JP2001128175A - Device and method for converting image information and recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To further reduce the code rate of image compression information in accordance with the band width or storage capacity of a transmission route. SOLUTION: The image information converting device fetches high-bit rate image compression information to an code buffer 1, analyzes the image compression information by means of a compression information analyzer 2, and decodes the information by means of a variable-length decoder 4. Then the device inversely quantizes the decoded information and, after further reducing the code rate of the information by performing band limiting by means of a band limiter 7, again quantizes the information by means of a quantizer 8, again encodes the information by means of a variable-length encoder 11, and outputs the encoded information from another code buffer 10 as image compression information. A target code rate calculator 54 controls the quantizer 8 by setting a target code rate based on the code rate of the header section of each frame stored in a header code rate buffer 52 and the complexity calculated by means of a complexity calculator 53.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus and method for converting image information, and a recording medium, and more particularly to an MPEG (Moving) method.
When transmitting image compression information (bit stream) compressed by orthogonal transform such as discrete cosine transform through a network medium such as satellite broadcasting, cable television, or the Internet, such as a Picture Experts Group, or an optical disc The present invention relates to an image information conversion apparatus and method for further compressing and reducing the amount of code of data when processing on a storage medium such as a magnetic disk, and a recording medium.

[0002]

2. Description of the Related Art In recent years, image information is handled as digital data. At that time, the purpose of transmitting and storing information with high efficiency is to utilize orthogonality such as discrete cosine transform and motion by utilizing redundancy inherent in image information. 2. Description of the Related Art Devices that comply with a scheme such as MPEG that compresses image information by compensation are becoming widespread in both information distribution at broadcast stations and information reception in ordinary households.

[0003] In particular, MPEG-2 (ISO / IEC13818-2) is defined as a general-purpose image encoding method, and is a standard covering both interlaced scan images and progressive scan images, as well as standard resolution images and high definition images. It is expected to be used in a wide range of professional and consumer applications.

However, in order to achieve good image quality even if the MPEG-2 compression method is used, for example, a standard resolution interlaced scan image having 720 × 480 pixels is 4 to 8 Mbps and 1920 × 1080. In the case of a high-resolution interlaced scanning image having pixels, it is necessary to allocate a code amount (bit rate) of 18 to 22 Mbps as image information.

However, when transmitting such a compressed image having an enormous amount of code as network media such as satellite broadcasting and cable television, the deterioration of image quality is minimized in accordance with the bandwidth of the transmission path. It is necessary to further compress the image information while suppressing it. Further, when processing such a compressed image having a huge code amount on a storage medium such as an optical disk or a magnetic disk, the code amount is further reduced while minimizing image quality deterioration according to the capacity of the recording medium. It is necessary to perform reduction, that is, image compression.

The necessity of such code amount reduction is not only for high-resolution images but also for standard-resolution images (for example, when the image frame is 72
(30 pixels interlaced scanning image of 0 pixels x 480 pixels)
Is also expected to occur when processing is performed on network media or storage media as described above.

[0007]

As means for solving the above-mentioned problems, hierarchical coding (scalability) and image information conversion (transcoding) can be considered. Regarding the former, MPEG-2 has standardized the SNR (Signal Noise Ratio) scalability, which has
It is possible to hierarchically encode SNR image compression information and low SNR image compression information. However, in order to perform hierarchical coding, it is necessary that the constraint condition of the bandwidth or the storage capacity is known at the time of coding.

However, in an actual system, these constraint conditions are often unknown, so that the latter image information conversion (transcoding) has a higher degree of freedom than the actual system. It can be said that
Therefore, it is desired to realize an image information conversion method and an apparatus for executing the method, which minimize image deterioration and enable further compression of image information according to the bandwidth of a transmission path.

The image information converter basically decodes or partially decodes the input image compression information, further compresses the decoded image information (reduces the code amount), and encodes the decoded image information again. However, there are two possible configurations of passing information (pixel data) from the decoding unit to the encoding unit in the spatial domain or in the frequency domain.

The former has a large amount of arithmetic processing, but can minimize the deterioration of the decoded image of the compressed information to be output, and is mainly used for applications such as broadcasting equipment. On the other hand, the latter causes a slight deterioration in image quality as compared with the former, but can be realized with a smaller amount of processing, and is mainly used for applications in consumer appliances.

However, in any of the configurations, the image information conversion device has a similar configuration to the image information encoding device that compresses image information using MPEG or MPEG-2, but the image information of the original image is input. In contrast to the image information encoding device, the image information conversion device is configured by directly applying the device configuration of the image information encoding device in order to receive the image compression information compressed by the image information encoding device. Can not do it. For example, in an image information encoding device, GOP (Gr.
Although a structure of “up of picture” is given, the image information conversion apparatus cannot predict what type of GOP is input, and has a problem that it cannot be understood unless the input image compression information is analyzed.

Further, in the image information coding apparatus, for example, the document “Bit-rate con
trol for MPEG encoders ”(G. Keesman, I. Shah and RK
lein-Gunnewiek, Signal Processing Image Communicati
on 6, pp. 545-560, 1995), the frame code amount is calculated by the code amount of the discrete cosine transform coefficient part depending on the quantization width,
There has been proposed a method of allocating the code amount separately to the code amount of the header portion which does not depend on the quantization width. However, the configuration of the image information encoding device cannot be directly applied to the image information conversion device. There are also problems.

[0013] An object of the present invention is to minimize the deterioration of image quality and to further reduce the code amount even if the structure such as the GOP length of the input image compression information is unknown. An image information conversion apparatus capable of reducing the amount of code according to the bandwidth and storage capacity of a transmission path, and further capable of further reducing the amount of code without increasing the amount of computation and the circuit scale; and A method and a recording medium are provided.

[0014]

An image information conversion apparatus according to claim 1 takes in image compression information encoded by orthogonal transformation and motion compensation, further reduces the code amount, and outputs the image information. In the apparatus, detecting means for detecting the amount of code assigned to the header portion of the image compression information,
A first calculator for calculating a code amount to be assigned to the orthogonal transform coefficient of the image compression information based on the code amount assigned to the header portion detected by the detector, and a first calculator that calculates the code amount. A second calculating means for calculating a target code amount based on the code amount allocated to the orthogonal transform coefficient; and a code for requantizing the image compression information based on the target code amount calculated by the second calculating means. Control means for controlling the amount.

The orthogonal transform may be a discrete cosine transform.

When the target code amount for the header portion is T _header and the target code amount for the orthogonal transform coefficient is _Tcoef , the second arithmetic means sets the target code amount T for each frame of the image compression information to T _header.

