JP2000165861A - Moving image decoding device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To decrease memory consumption and memory access amount by providing a memory for storing compressed image data outputted from a data compressing means and a data extending means for extending the compressed image data outputted from the memory. SOLUTION: A data compressing means 11 is provided between an adding means 24 and a memory 25, and restored image data S24 outputted from the adding means 24 are compressed and stored in the memory 25. Then, a data extending means 12 is provided between the memory 25 and a motion compensating and interpolating means (IMC) 26, and compressed reference data S25 read out of the memory 25 are restored and outputted to the IMC 26. Thus, since the data compressing means 11 is provided, the code amount (bit amount) of reference image data is compressed and reference images can be stored in a little memories. Therefore, since the memory capacity is reduced, the memory access amount can be reduced.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to encoding and
The present invention relates to an effective reduction of a memory used in a decoding apparatus for encoding and decoding an image.

[0002]

2. Description of the Related Art Conventionally, there has been a technique described in the following literature as a technique in such a field. Reference: Hiroshi Fujiwara, "Latest MPEG Textbook", ASCII Publisher, pp129-pp155.

The above-mentioned document describes an encoding method of MPEG2 which is an international standard of moving image encoding. FIG. 2 is a configuration diagram of a conventional video decoding method. In FIG. 2, the input of a compression-encoded image signal is a variable length decoding
The control information s21-2 such as a variable length decoded motion vector of the VLD 21 is connected to a motion compensation interpolation means (hereinafter abbreviated as IMC) 26, and the VLD 26 is abbreviated.
Is connected to an inverse quantization means (hereinafter abbreviated as IQ) 22 and an output s22 of the IQ 22 is connected to an inverse discrete cosine transform means (hereinafter abbreviated as IDCT) 23. The output s23 of the IDCT 23 is connected to the adding means 24, the output s24 of the adding means is connected to the external terminal and the memory 25, and the output s25 of the memory is
And the output of the IMC is connected to the adding means 24.

[0004] When a compression-encoded image signal such as MPEG2 is input, the VLD 21 decodes the encoded image signal and converts information s21-2 such as a motion vector into an IMC2.
6 and outputs the image data signal s21-1 to the IQ22. The IQ 22 performs inverse quantization on the input image data and outputs an inversely quantized image signal. IDCT2
In step 3, IDCT is performed on the input image signal to restore the image data. The IMC 26 reads reference image data from the position indicated by the motion vector from the memory 25 according to the motion vector information of the input image.
In the adding means, the read reference image data and the ID
Outputs the sum with the image data output by CT. This is the current image in which the sum of the reference image data and the IDCT image data has been restored. In the memory 25, the restored current image is stored and used as a reference image for the next image.

[0005]

However, such a moving image restoration method requires a large memory and a large number of memory accesses to store the reference image. For large images such as HDTV, the burden on memory and memory access is quite large. Therefore, a larger amount of memory and faster memory access are required, resulting in a more expensive device.

[0006]

SUMMARY OF THE INVENTION According to the present invention, there is provided a conventional moving image restoring method and apparatus, wherein image data is compressed by a simple and efficient compressing means before storing the restored image in a memory. By restoring the reference image when reading the reference image from the memory, the amount of memory and the amount of memory access are reduced.

[0007]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiment 1 Hereinafter, a moving image restoration method and apparatus according to Embodiment 1 of the present invention will be described in detail.

<Structure> FIG. 1 is a diagram showing the structure of a moving image restoration apparatus according to a first embodiment of the present invention. The moving image restoring apparatus according to the first embodiment of the present invention is different from the conventional moving image restoring apparatus shown in FIG. 2 in that the data compressing means 11 is provided between the adding means 24 and the memory 25, and the restoration output from the adding means 24 is provided. The image data s24 is compressed and stored in the memory 25, and the data decompression means 12 is stored between the memory 25 and the IMC 26.
And restores the compressed reference image data s25 read from the memory 25 and outputs it to the IMC 26.

<Operation> In FIG. 1, VLD 21 and IQ
22, IDCT 23, adding means 24, and memory 2
5 and the IMC 26 perform the same operation as that of the conventional video decoding device.

As shown in FIG. 3, the data compressing means 11 comprises a block dividing means 31, a data converting means 32 and an encoding means 33. When the restored image data is input from the adding means, the block dividing means 31 divides the image data into m × n blocks and outputs the divided data to the data converting means 32. The data conversion unit 32 performs conversion such as, for example, an Hadamard conversion on the m × n blocks, and outputs the converted data to the encoding unit 33. Two-dimensional Hadamard transform can be realized by combining vertical and horizontal one-dimensional Hadamard transform. Further, Adamaru conversion of 1-dimensional 2 ^K pieces of data can be expressed as (1).

