such as DCT is performed, the transform coefficient of the transformed image data is biased to a low-frequency region where the level fluctuation of the image is small, and a transform coefficient of a high-frequency region such as an edge where the shading of the image changes can be ignored. Becomes The transform coefficients have a specific distribution in the screen area,
This is called spatial domain data. Therefore, since image data transmits only non-zero transform coefficients, image compression becomes possible. By the inverse DCT processing, an original image substantially before conversion is reproduced from these conversion coefficients.

The transcoder 2 includes an image memory unit 7 for storing the converted pixel data 23, and a pixel sub-sampling unit 10 for thinning out the pixel data 25 read from the image memory and outputting corrected image data 29.
The picture type data 24 is determined, and if the picture type data is other than the B picture, the picture determination unit 9 that outputs the start instruction signals 27 and 28, and the motion vector data 22 is corrected by the start instruction signal 27 and the correction vector And a vector corrector 8 for outputting data 26. The encoder 3 varies the DCT unit 13 for compressing the corrected image data 29, the quantization unit 12 for quantizing the compressed pixel data 16 obtained by image compression, and the quantized pixel data 17 together with the correction vector data 26. Long encoded H.264 And a variable-length encoding unit 11 that outputs a H.263 bit stream 30.

FIG. 2 is a view for explaining MPEG2 image compression data.

Referring to FIG. 2, compressed image data encoded by MPEG2 includes a sequence layer, GOP
(Group Of Picture) layer, picture layer, slice layer, macroblock layer, and block layer. Picture layer is 704 for NTSC
It is intended for a screen of × 480 pixels. However, PAL
The method targets a screen of 720 × 576 pixels. The slice layer is composed of an arbitrary number of macro blocks (MB). Here, one macroblock is composed of 16 × 16 pixels. Therefore, a screen of 704 × 480 pixels is composed of 44 × 30 macro blocks. The macro block layer is composed of four blocks, and one block is composed of 8 × 8 pixels. As mentioned above, 1
The screen is divided into slices, and one slice is further divided into a macroblock layer and a block layer. In addition, each screen (picture) has a picture type of I, P, and B, and a GOP is formed by the plurality of pictures.
The layers are composed.

Here, the I-picture is an abbreviation of an intra-coded picture (intra-frame coded picture), and is a picture in which all data for one picture frame is generated without performing inter-frame prediction processing. Predi
An abbreviation of an active coded image (inter-frame forward prediction coded image), which is a picture generated by performing an inter-frame prediction process from an I-picture or a previous P-picture, and a B-picture is a bidirectionally-coded picture.
Abbreviation of Predictive coded image (bidirectional predictive coded image), which indicates a picture generated by performing an inter-frame prediction process from a past picture and a future picture.

The MPEG1 bit stream is MP
It has the same structure of six layers as the EG2 bit stream, and there is no difference in the layer structure. However, compared to the MPEG1 bit stream, the MPEG2 bit stream has added functions such as support for various image formats, enhancement of error tolerance, and function division by profile and level.

FIG. FIG. 3 is a diagram illustrating 263 image data.

H. 261 is a video conference by ISDN
A video coding standard developed mainly for videophone use, whereas H.264 is a video coding standard. H.263 is for the GSTN videophone system. 261 is an improved coding method. H. For H.261, motion compensation with half-pixel accuracy has been introduced, variable-length coding has been improved, overhead information such as headers has been reduced, and the prediction efficiency of motion vector information has been improved. is there.

Referring to FIG. 3, in order to compress image information of one screen, the screen is divided into 12 like a picture layer, and further divided into a GOB layer, a macroblock layer, and a block layer. One screen (picture) is CIF
(Common Intermediate Form
At) video format (352 × 288 pixels) is divided into 12 GOBs. One GOB is composed of 33 macroblocks, and one macroblock is composed of an arbitrary number of blocks. H. The screen of H.263 has a standard size of 360.times.288 pixels, but stuffs 8 pixels in the horizontal direction (inserts "0" data to make a black screen) to make 352.times.288 pixels.

Next, the operation will be described in detail with reference to FIGS. 1, 2 and 3.