(Equation 9) It can be calculated as

[0017] The second computing means, a code amount of the header portion which is detected by the detection means may be a T _{he ADER.}

A calculating means for calculating a complexity for each frame from an average quantization scale of the captured image compression information for each frame and a code amount allocated to an orthogonal transform coefficient of each frame; The second calculating means can determine the target code amount T _coef based on the complexity calculated by the calculating means.

[0019] The information processing apparatus may further include determining means for determining a pseudo GOP structure from an interval between I pictures in the image information obtained by decoding the image compression information.

The first computing means calculates N as a pseudo GOP
, The number of frames included in the frame, bit_rate is the code amount of the image compression information to be output, picture_rate is a value indicating how many frames of the output image are displayed per second, and the first frame of the pseudo GOP is reproduced. When encoding, pseudo G
The allocated code amount R for the uncoded frame in the OP is

(Equation 10) Is updated as a pseudo-code in the input image compression information.
The total code amount R _head for the _header of the frame included in the GOP is

[Equation 11] And the allocated code amount R _coef for the orthogonal transform coefficient of the _uncoded frame in the pseudo GOP is calculated as

(Equation 12) It can be calculated as

After the respective frames are re-encoded, the first arithmetic means determines the allocated code amount R _coef when the amount allocated to the orthogonal transform coefficient is S _coef among the generated code amounts of the frame. To

(Equation 13) Can be updated as

[0022] The second operation means is provided with an allocated code amount R _coef.
Is used to calculate the target code amount T _coef for each frame.

[0023] The second calculating means is configured to calculate a target code amount T _coef.
Can be calculated using the values of the parameters Kp and Kb according to the image.

The second calculating means adaptively calculates the value of the parameter using the complexity of each frame of the input image compression information.

The second calculating means calculates the complexity of each frame by calculating the average quantization scale of each frame in the input image compression information and the code amount allocated to the orthogonal transform coefficient within the frame. The calculation can be performed by using this.

When Xi, Xp, and Xb are the complexities of the I picture, P picture, and B picture constituting the pseudo GOP, respectively, and 1 / (1 + m) is a predetermined value, the second calculating means , Parameters Kp and Kb

[Equation 14]

(Equation 15) Can be calculated by

When the average quantized scale code in the frame of the input image compression information is Q, and the amount of the total code amount allocated to the orthogonal transform coefficient is S _coef , the second calculating means City X

(Equation 16) Can be calculated by

[0028] The second arithmetic means may set the value of 1 / (1 + m) to 0.6 to 1.2.

[0029] The second arithmetic means may set the value of 1 / (1 + m) to 1.0.

The second calculating means may execute the exponent calculation for obtaining the parameters Kp and Kb by referring to a preset table.

The second operation means includes a parameter Kp, which gives a virtual buffer initial value at the beginning of the sequence.
As values of Kb, Kp = 1.0, Kb = 1. A value of 4 can be used.

The image information conversion method according to claim 19 is
In an image information conversion method of an image information conversion device that takes in image compression information encoded by orthogonal transformation and motion compensation, further reduces the amount of code, and outputs the code amount, the code amount allocated to the header portion of the image compression information is A detecting step of detecting, a first calculating step of calculating a code amount to be assigned to the orthogonal transform coefficient of the image compression information based on the code amount assigned to the header portion detected by the processing of the detecting step; A second calculation step of calculating a target code amount based on the code amount to be assigned to the orthogonal transform coefficient, calculated by the processing of the calculation step, and a target code amount calculated by the processing of the second calculation step. hand,
And a control step of controlling a code amount at the time of requantization of the image compression information.

According to a twentieth aspect of the present invention, there is provided an image information conversion apparatus which fetches image compression information coded by orthogonal transformation and motion compensation, further reduces the code amount, and outputs the image compression information. In the controlling program, a detection step of detecting a code amount allocated to a header portion of the image compression information, and an orthogonalization of the image compression information based on the code amount allocated to the header portion detected by the processing of the detection step. A first calculation step of calculating a code amount to be assigned to a transform coefficient, and a second calculation step of calculating a target code amount based on the code amount to be assigned to an orthogonal transform coefficient, which is calculated by the processing of the first calculation step Controlling the code amount at the time of requantization of the image compression information based on the target code amount calculated by the processing of the second calculation step Characterized in that it comprises a step.

[0034] In the image information conversion device according to the first aspect, the image information conversion method according to the nineteenth aspect, and the program recorded on the recording medium according to the twentieth aspect, the code assigned to the header portion is used. A code amount to be assigned to the orthogonal transform coefficient of the image compression information is calculated based on the amount.

[0035]

DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram of an image information conversion device (transcoder) according to the present invention. The image information conversion apparatus according to this embodiment is configured to transfer information (pixel data) from the decoding unit to the encoding unit in the frequency domain, and to perform encoding of high-bit-rate image compression information. A buffer 1, a compression information analyzer 2 connected to an output of the code buffer 1, and an information buffer 3 for storing data such as a quantized value in high bit rate image compression information analyzed by the compression information analyzer 2. A variable-length decoding device 4 that takes in the high-bit-rate image compression information that has passed through the compression-information analysis device 2 and performs variable-length decoding, and an inverse quantization device 5 that inversely quantizes the variable-length-decoded image data. Prepare.

Further, the image information conversion apparatus includes an adder 6 for calculating a difference between an output of the inverse quantization apparatus 5 and a motion compensation error correction value described later, and a high bit rate output from the adder 6. Band limiting device 7 for reducing the code amount of image information
A quantizing device 8 for quantizing image data output from the band limiting device 7, a variable-length encoding device 11 for variable-length encoding image data output from the quantizing device 8, and a variable-length code Code buffer 10 which takes in the low bit rate image compression information output from the quantization device 11 and outputs it to the outside, and the code amount control which controls the quantization device 8 based on the state of the code buffer 10 and the information stored in the information buffer 3. Device 9.

The image information conversion device shown in FIG. 1 is provided with a motion compensation error correction device 12 in addition to the above configuration. The motion compensation error correction device 12 takes in the output data of the quantization device 8 and performs inverse quantization on the inverse quantization device 13.
And an adder 14 for calculating the difference between the output of the inverse quantizer 13 and the output of the adder 6 (adding data of the opposite polarity of the output of the adder 6), and inverting the output of the adder 14. An inverse discrete cosine transform device 15 for performing discrete cosine transform, a video memory 16 for storing output data of the inverse discrete cosine transform device 15, a motion compensation prediction device 17 for performing motion compensation prediction based on the contents of the video memory 16, Discrete cosine transform device 1 that performs discrete cosine transform on the output of compensation prediction device 17 to generate a motion error correction value, and outputs it to adder 6
8 is provided. If it is desired to further reduce the circuit scale even if some image quality degradation is allowed, the motion compensation error correction device 12 can be omitted.