[0011]

(Equation 1)

In the equation (1), [x] represents input image data, [X] represents image data subjected to an Hadamard transform, [H ₂ ^K ] represents an Hadamard transform coefficient, and [H ₂ ^K ] represents an equation (2). Is required.

[0013]

(Equation 2)

The coding means 33 performs adaptive fixed code length coding on the input converted data. The fixed code length coding is a method of allocating a fixed code amount (the number of bits) to each data in a block and adaptively coding each data according to the value of the data. . FIG. 4 is an explanatory diagram of the operation of the adaptive fixed code length coding method. Hereinafter, the adaptive fixed code length encoding will be described with reference to FIG. First, a fixed code length B and a plurality of thresholds Tn are prepared for data in a block. When there are two thresholds, one bit within the fixed code length is used as an identification bit, and when there are four thresholds, two bits are used as an identification bit. For any given data,
First, it is checked whether or not the absolute value of the data has exceeded a given first threshold T1, and if not, the identification bit is set to, for example, “0” and the data is set to Bt ( (If the identification bit is t bits). If so, a check is made as to whether a value obtained by subtracting the first threshold from the absolute value of the data has exceeded the second threshold, and if the value has not exceeded the second threshold, the identification bit is set to " 1 ", and similarly encoded.
Repeat this work. If the data exceeds the last threshold value Tn, the data is clipped and encoded so that the data does not exceed the threshold value (absolute value = Tn-1).
As an encoding method, for example, for the corresponding threshold value Tk, the quantization coefficient Qk is calculated so that the range from -Tk to + Tk is equally divided by 2 ^Bt. Each
There is a method of indexing from 0 to 2 ^Bt -1 and outputting an index in a certain range to data included in the range.

As shown in FIG. 5, the data decompression means 12 is composed of a decoding means 41 for decoding image data which has been divided into blocks and encoded, a data inversion means 42 and a block integration means 43. When the compressed image data input from the memory 25 is input, the decoding means 41 performs the reverse operation of the coding means 33 and decodes the coded data. First, the identification bit of the encoded data is used to check the range of the threshold value of the data, and the index bit of the data is used to check which range of the data is within the threshold value. . For example, the data is Xk and Xk + 1 (Xk <Xk
If the value falls within the range of +1), the data is output as Xk + Zk, and the sum of the threshold values subtracted by the encoding means is output. Here, Zk is a (Xk + 1-Xk) (0 <=
a <= 1 arbitrary fixed value).

In the data conversion means 42, the decrypted m
The Hadamard inverse transform is performed on the xn block data, and the data after the inverse transform is output to the block integrated encoding 43. The two-dimensional inverse Hadamard transform can also be realized by combining the one-dimensional vertical and horizontal inverse Hadamard transform, like the Adamal transform. Further, Adamaru inverse transform of a one-dimensional 2 ^K pieces of data can be expressed as (3).

[0017]

(Equation 3)

Although the Hadamard transform is used here, other transforms such as DCT, Wavelet, and DFT (Fourier transform) can be used. However, since these arithmetic processes are large in volume, the processing load can be further reduced by using the Adamal transform that can be realized only by addition and subtraction.

The block integrating means 43 integrates the inversely transformed mxn block data, and returns the data divided by the block dividing means to the original structure.

<Effects> As described above in detail, according to the first embodiment of the present invention, by providing the data compression means, the code amount (bit amount) of the reference image data is compressed, and the memory capacity is reduced. Can store the reference image. ,
As the amount of memory decreases, the amount of memory access also decreases. Further, image data can be converted into a frequency band by performing a conversion such as an Hadamard transform, and the coding unit can determine a code amount and a threshold value according to the characteristics of each frequency component, thereby improving efficiency. Encoding can be performed. Furthermore, by performing adaptive fixed code length coding on each frequency component, the code amount of the entire block is fixed, and it is possible to easily calculate which block is contained in the memory and where. For this reason, random accessibility is improved.

In addition to this, it is possible to perform appropriate coding on each data according to the value of the data, and thus it is possible to simultaneously perform coding with high coding efficiency.