The picture determining unit 9 determines the picture type based on the picture type data 24. If it is a B picture type, no processing is performed, and if it is an I picture type or a P picture type, the vector corrector 8
, And outputs a start instruction signal 28 to the pixel sub-sampler 10.

Upon receiving the activation instruction signal 27, the vector corrector 8 corrects the motion vector for one screen stored in the image memory unit 7. The equation for correcting the motion vector is shown in equation (1).

Horizontal direction: Horizontal vector after correction = (H.263 horizontal pixel number / MPEG2 horizontal pixel number) × (horizontal vector before correction) Vertical direction: Vertical vector after correction = (H.263 vertical) The number of pixels in the direction / the number of pixels in the MPEG-2 vertical direction × (vertical vector before correction)-(1) A motion vector is calculated for each macroblock according to the equation (1) and correction is performed. Here, as described above, the macroblock is one screen image data of 704 pixels × 480 pixels for NTSC, which is divided into 44 blocks × 30 blocks, where 16 pixels × 16 pixels are one block. Performs the discrete cosine transform (DCT) compression processing on a block-by-block basis.
Indicates a unit.

When spatial resolution conversion is involved, it is necessary to sample the motion vector of each macroblock as in the pixel sub-sampling device 10. The sampling method of the motion vector is the same as the pixel subsampling. That is, the pixel sub-sampling unit 10 performs sampling of pixels by the following method.

Horizontal direction: Horizontal direction thinning interval = MPEG2 horizontal pixel number / H. 263 horizontal pixel count --- (2) Vertical direction: vertical thinning interval = MPEG2 vertical pixel count / H.264. 263 horizontal pixel count --- (3) vertical thinning interval = (MPEG2 vertical pixel count / 2) /H.264. 263 Number of pixels in horizontal direction --- (4) If the scanning method of MPEG2 is progressive, the thinning interval in the horizontal and vertical directions is determined by the formulas (2) and (3). If the scanning method of MPEG2 is interlaced, the vertical direction is reduced by ２ by ignoring either the top field or the bottom field data.
Thin out. Thereafter, the thinning intervals in the horizontal and vertical directions are determined by the equations (2) and (4). The value obtained from each equation is rounded to the nearest whole number to obtain an integer.

Here, the progressive scanning method (non-interlace) is called progressive, and the interlaced scanning method is called interlace. In the interlaced scanning method, there are two fields, that is, a field that scans an odd-numbered scan line and a field that scans an even-numbered line. Fields of odd lines are called top fields, and fields of even lines are called bottom fields.

The sampling of pixels in the horizontal and vertical directions is performed according to the value of the thinning interval. The correction vector data 26 corrected by the vector corrector 8 is output to the variable length encoding unit 11 in the encoder 3. The corrected image data 29 sampled by the pixel sub-sampler 10 is output to the DCT unit 13 of the encoder 3. The DCT unit 13 performs compression processing on the input corrected image data 29 by discrete cosine transform, and outputs compressed pixel data 16. The quantization unit 12 quantizes the compressed pixel data 16 and outputs the quantized pixel data 17. The variable-length encoding unit 11 outputs the converted bit stream 30 by combining the quantized pixel data 17 with the correction vector data 26.

In the vector corrector 8, (H.
When the coefficient value of (H.263 horizontal pixel number / MPEG2 horizontal pixel number) and the coefficient value of (H.263 vertical pixel number / MPEG2 vertical pixel number) are obtained in advance, and the motion vector data 22 is input. , The correction vector data 26 is immediately calculated.

Next, from the MPEG2 bit stream 21 to H.264. The conversion operation when converting to the H.263 bit stream 30 will be described. This conversion operation is roughly divided into two processes, a pixel size conversion process and a motion vector conversion process.

FIG. 4 is a diagram for explaining the conversion of the pixel size. FIG. 4A is a diagram for explaining thinning-out by a picture type, and FIG. 4B is a diagram for explaining vertical and horizontal thinning-out.

The pixel size of MPEG2 is 7 in PAL format.
20 × 576 pixels, H.264. Since the pixel size of H.263 is 240.times.352 pixels obtained by performing stuffing of 8 pixels, it is necessary to thin out MPEG2 pixels in order to match the number of pixels.