In the following, first, the motion compensation error correction device 1
The principle of operation of the image information conversion apparatus in which step 2 is omitted will be described.

Image compression information (bit stream) having a large code amount (high bit rate) is first input to the code buffer 1 and stored. This image compression information is
VBV (Video Buffering Verifier) specified by EG-2
Therefore, no overflow or underflow occurs in the code buffer 1.

Next, the image compression information stored in the code buffer 1 is sent to the compression information analyzer 2. The compression information analysis device 2 extracts necessary information from the image compression information according to a syntax (syntax) defined by MPEG-2. The image information conversion device according to the present embodiment executes the following re-encoding processing according to the extracted information. In particular,
As described later, pi required for the operation of the code amount control device 9 is
Information such as cture_coding_type and a quantization value (quantization scale: q_scale) for each macroblock is stored in the information buffer 3.

The variable length decoding device 4 first decodes the data encoded as the difference value from the adjacent block with respect to the DC component of the inter macroblock, and executes the run (with respect to the continuous 0 coefficient) for the other coefficients. ) And the level (non-zero coefficient) is subjected to variable length decoding to obtain quantized one-dimensional discrete cosine transform coefficients.

Next, the variable length decoding device 4 rearranges the quantized discrete cosine transform coefficients as two-dimensional data based on the information on the scanning method of the image extracted by the compression information analysis device 2. MPEG-2 has two scanning methods (zigzag scanning method and alternate scanning method).
FIG. 2 shows a list of the scan order of the zigzag scan method, and FIG. 3 shows a list of the scan order of the alternate scan method.

The quantized discrete cosine transform coefficients rearranged into two-dimensional data are inversely quantized by the inverse quantization device 5 based on the information on the quantization width and the quantization matrix extracted by the compression information analysis device 2. Quantization is performed.

With respect to the discrete cosine transform coefficient obtained as the output of the inverse quantization device 5, the band limiting device 7 reduces the high-frequency component coefficient in the horizontal direction for each block. 4A and 4B show an example of processing in the band limiting device 7. FIG.

In the example of FIG. 4, the luminance signal is
As shown in FIG. 4A, of the 8 × 8 discrete cosine transform coefficients, only the values of 8 × 6 coefficients which are low-frequency components in the horizontal direction are stored, and the remaining values are “0” (FIG. 4). (In the example of (A), a white circle). As for the color difference signal, as shown in FIG. 4B, of the 8 × 8 discrete cosine transform coefficients,
The values of only the 8 × 4 coefficients, which are horizontal low-frequency components, are stored, and the rest are replaced with “0” (open circles in the example of FIG. 4B).

When the image compression information input to the code buffer 1 is that of an interlaced scan image, in the frame discrete cosine transform mode, information relating to the time difference between fields includes the vertical high-frequency component of the discrete cosine transform coefficient. Is included, and performing the restriction leads to a significant deterioration in the image quality. Therefore, the band restriction in the vertical direction is not performed. In addition, as shown in this example, image quality deterioration is reduced by performing a greater band limitation on a color difference signal that is less noticeable to human eyes than a luminance signal that is more noticeable to human eyes. Requantization distortion can be reduced while minimizing it.

The processing in the band limiting device 7 is shown in FIG.
It is not limited to those shown in (A) and (B). For example,
A similar effect can be obtained by multiplying the high-frequency component in the horizontal direction of the discrete cosine transform coefficient by a weight coefficient prepared in advance instead of replacing it with “0”.

The 8 × 8 discrete cosine transform coefficients output from the band limiting device 7 are quantized by the quantizing device 8. The quantization width used at that time is determined by the code amount control device 9.

Before explaining the operation principle of the code amount control device 9, first, for comparison with the image information conversion device according to the present embodiment, MPEG-2 Test Mode 15 (ISO / IEC JTC1 / S
MPE to which the method used in C29 / WG11 N0400) is applied
The G-2 image information encoding device will be described.

In the MPEG-2 image information encoding apparatus, the following first
The control by the step, the second step, and the third step is performed. In the first step, the allocated bit amount is allocated to each picture in a GOP (Group of Picture). This allocation is performed based on the bit amount allocated to a picture that has not been encoded in the GOP including the allocation target picture.

In the second step, in order to make the amount of bits allocated to each picture obtained in the first step coincide with the actual code amount, based on the capacities of three types of virtual buffers independently set for each picture. A quantized scale code is obtained by feedback control on a macroblock basis.

In the third step, the quantized scale code obtained in the second step is finely quantized in a flat portion where deterioration is visually conspicuous, and is coarser in a complicated portion of a pattern where deterioration is relatively inconspicuous. As with quantization, it is changed by the activity of each macroblock. In a practically used MPEG-2 image information encoding apparatus, the code amount is controlled by an algorithm according to the method defined in Test Model 5.

However, there are two problems in applying the above-described method applied to the image information encoding device to the encoding unit of the image information converting device as shown in FIG. 1 as it is. The first problem is related to the above-mentioned first step. That is, in the MPEG-2 image information encoding device, since the GOP structure is given in advance, the operation of the first step can be performed based on the GOP structure, whereas the image information shown in FIG. In the conversion device, GO
The P structure is not given in advance, and the GOP structure becomes clear only by analyzing all the information for one GOP in the input image compression information.

Further, the length of a GOP is not always fixed, and a practically used MPEG-2 image information encoding apparatus detects a scene change and adapts a GO according to the scene change.
Since the length of P is controlled in the image compression information, there is an image information conversion device that receives the image compression information compressed by the image information encoding device as an input. Can not handle. .

The second problem is related to the above-mentioned third step. That is, in the MPEG-2 image information encoding device, the activity is calculated using the luminance signal pixel value of the original image. However, in the image information conversion apparatus shown in FIG. 1, image compression information based on MPEG-2 is input, and the luminance signal pixel value of the original image cannot be known.

Therefore, the image information conversion apparatus according to the present embodiment operates according to the procedure shown in FIG. 5 in order to solve the above-described first and second problems. First, in step S1, a pseudo GOP is determined as follows, and code amount control is performed based on the pseudo GOP. The pseudo GOP is a pseudo GOP composed of one I picture and a plurality of P pictures and B pictures. Its length is variable,
It depends on how an I picture is detected in the input image compression information.