In the IMC and the like, since it is necessary to take out arbitrary image data, random accessibility is extremely important. In a flat image, the value of the high-frequency component is small, but the image is also sensitive to noise. Therefore, fine quantization is required for the small high-frequency component. On the other hand, an image that changes drastically has a large high-frequency component, but is insensitive to noise. Therefore, even if quantization is large, deterioration in image quality is hardly perceived visually, so that the code amount can be reduced. . Therefore, efficient encoding can be realized by performing adaptive encoding.

Embodiment 2 <Structure and Operation> In the moving picture decoding apparatus of Embodiment 2 of the present invention, the data compression means 11 of the moving picture decoding apparatus of Embodiment 1 of the present invention uses the luminance component ( Y) and the color difference components (U, V) are Y: U: V = 4: 1: 1, and the color difference components U and V
Halves both vertically and horizontally compared to the luminance component,
The ratio of the luminance component (Y) and the color difference components (U, V) of a color image is Y:
In the case of U: V = 4: 2: 2, the color difference components U and V are divided into blocks so that each becomes only half the horizontal compared to the luminance component,
Conversion and encoding are performed such that the sum of the allocated code amounts (the number of bits) of the luminance component block and the corresponding chrominance component block becomes constant. In particular, when the data bit width of the memory is fixed, the sum of the code amounts is set to be a multiple of the data bit width of the memory. For example, when the data bit width is 8 bits, the sum of the code amounts of the luminance component block and the chrominance component block is a multiple of 8, and when the data bit width is 16 or 32 bits, the luminance component The sum of the code amounts of the block and the block of the color difference component is set to be a multiple of 16 or 32, respectively.

<Effect> The luminance component and the chrominance component are treated in a color image in an integrated manner, and by performing block division and encoding, the amount of memory access can be reduced and the encoding efficiency can be increased. it can. For example, if the size of the color difference component block is the same as that of the luminance component block,
The random access property of the color difference component in the IMC is reduced. Since the luminance component and the chrominance component have different component ratios, when only some pixels in the block are required, the reading of pixel data (overhead) that does not require the chrominance component increases compared to the luminance component. This is caused by

On the other hand, when the color difference component block is reduced,
Since the code amount allocated to the block is limited by the data bit width of the memory, efficient coding becomes difficult.
For example, assuming that the block size is 2 × 2 and the data bit width of the memory is 32 bits, even if a code amount of 32 bits or less is allocated to the block, 32 bits are read by one access. Is wasted. Therefore, as in the specific example 2, by treating the luminance component and the color difference component in an integrated manner, for example, the luminance component is
If a 4 × 4 block, a chrominance component is a 2 × 2 block, and the data bit width of the memory is 32 bits, the total code amount of the luminance component and the chrominance component may be a multiple of 32 bits. 98 bits in a block,
Each of the color difference components U and V can be assigned like 15 bits, and efficient coding can be realized. Further, by setting the block size of the color difference component as in this specific example, since the luminance component and the color difference component represent the same image area, the access overhead for IMC and image display can be reduced.

<< Embodiment 3 >><Structure and Operation> In the moving picture decoding apparatus of Embodiment 3 of the present invention, the data compression means 11 of the moving picture decoding apparatus of Embodiments 1 and 2 of the present invention sets a threshold value. It is set to be a power of two.

<Effect> By setting the threshold value to a power of 2, it is not necessary to perform division for decoding to obtain an index during encoding and multiplication for decoding from the index to image data during decoding. Since it can be performed, it is useful for simplifying and speeding up the apparatus. For example, when the threshold value is 2 ^k, since the range of the threshold value is from -2 ^k to 2 ^k -1, the data in the threshold value can be expressed by k + 1 bits.
If the data within the threshold is coded with B bits, the data within the threshold is actually shifted to the left (in a direction of decreasing the absolute value) by (k + 1-B) bits, and B bits are shifted from the lower order. You only have to take it out. The extracted B-bit code corresponds to the index. The decoding means can decode the code by shifting the code for the B bits to the right by (k + 1-B) bits and adding the bias Zk.

Embodiment 4 <Structure and Operation> The moving picture decoding apparatus according to Embodiment 4 of the present invention uses the data decompression means 12 of the moving picture decoding apparatus according to Embodiment 3 of the present invention to was after 2 ^m times to add a bias Zk, and restore it with 1/2 ^m times after inverse transformation.