As shown in FIG. At 263, since there is no B picture, it is deleted, and the deleted frame is set as a dropped frame. Also, as shown in FIG.
For vertical thinning, 720 × 576 pixels are divided into a top field of 720 × 288 pixels and a bottom field of 720 × 288 pixels, and the bottom field of 720 × 288 pixels is deleted. Further, 720 × 240 is obtained by thinning out the 48 pixels below the top field.
Pixels. Also, as the thinning in the horizontal direction, 720 × 2
By thinning out every other 40-pixel macroblock (MB), a screen of 360 × 240 pixels is obtained.
Perform stuffing of 0 to 8 pixels, 352 ×
H. 240 pixels 263 pixels.

Next, the motion vector conversion processing will be described.

FIG. 5 is a diagram for explaining the motion vector correction. FIG. 5A shows the vector correction of MPEG2, and FIG.
FIG. 263 shows a vector correction.

In FIG. 5A, an arbitrary pixel (indicated by a triangle in the figure) of the previous image of MPEG2 is shifted by H pixels in the horizontal direction and by V pixels in the vertical direction. In this case, the conversion of the horizontal direction vector and the vertical direction vector is performed according to H.264 of FIG. In the image of H.263, the corrected horizontal vector H ′ and the corrected vertical vector V ′ are respectively H ′ = H / 2,
It is converted as V '= V / 2.

As described above, MPEG2 to H.264 are used. When the conversion to H.263 is performed, the luminance signal is obtained as follows when the coefficient value is obtained by the equation (1).

H. 263 horizontal pixels / MPEG2 horizontal pixels = 352/720 = 0.49. H. 263 vertical pixels / MPEG2 vertical pixels = 288/576 = 0.5. Similarly, a thinning interval is obtained for the pixel sub-sampling device 10. If the scanning method of MPEG2 is interlaced, the following values are obtained from equations (2) and (4).

Horizontal thinning interval = MPEG2 horizontal pixel count / H. 263 number of horizontal pixels = 720/352 = 2.05. Vertical thinning interval = (MPEG2 vertical pixel count / 2) / H. 263 pixels in the vertical direction = (576/2) / 288 = 1. By rounding the decimal places, the horizontal thinning interval = 2 and the vertical thinning interval = 1 are obtained.

In the case of interlacing, the vertical thinning is completed by ignoring the top field or the bottom field in the vertical direction. The vector corrector 8 outputs the corrected vector data 26 after the correction to the variable length coding unit 11.

The corrected image data 29 is output to the DCT unit 13 by the pixel sub-sampling unit 10, and
2, through the variable length coding unit 11, It is output as a H.263 bit stream 30.

Next, in the case where the scanning method is an interlaced bit stream, when one frame of data is input to the variable length decoding unit 4, the variable length decoding unit 4 decodes the MPEG2 bit stream 21. , Frequency domain pixel data 14, motion vector data 22, and picture type data 24. The motion vector data 22 is output to the vector corrector 8. Motion vector data 22
Is displayed with 16-bit data in each of the horizontal and vertical directions of one pixel. The picture type data 24 is output to the picture determiner 9. The picture type data 24 is I picture if 8-bit data is "01H (00000001)" in hexadecimal (hexadecimal notation), and if it is "02H (00000010)".
If the P picture is “03H (0000011)”, it is preset to be a B picture.

The pixel data 23 in the spatial region output from the inverse DCT unit 6 is 32-bit data and has a luminance level of -255.
From +255. The pixel data 23 is stored in the image memory 7 for one frame. If the picture type data 24 is other than "03H" of the B picture, the picture determination unit 9 sends the activation instruction signals 27 and 2 to the vector corrector 8 and the pixel sub-sampling unit 10.
8 is output. These start instruction signals 27 and 28 use pulse signals of an arbitrary width.

The vector corrector 8 calculates the motion vector data 2
2 is corrected. If the number of pixels of the horizontal motion vector is 6 and the number of pixels of the vertical motion vector is 5,
From equation (1), the corrected horizontal vector is 3, and the corrected vertical vector is 2.5. This corrected vertical vector is rounded to three after the decimal point.