FIG. 6 is a diagram for explaining how a pseudo GOP is determined. Picture_co as shown in FIG.
The information buffer 3 of FIG. 1 has a circular buffer for storing ding_type. The circular buffer has a capacity to store 256 picture_coding_types equal to the maximum number of frames that can be included in one GOP defined by MPEG. Each element of the circular buffer stores an initial value in advance.

In FIG. 6, the information of each frame included in the input image compression information is P, B, B, I, B, B
Up to the next P picture. At this time, first, picture_coding_type for several frames is pre-read by the feedforward buffer included in the compression information analyzer 2 (FIG. 1), and the elements of the circular buffer are updated. The size of the feedforward buffer is arbitrary, but is 6 frames in the example of FIG.

Next, as shown in FIG. 6, a pointer a pointing to the current I picture and the next I picture (in this case, this I picture is The length of the pseudo GOP is determined by referring to the pointer b indicating the pseudo GOP.

Finally, a pointer d pointing to the last frame of the feedforward buffer and a pseudo
A pointer c indicating the first frame is determined from the length of the GOP, and the configuration of the pseudo GOP is determined in the code amount control device 9 as shown in FIG. 6 (step S1 in FIG. 5).

The structure of the pseudo GOP determined in this way is as follows: ｛B ₁ , B ₂ , P ₁ , B ₃ , B ₄ , I ₁ , B ₅ , B ₆ ,.
If P _L , B _M-1 , B _M様, the size of the pseudo GOP L_pgop is

[Equation 17] Given by At this time, the code amount control device 9 sets the target code amounts (target bits) Ti, Tp, and Tb of the I picture, the P picture, and the B picture, respectively.

(Equation 18)

[Equation 19]

(Equation 20) (Step S2 in FIG. 5). Where Ω is
In the pseudo GOP, it indicates the frame to be processed, the frame rate is F, the code amount (bit rate) of the output image compression information is B, and the generated code amount is gene.
If rated_bit, R in (Equation 18) to (Equation 20)
Is

(Equation 21)

(Equation 22) It is. In (Equation 22), Θ indicates a frame that has already been processed in the pseudo GOP. Also, X
X (•) such as (I) and X (P) is a parameter (global complexity measure) representing the complexity of each frame, and is pre-analyzed (pre-parsed) by the compression information analyzer 2.
, The total code amount (number of bits) S and the average quantization scale code Q of the frame are calculated in advance,

(Equation 23) Is required. The values of the parameters Kp and Kb are
The initial value of the virtual buffer at the beginning of the sequence is given, whereby the target code amount is controlled by (Expression 18) to (Expression 20) in the code amount control device 9 (Step S3 in FIG. 5). The values of the parameters Kp and Kb are the ratios of the quantized scale codes of the P-picture and the B-picture based on the quantized scale code of the I-picture as defined in MPEG-2 Test Mode 15, respectively.
It is assumed that the overall image quality is always optimized when Kp = 1.0 and Kb = 1.4.

The second problem described above is solved as follows. The quantization scale Q of each macroblock in the image compression information input to the code buffer 1 is calculated at the time of encoding by using the luminance signal pixel value of the original image. Therefore, when the compression information analysis device 2 performs preparsing, first, the quantization scale Q and the code amount (the number of bits) B of each macroblock in the frame are extracted and stored in the information buffer 3, and The average value E (Q), E (B) of the whole Q and B or the average value E (QB) of the product is calculated in advance, and the value is also stored in the information buffer 3.

The code amount control device 9 calculates the normalized activity N_act based on the information such as Q and B stored in the information buffer 3 and the average value thereof by one of the following equations (FIG. 5). Step S4).

(Equation 24)

(Equation 25)

(Equation 26)

Of these, (Equation 25) and (Equation 26) are equivalent processing. When the image quality is evaluated by SNR (Signal Noise Ratio), (Equation 24) has higher image quality, but the subjective image quality is better given by (Equation 25) or (Equation 26).

[0066] Incidentally Now, the quantized values in the input image compression information for a macroblock in Q _l, the quantization value for the output image compression information calculated by the code amount control unit 9 becomes Q _2. Since the image information conversion device of FIG. 1 is for reducing the code amount, Q ₁ > Q ₂
In the case of, the macroblock that has been coarsely quantized once has been requantized, resulting in finer quantization.

The distortion due to coarse quantization is not reduced by fine requantization. Further, fine requantization means that more bits are used for this macroblock, which leads to a reduction in the number of bits allocated to other macroblocks and further deterioration of image quality. For this reason, the code amount control device 9
It is determined whether or not Q ₁ > Q ₂ (step S in FIG. 6).
5) If YES, set Q ₂ = Q ₁ .

The discrete cosine transform coefficients quantized by the quantization device 8 are subjected to variable length coding by the variable length coding device 11. At this time, for the DC component of the discrete cosine transform coefficient, the difference is encoded using the DC component coefficient of one block before as a predicted value, and for the other components, a predetermined scanning method (the zigzag scanning method in FIG. Alternatively, after rearranging into two-dimensional array data based on the alternate scan method of FIG.
Variable length coding is performed using a pair of the number of coefficients (run) and a non-zero coefficient (level) as an event. If all subsequent coefficients are “0” in the scan order within the block, EOB
A code called (End of Block) is output, and the variable length coding for the block is terminated.

Assume that the coefficient of a certain block in the image compression information input to the code buffer 1 is as shown in FIG. In FIG. 7, black circles indicate non-zero coefficients, and white circles indicate zero coefficients. If the horizontal high-frequency component of the discrete cosine transform coefficients is reduced as shown in FIG. 4A, the distribution of non-zero coefficients is as shown in FIG. 7B. When this is re-encoded as it is in the zigzag scan of FIG. 2 (the scan order is indicated by a dotted arrow in FIG. 2),
The non-zero coefficient in the last scan order (the symbol z in FIG. 7B)
Indicated by ) Is "50".

Therefore, the scan conversion is performed, and the scan is performed again by the alternate scan in FIG. 3 (the scan order is indicated by a dotted arrow in FIG. 3), so that the scan number of the last non-zero coefficient z is “44”. become. This allows EOB
Signals can be set with a faster scan number than in the case of zigzag scanning, and a finer value can be assigned as the quantization width accordingly, reducing quantization distortion due to requantization. Becomes

Next, the operation principle of the motion compensation error correction device 12 shown in FIG. 1 will be described.