<Effect> As described in the first embodiment, in order to reduce the quantization noise, the decoding means needs to add a bias so that the decoded data becomes the center of the probability distribution within the range of the data. There is. On the other hand, a hardware device generally performs an integer operation. In this case, in the encoding, for example, when only one bit is shifted to the left, that is, when the quantization coefficient is 2, the decoding means shifts one bit to the right. Therefore, the value that the bias can take is only 0 or 1. Setting the bias to 0 means that the image data is divided by the quantization coefficient 2 and rounded down, and setting the bias to 1 means that the image data is divided by the quantization coefficient 2 and rounded up. I do. Each is the end of the probability distribution of the data. However, by multiplying the image data by 2 ^m at the decoding stage, even if only one bit is shifted to the left in the encoding, m + 1
By shifting to the right of the bit, the value that the bias can take can be any value from 0 to 2 ^{m + 1} -1 and can be closer to the center value of the probability distribution of the data. Therefore, the quantization error is reduced, and the image quality is improved.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a video decoding device according to Embodiment 1 of the present invention.

FIG. 2 is a configuration diagram of a conventional video decoding device.

FIG. 3 is a configuration diagram of a data compression unit according to the present invention.

FIG. 4 is an explanatory diagram of adaptive fixed code length encoding according to the present invention.

FIG. 5 is a configuration diagram of data decompression means according to the present invention.

[Explanation of symbols]

 11, data compression means 12, data expansion means 21, variable length decoding means (VLD) 22, inverse quantization means (IQ) 23, discrete cosine inverse transformation means (IDCT) 24, addition means 25, memory 26, motion compensation interpolation Means (IMC) 31, block division means 32, data conversion means 33, encoding means 41, decoding means 42, data inverse conversion means 43, block integration means

 ──────────────────────────────────────────────────続 き Continued on the front page F term (reference) 5C053 GA11 GB07 GB19 GB21 GB22 GB25 GB26 GB32 GB40 KA01 KA22 5C059 KK02 KK08 KK15 LC08 MA22 MC11 ME01 ME13 PP04 PP14 TA11 TA16 TA60 TB15 TC18 TD12 UA02 UA05 UA31 5C044 BA53A BA57 CA02 CA22 CA27 DA01 DA02 9A001 BB02 BB03 BB04 DD08 EE02 EE04 EE05 FF01 HH25 HH27 HH30

Claims (10)

[Claims] A decompression unit for decompressing the compression-encoded image data; a data compression unit for compressing the decompressed image data output from the decompression unit; and a compressed data output from the data compression unit. A moving picture decoding apparatus comprising: a memory for storing image data; and a data decompression unit for decompressing compressed image data output from the memory. 2. The data compression unit includes: a block division unit configured to divide the restored image data into blocks; and a data conversion unit configured to perform data conversion independently on each of the blocks divided by the block division unit. The moving picture decoding apparatus according to claim 1, further comprising: coding means for coding data converted by the data converting means. 3. The data decompression means: decoding means for decoding the compressed image data output from the data compression means; data inverse conversion means for performing an inverse conversion on the image data decoded by the decoding means; The moving picture decoding apparatus according to claim 1, further comprising: a block integrating unit that integrates and restores the data block inversely transformed by the data inverse transforming unit. 4. The moving picture decoding apparatus according to claim 2, wherein said data conversion means and data inverse conversion means use orthogonal transform and orthogonal inverse transform. 5. The moving picture decoding apparatus according to claim 2, wherein said data conversion means and data inverse conversion means use an Hadamard transform and an inverse transform. 6. The encoding unit according to claim 1, wherein the image data is divided into a plurality of classes using a threshold value, and the image data of each class is subjected to different quantization to perform encoding of a fixed code length. 6. The moving picture decoding apparatus according to claim 2, wherein: 7. The moving picture decoding method and apparatus according to claim 6, wherein said encoding means sets a threshold value to a power of 2, and performs a shift operation of quantization and inverse quantization. . 8. The image processing apparatus according to claim 1, wherein the decoding means multiplies the decoded image data by a power of 2, and the image data after dequantization takes a value near the center of the probability distribution even in an integer operation. Item 8. The video decoding device according to any one of Items 2 to 7. 9. The encoding unit assigns different code amounts to individual components in each block divided by the block division unit, and the total code amount of the block is a data bit width of the memory. The moving picture decoding apparatus according to claim 2, wherein the video decoding apparatus is set so as to be an integral multiple of the moving picture decoding apparatus. 10. The block dividing means divides a chrominance component block of a color image so as to indicate the same image region as the luminance component block, and divides the chrominance component block and the chrominance component block indicating the same image region. 10. The moving image according to claim 2, wherein a block obtained by integrating the two is set as one encoded block, and the total code amount of the integrated block is set to be an integral multiple of the data bit width of the memory. Image decoding device.