That is, from the equation (1), the horizontal direction: (360/720) × 06 = 3 (corrected horizontal vector) Vertical direction: (240/576) × 05 = 2.1 (Direction Vector) The pixel sub-sampler 10 performs sampling by ignoring data of a bottom field with respect to vertical sub-sampling. For horizontal sampling, pixel data sampling is performed every other pixel as calculated by equation (1). Similarly, the motion vector is sampled for each macroblock.

The corrected motion vector data 26 of the motion vector is output to the variable length coding unit 11. The corrected image data 29 after the sub-sampling is output to the DCT unit 13. The corrected image data 29 is subjected to image compression by the DCT unit 13 and then by the quantization unit 12 and the variable-length encoding unit 11,
H. It is output as a H.263 bit stream 30.

The MPEG bit stream 21 is M
Instead of the PEG2 bit stream, an MPEG1 bit stream is also possible.

[0054]

As described above, the moving picture transcoder according to the present invention can perform pixel sampling and vector correction inside the transcoder, and thus has the effect of transforming moving pictures having different spatial resolutions. doing.

[Brief description of the drawings]

FIG. 1 is a block diagram showing one embodiment of a moving image transcoder according to the present invention.

FIG. 2 is a diagram illustrating MPEG2 image compression data.

FIG. FIG. 3 is a diagram illustrating 263 image data.

FIG. 4 is a diagram illustrating pixel size conversion.

FIG. 5 is a diagram illustrating motion vector correction.

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 decoder 2 transcoder 3 encoder 4 variable length decoding unit 5 inverse quantization unit 6 inverse DCT unit 7 image memory unit 8 vector corrector 9 picture decision unit 10 pixel subsampling unit 11 variable length encoding unit 12 quantization unit 13 DCT unit 14 Pixel data 15 Pixel data 16 Compressed pixel data 17 Quantized pixel data 21 MPEG2 bit stream 22 Motion vector data 23 Pixel data 24 Picture type data 25 Pixel data 26 Correction vector data 27, 28 Start-up instruction signal 29 Corrected image data 30 H. 263 bit stream

Claims (6)

(57) [Claims] 1. A decoding the moving image data obtained by image compression, this
Is an intra-frame coded image (I-picture),
Inter-frame forward predictive coded image (P-picture), bi-directional
Any picture of the directional prediction coded image (B picture)
Image identification data and motion vector indicating the
Image decoding means for outputting data and compressed pixel data and outputting image data reproduced from the compressed pixel data;
The corrected image data that has been subjected to the thinning processing of the image data is output, and the picture identification data
It is determined whether the image is a
Decimation processing and motion vector processing for motion vector data
Output the correction vector data calculated by the
That an image converting unit; the correction image data and image compression,
Video encoding means for encoding the compressed image together with the correction vector data into images having different resolutions. 2. The image decoding means decodes the moving image data, and the decoded image is an I-picture.
Any of A, P and B pictures
The image identification data indicating the type and the motion vector
A variable length decoding unit that outputs the compressed data and the compressed pixel data ; a dequantizing unit that outputs a dequantized pixel of the compressed pixel data; and decompresses the dequantized pixel.
2. The moving image transcoder according to claim 1 , further comprising: an inverse orthogonal transform unit that outputs the reproduced image data. 3. An image memory unit for storing the reproduced image data, wherein the image conversion unit performs a thinning process of image data read from the image memory unit by activating a first control signal, and A pixel sampling unit for outputting data; judging which picture type image is based on the image identification data
And the determination result by without processing if the image of a particular type, the image determining unit and for outputting a control signal and a second control signal if the image of the first other types; the second the control signal, to said motion vector data
A vector correction unit that outputs the correction vector data calculated by the decimation process and the motion vector correction formula ;
Motion picture code conversion apparatus 請 Motomeko 1 or claim 2, wherein the having the. 4. An orthogonal transformation unit for performing image compression processing on the corrected image data and outputting compressed pixel data; a quantization unit for quantizing the compressed pixel data and outputting quantized pixel data. The moving image code conversion according to claim 1, 2, or 3, further comprising: a variable length coding unit that codes the quantized pixel data together with the correction vector data into images having different resolutions. apparatus. 5. The moving picture code converter according to claim 2 , wherein said inverse orthogonal transform unit is an inverse discrete cosine transform. 6. The moving picture transcoder according to claim 4 , wherein said orthogonal transform unit is a discrete cosine transform.