First, the cause of the motion compensation error will be described. The pixel value of the original image and O, and the quantization width of the input image compression information for the pixel value and Q _1, the output compressed image information after re-encoding, the quantization width for the pixel value and Q _2. When decoded with the quantization widths Q _l and Q ₂ ,
The pixel values of the reference image are represented as L (Q ₁ ) and L (Q ₂ ), respectively.

In the MPEG-2 image information coding apparatus, the pixels of the inter macroblock are first expressed by OL (Q ₁ )
Is calculated, and a discrete cosine transform is applied to the difference value to encode the difference value. On the other hand, an MPEG-2 image information decoding device which receives, via a network or the like, output image compression information whose code amount has been further reduced by an image information conversion device in which the motion compensation error correction device 12 shown in FIG. 1 is omitted. When decoding the received compressed image information, the discrete cosine transform coefficient in the compressed image information is regarded as a code obtained by performing discrete cosine transform on OL (Q ₂ ). In the image information conversion device having the configuration shown in FIG. 1, generally, Q ₁ = Q ₂ does not hold. Since such a phenomenon occurs in a P picture and a B picture, an error accompanying motion compensation occurs.

Further, the picture quality deterioration occurring in the P picture is propagated to the succeeding P picture and the B picture which refers to the picture, which leads to further picture quality deterioration. With this principle,
As the position following the GOP accumulates errors due to motion compensation, the image quality deteriorates, and a phenomenon (drift) occurs in which the image quality returns to good at the beginning of the next GOP.

Therefore, the motion compensation error correction device 12 is provided, and its operation is as follows. The quantized discrete cosine transform coefficient output from the quantization device 8 is transmitted to the variable length coding device 11 and also to the inverse quantization device 13, where the quantization width and the quantization matrix are calculated. Inverse quantization is performed based on the information about The difference between the discrete cosine transform coefficient output from the inverse quantizer 13 and the discrete cosine transform coefficient output from the inverse quantizer 5 is calculated in the adder 14, and this output is input to the inverse discrete cosine transform device 15. And subjected to an inverse discrete cosine transform. The output is stored in the video memory 16 as motion compensation error correction information.

The motion compensation prediction mode information (field motion compensation prediction mode or frame motion compensation prediction mode, and forward prediction mode, backward prediction mode, or bidirectional) in the image compression information input to the image information conversion apparatus. Prediction mode) and motion vector information (analyzed and extracted by the compression information analysis device 2)
Is supplied to the motion compensation prediction device 17 based on the motion compensation error information in the video memory 16, and the data generated thereby becomes an error correction value in the spatial domain. By performing discrete cosine transform in the discrete cosine transform device 18 that receives the correction value as an input, an error correction value in the frequency domain can be obtained.

In the inverse discrete cosine transform device 15 and the discrete cosine transform device 18, for example, a document “A fast computa
tional algorithm for the discrete cosine transfor
m ”(IEEE Trans.commun., vol.25.no.9, pp.1004-1009,19
It is possible to apply a fast algorithm as shown in 77).

In the inverse discrete cosine transform device 15 and the discrete cosine transform device 18, the horizontal discrete cosine transform coefficient is replaced with the high-frequency component coefficient “0” in the band limiting device 7. By omitting the inverse discrete cosine transform and the discrete cosine transform for, it is possible to reduce the circuit scale and the amount of arithmetic processing.

Further, since the deterioration of the color difference signal in the image has a special characteristic that it is difficult for the human eye to understand the deterioration of the luminance signal, the above-described motion compensation error correction is applied only to the luminance signal. Thus, the circuit scale and the amount of arithmetic processing can be significantly reduced while keeping the image quality deterioration to a minimum.

The error in the P picture propagates to the B picture, but the error in the B picture does not propagate any further. On the other hand, a B picture includes a bidirectional prediction mode and requires a large amount of calculation processing. Therefore, by performing the motion compensation error correction only on the P picture, the circuit scale and the amount of arithmetic processing can be significantly reduced while keeping the image quality deterioration to a minimum. By not performing the processing on the B picture, the capacity of the video memory 16 can be reduced.

Further, the motion compensation error correction device 1 of FIG.
In No. 2, all components of 8 × 8 discrete cosine transform coefficients are used as error correction components. In particular, the discrete cosine transform mode is a frame discrete cosine transform mode,
If the scanning method of the input image compression information is interlaced scanning, ignoring the error of the high-frequency component in the vertical direction leads to significant image quality degradation, but in the horizontal direction, among the 8th-order discrete cosine transform coefficients, , High-frequency 4 components (0th to 7th
There is almost no degradation in image quality due to ignoring the fourth to seventh order components of the next discrete cosine transform coefficient.

By utilizing this fact, it is possible to reduce the amount of arithmetic processing in the inverse discrete cosine transform unit 15 and the discrete cosine transform unit 18 and the capacity in the video memory 16 while minimizing image quality deterioration.

That is, in the inverse discrete cosine transform unit 15 and the discrete cosine transform unit 18, the normal eighth-order processing (eighth-order inverse discrete cosine transform and 1/2) is performed in the vertical direction.
In the horizontal direction, among the 8th-order discrete cosine transform coefficients, four low-frequency components (0th to 3rd order among 0th to 7th order discrete cosine transform coefficients) are applied. Is performed using only the component (i). As a result, the horizontal resolution of the video memory 16 is halved, and its capacity can be reduced. However, in this case, the motion compensation prediction device 17 needs a motion compensation process with quarter-pixel accuracy. This processing is performed according to the value of the motion vector in the image compression information as shown in FIG.
By performing linear interpolation in units of 1/4 pixel, it is possible to sufficiently suppress image quality deterioration due to a motion compensation error.

The processing in the horizontal direction is as follows.
There are two approaches.

The first method is the inverse discrete cosine transform device 1
5 performs a fourth-order inverse discrete cosine transform on only the low-frequency fourth-order coefficient among the eighth-order discrete cosine transform coefficients, and the discrete cosine transform unit 18 calculates a pixel generated by the motion compensation from the video memory 16. Fourth-order discrete cosine transform is performed in the horizontal direction on the 4 × 8 error correction values of each block in the region, thereby outputting 4 × 8 error correction values in the frequency region. That is.

By using a high-speed algorithm for the fourth-order inverse discrete cosine transform and discrete cosine transform, it is possible to further reduce the processing amount. FIG. 9 shows a known Wang algorithm (literature: Zhong de Wang., “Fast Algorithms for t
he Discrete W Transform and for the Discrete Fouri
er Transform ", IEEE Tr. ASSP-32, No. 4, pp. 803-816, Aug.
FIG. 18 is a diagram illustrating a fourth-order discrete cosine transform / inverse discrete cosine transform process based on 1984). In FIG. 9, data F
(0) to F (3) are input and data f (0) to f (3) are input.
By making (3) an output, an inverse discrete cosine transform is realized, and by inputting data f (0) to f (3) and outputting data F (0) to F (3), a discrete cosine transform is obtained. Conversion is realized.

For example, the data F (0) and F (2) are input to the adders 31 and 34. The adder 31 adds the data F (0) and the data F (2),
Output to The arithmetic unit 32 outputs the output of the adder 31 (F (0)
+ F (2)) is multiplied by a coefficient 1 / √2, and output to the adder 33 and the adder 44.

The adder 34 adds (subtracts) the data F (2) to the data F (0) with the opposite polarity and outputs the result to the computing unit 35. The arithmetic unit 35 outputs the output of the adder 34 (F (0) -F
(2)) is multiplied by a coefficient 1 / √2, and output to the adder 36 and the adder 39.

The adder 33 adds the output of the arithmetic unit 32 and the output of the adder 43 and outputs the result as data f (0).

The adder 36 adds the output of the arithmetic unit 35 and the output of the adder 38 and outputs the result as data f (1).

The data F (3) is supplied to the arithmetic unit 37 and the adder 4
1 is input. The arithmetic unit 37 converts the data F (3) into
The calculation shown in (Equation 27) is performed, and the result is output to the adder 38.

[Equation 27]

The arithmetic unit 40 calculates the data F (1)
The calculation represented by (Equation 28) is performed, and the result is output to the adder 43.

[Equation 28]

The adder 41 calculates the data F (3)
The data F (1) is added (subtracted) with the opposite polarity, and the arithmetic unit 4
Output to 2. The arithmetic unit 42 performs an operation represented by (Equation 29) on the data input from the adder 41, and outputs the data to the adder 38 and the adder 43.

(Equation 29)

The adder 38 adds (subtracts) the output of the arithmetic unit 42 to the output of the arithmetic unit 37 with the opposite polarity (subtraction), and outputs the result to the adders 36 and 39. The adder 39 adds (subtracts) the output of the adder 38 to the output of the arithmetic unit 35 with the opposite polarity (subtracts), and outputs the result as data f (2).

The adder 43 adds the output of the arithmetic unit 40 and the output of the arithmetic unit 42 and outputs the result to the adders 33 and 44. The adder 44 adds (subtracts) the output of the adder 43 to the output of the arithmetic unit 32 with the opposite polarity (subtraction), and outputs the data f.
Output as (3).

In a second method of processing in the horizontal direction, the inverse discrete cosine transform unit 15 performs an 8th-order inverse discrete cosine transform by replacing the high-frequency 4 coefficients with “0”, and performs a thinning process or an averaging process. Then, the error correction values of the four points in the pixel area are output, and the discrete cosine transform unit 18 outputs the four error correction values of the pixel area obtained by the motion compensation of the motion compensation prediction unit 17.
The error correction value of the point is set to 8 points by interpolation processing, and after performing a discrete cosine transform on this point, the low frequency band up to the fourth order is extracted to output error correction values in 4 × 8 frequency domains. . In the processing by each of the inverse discrete cosine transform device 15 and the discrete cosine transform device 18, a matrix equivalent to a series of processes is calculated in advance, and the matrix is directly applied to each input coefficient, thereby further increasing the processing amount. Can be reduced.

When the inverse discrete cosine transform unit 15 replaces the high-frequency four coefficients with “0” and performs the eighth-order inverse discrete cosine transform, the process is a thinning process, and the process is an averaging process. the series of processes equivalent matrix (individually sequentially multiplying matrices a, B, and C, previously all matrix a * B * C where obtained by multiplying) was respectively iD ₄ _ _deci, and iD ₄ _ _ave ,
In each case, a matrix equivalent to a series of processes performed in the discrete cosine transform unit 18 is represented by D ₄ _ _deci , D
_{4 _ave} . The matrices iD ₄ _ _deci and iD _{4 _ave} are shown in FIG.
(A) and (B) show. For D ₄ _ _deci and D ₄ _ _ave ,

[Equation 30]

(Equation 31) Is established. Here, ^t () represents a transposed matrix. In FIG. 10, iD ₈ represents an eighth-order inverse discrete cosine transform coefficient.

Further, it is generally known that the deterioration of the color difference signal is less noticeable to human eyes than the luminance signal. Therefore, the processing amount of the color difference signal can be further reduced. That is, in the inverse discrete cosine transform device 15 and the discrete cosine transform device 18,
As for the error correction component of the color difference signal among the 4 × 8 error correction signals, as shown in FIG. 11, only the low frequency coefficients in the vertical direction (for example, ぱ 4 × 4) are used for correction, and the remaining By replacing the high-frequency coefficient with “0”, the amount of arithmetic processing involved in error correction can be further reduced.

FIGS. 12 to 14 show CCIR (Consultative).
Committee on InternationalRadio) (Current ITU (I
test sequence “Mobile & Calendar”, which is one of the standard images specified for use in image quality evaluation in the nternationnal Telecommunication Union) -R)
Is the code amount (the number of bits) and the quantization width (Q_scale) when encoding is performed with a fixed quantization width without performing the code amount control.
FIG. FIG. 12 is a diagram illustrating the relationship between the code amount and the quantization width when the quantization width in the I picture is changed from “14” to “112”. FIG.
Shows the relationship between the code amount and the quantization width when the quantization width of the reference I picture is fixed at “14” and the quantization width of the P picture is changed from “14” to “112”. FIG. FIG. 14 shows that the quantization width of the reference P picture is fixed to “14” and the quantization width of the B picture is “18”.
FIG. 10 is a diagram showing a relationship between a code amount and a quantization width when changing from “1” to “112”.

In the MPEG-2 Test Model 5 and the image information conversion apparatus shown in FIG. 1, the product of the average quantization scale code and the generated code amount used for coding each picture when performing the code amount control is: The processing is performed on the assumption that the value becomes constant for each picture type unless the screen changes.

However, as can be seen from FIGS. 12 to 14, the amount of generated code in each of the I picture, P picture, and B picture does not depend on a portion depending on the quantization width (a portion near the horizontal in the drawings). It is divided into parts (parts near the vertical in the figure). The former relates to a quantized discrete cosine transform coefficient, and the latter relates to header information such as a motion vector. Although the product of the average quantization scale code used to encode each picture and the code amount (number of bits) allocated to the discrete cosine transform coefficient of the generated code amount is constant, each picture is encoded. Is not always constant.

The above fact indicates that when the quantization width is sufficiently small, the code amount (bit number) assigned to the header information is negligible compared to the code amount (bit number) assigned to the discrete cosine transform coefficient. Therefore, when coding with a lower code amount (low bit rate), the code amount allocated to the header information is ignored. Indicates that the ratio will not be possible. In particular,
Since the B picture includes the bidirectional prediction mode,
For some pictures, the code amount allocated to the motion vector increases to a non-negligible degree, hinders stable code amount control, and may be a cause of buffer failure.

In the MPEG-2 image information encoding apparatus, as a method for solving this problem, the above-mentioned document “Bit-rate
control for MPEG encoders "(G. Keesman, I. Shahand R.
Klein-Gunnewiek, Signal Processing Image Communicat
ion 6, pp. 545-560, 1995), the frame code amount is separated into the code amount of the discrete cosine transform coefficient part that depends on the quantization width and the code amount of the header part that does not depend on the quantization width. Techniques for performing code amount distribution have been proposed.

The code amount distribution of the header section according to this proposal is as follows.
Derived from the pre-analysis by the coding process in the fixed quantizer, the code amount distribution in the discrete cosine transform coefficient part is
Similar to the method described in ode15, the calculation is performed based on the parameter X, but the parameter X is calculated by the product of the generated code amount of the discrete cosine transform coefficient unit and the quantization width.

The proposal of the code amount distribution for the MPEG-2 image information coding apparatus has a problem that the calculation processing amount and the circuit scale required for the coding process in the fixed quantizer are increased, and this problem is solved. Otherwise, it is difficult to apply to the image information conversion device of FIG. The code amount is appropriately distributed while avoiding an increase in the amount of arithmetic processing and the circuit scale.

FIG. 15 is a block diagram of the image information conversion apparatus of the present invention. In this embodiment, the compression information analyzer 2
By performing the above-described code amount distribution using the data stored in the information buffer 3, the increase in the amount of arithmetic processing and the increase in the circuit scale are avoided. Therefore, the configuration of the code amount control device 51 corresponding to the code amount control device 9 of FIG. 1 is different from the embodiment of FIG. The code amount control device 51 includes a header code amount buffer 52 and a Complexi
A ty calculator 53 and a target code amount calculator 54 are provided. Other code buffer 1, compression information analysis device 2, variable length decoding device 4, inverse quantization device 5, adder 6, band limiting device 7, quantization device 8, code buffer 10, variable length coding device 11, and The configuration of the motion compensation error correction device 12 is the same as the embodiment of FIG.

In this image information conversion device, similarly to the image information conversion device of FIG. 1, the compression information analysis device 2 analyzes the syntax of the image compression information (bit stream), and encodes the encoded information necessary for the following processing. The information necessary for the following processing is extracted from the extracted information and stored in the information buffer 3. The variable-length decoding device 4 performs variable-length decoding of the variable-length-coded discrete cosine transform coefficients according to the information (zigzag scan or alternate scan) regarding the scanning method extracted from the compression information analysis device 2.

The inverse quantization device 5 inversely quantizes the discrete cosine transform coefficients according to the information (quantization scale and quantization matrix) relating to the quantization extracted from the compression information analysis device 2. The band limiting device 7 lowers the resolution of the image by setting the high-frequency component coefficient of the discrete cosine transform coefficient output from the inverse quantization device 5 to “0” or multiplying by a weight coefficient. The quantization device 8 re-quantizes the discrete cosine transform coefficients using a quantization width corresponding to the image information amount (target bit rate) of the image compression information (bit stream) to be output. The variable length coding device 11 performs variable length coding of discrete cosine transform coefficients.

As in the code amount control device 9 shown in FIG. 1, the code amount control device 51 provides the compressed image information after the variable length decoding so that the code buffer 10 does not overflow or underflow. The obtained target code amount and the quantization width extracted from the information buffer 3 are controlled.

Further, a header code amount buffer 52, which is a component of the code amount control device 51 according to the present embodiment, stores a code for a header portion assigned to each frame in the input image compression information. Stores information about the quantity.
Further, the Complexity calculator 53 calculates
The complexity of each frame is calculated from the code amount and the quantization scale assigned to the discrete cosine transform coefficient of each frame in the input image compression information stored in the information buffer 3.

The target code amount calculator 54 calculates the code amount for the discrete cosine transform coefficient optimum for each frame in the image compression information to be output, based on the calculation result of each frame calculated by the complexity calculator 53. Calculated, and furthermore, the value (the number of bits) stored in the header code amount buffer 52 is added to this code amount, thereby obtaining a target code amount (target bit) of the frame. The motion compensation error correction device 12 stores the motion compensation error of each macroblock in the video memory 16, and determines the stored error value according to the motion vector extracted from the input image compression information and the information on the prediction mode. Take out and correct the error.

Next, the operation will be described with reference to the flowchart of FIG. Here, it is assumed that R is a code amount allocated to an uncoded frame in the pseudo GOP shown in FIG. At the beginning of the pseudo GOP, the target code amount calculator 54 calculates R as follows in the same manner as defined in MPEG-2 Test Mode 15 (step S2).
1).

(Equation 32)

Here, N is the number of pictures (the number of frames) in the pseudo GOP, bit_rate is the code amount of the output image compression information, and picture_rate is the number of frames of the output image to be displayed per second. Is the value to represent. The initial value of R at the beginning of the sequence is "0".

Next, the target code amount calculator 54 outputs, via the header code amount buffer 52, the code amount T _header detected by the compression information analyzer 2 and assigned to the header portion of each frame in the pseudo GOP. Then, the total R _{head is obtained} by the following equation (step S22).

[Equation 33]

The target code amount calculator 54 determines in step S21
The code amount R _coef assigned to the discrete cosine transform coefficient among the code amount R given by the processing of

[Equation 34] (Step S23).

The complexity calculator 53 calculates (Equation 2)
According to 3), the complexity X for the frame is calculated as follows (step S24).

(Equation 35)

Here, Q is the average quantization scale code in the input image compression information, as in (Equation 23), and S _coef is the value actually assigned to the frame in the input image compression information. Of the total code amount (the number of bits) assigned to the discrete cosine transform coefficient.

The target code amount calculator 54 calculates the target code amount based on the complexity of each frame calculated by (Equation 35).
First, corresponding to (Equation 18) to (Equation 20), the portion of the target code amount (target bit) allocated to the discrete cosine transform coefficient is calculated as follows (step S25).

[Equation 36]

(37)

(38) Here, Kp = 1.0 and Kb = 1.4.

Next, the target code amount calculator 54 uses the information T _header stored in the _header code amount buffer 52 and T _coef calculated by (Equation 36) to ( _Equation 38) to obtain the target code amount. The target code amount T for the frame is calculated as follows, and the quantization device 8 is controlled in accordance with the calculated value (step S26).

[Equation 39]

The target code amount calculator 54 calculates the code amount R _coef
Is the amount S of the generated code amount that is assigned to the discrete cosine transform coefficient.
Assuming that _coef is represented by genered_bit_coef, it is updated by the following equation (step S27).

(Equation 40)

Thereafter, in step S28, it is determined whether or not the end of the process has been instructed.
The process returns to step S24, and the subsequent processes are repeated. If it is determined in step S28 that termination has been instructed, the target code amount calculator 54 terminates the processing.

By the way, in (Equation 36) to (Equation 38), Kp = 1.0 and Kb = 1.4, and fixed values are used. Application to volume control ”(Koto and Ota, IEICE Technical Report, IE-
95, DSP95-10, May 1995), and dynamically controlling Kp and Kb in accordance with the complexity of each frame in the input image compression information, thereby increasing the height. Image quality can be measured.

When using that method, the target code amount calculator 54 calculates Kp and Kb as follows.

(Equation 41)

(Equation 42)

The difference from the above document is that complexity
To calculate X, as shown in (Equation 35), the total code amount (number of bits) S (Equation 2)
Instead of 3), the code amount Scoef assigned to the discrete cosine transform coefficient of the frame is used.

1 / (1+) in (Equation 41) and (Equation 42)
Regarding the value of m), as described in the literature, by setting the value to about 0.6 to 1.2, the code amount allocation to each frame can be optimized. In particular, if the value of 1 / (1 + m) is set to 1.0,
Since the exponential calculation is not required for executing (Equation 41) and (Equation 42), high-speed execution is possible. Also, 1 / (1+
When the value of m) is set to a value other than 1.0, a table is prepared in advance, and an exponent operation is performed with reference to the table, thereby enabling high-speed realization of (Equation 41) and (Equation 42). .

Next, the target code amount calculator 54 calculates (Equation 3)
6) to (Equation 38), the target code amount T _i , _coef , for the discrete cosine transform coefficient of each frame.
T _{p, coef,} T _b, is calculated as follows _coef.

[Equation 43]

[Equation 44]

[Equation 45]

However, the parameters Kp and Kb are also required when the code amount control device 51 gives the initial value of the virtual buffer at the beginning of the sequence. In order to simplify the calculation, the values of Kp and Kb for providing the initial values are Kp = 1.0 and Kb = 1.4.

As described above, the image compression information according to the MPEG-2 method is input to the image information conversion apparatus according to the above-described embodiment. However, as in MPEG-1 and H.263, orthogonal transformation or orthogonal transformation is performed. If the image compression information is encoded by using the motion compensation and the motion compensation, the code amount can be reduced with the same configuration as the image information conversion device according to the embodiment of the present invention.

In each of the above-described embodiments, the image information conversion device has been described. However, these device configurations can be realized only by software, not hardware.
Also, an embodiment has been described in which information (pixel data) is passed from the decoding unit to the encoding unit in the frequency domain. However, the present invention can be applied to an image information conversion device that performs transmission in the spatial domain. .

[0130]

As described above, according to the image information conversion apparatus according to claim 1, the image information conversion method according to claim 19, and the program recorded on the recording medium according to claim 20, The code amount allocated to the orthogonal transform coefficient of the image compression information is calculated based on the code amount allocated to the header part, so that the code amount is appropriately distributed while avoiding an increase in the calculation amount and the circuit scale. It is possible to do.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of an image information conversion device according to the present invention.

FIG. 2 is a diagram illustrating a scan order of a zigzag scan method.

FIG. 3 is a diagram illustrating a scan order of an alternate scan method.

FIG. 4 is a diagram for explaining the operation of the band limiting device of FIG. 1;

FIG. 5 is a flowchart showing an operation procedure in the code amount control device of FIG. 1;

FIG. 6 is a diagram illustrating a process of determining a pseudo GOP structure in the code amount control device of FIG. 1;

FIG. 7 is a diagram for explaining an advantage of performing a scan method conversion by the variable length coding device of FIG. 1;

FIG. 8 is a diagram illustrating an operation principle of the motion compensation prediction apparatus when performing interpolation with １ ／ pixel accuracy by linear interpolation.

FIG. 9 is a block diagram illustrating a configuration of an apparatus that performs fourth-order discrete cosine transform / inverse discrete cosine transform processing based on Wang's high-speed algorithm.

FIG. 10 is a diagram showing a method of calculating a matrix equivalent to processing in the inverse discrete cosine transform device of FIG. 1;

11 is a diagram illustrating a method of further reducing the amount of code for a color difference signal in the inverse discrete cosine transform device and the discrete cosine transform device of FIG. 1;

FIG. 12 is a diagram illustrating a relationship between a code amount and a quantization width when the quantization width in an I picture is changed.

FIG. 13 shows a case where the quantization width of a reference I picture is fixed and P
FIG. 4 is a diagram illustrating a relationship between a code amount and a quantization width when a quantization width of a picture is changed.

FIG. 14 shows a case where the quantization width of a reference P picture is fixed and B
FIG. 4 is a diagram illustrating a relationship between a code amount and a quantization width when a quantization width of a picture is changed.

FIG. 15 is a block diagram illustrating a configuration of an image information conversion device according to the present invention.

FIG. 16 is a flowchart illustrating an operation of the image information conversion device in FIG. 15;

[Explanation of symbols]

REFERENCE SIGNS LIST 1 code buffer, 2 compression information analyzer, 3 information buffer, 4 variable length decoder, 5, 13 inverse quantizer, 6, 14 adder, 7 band limiter, 8 quantizer, 9, 51 code amount control Device, 10 code buffers,
Reference Signs List 11 variable length coding device, 12 motion compensation error correction device, 15 inverse discrete cosine transform device, 16 video memory, 17 motion compensation prediction device, 18 discrete cosine transform device, 52 header code amount buffer, 53 complexity calculator, 54 Target code amount calculator

 ──────────────────────────────────────────────────続 き Continued on the front page F term (reference) 5C059 KK35 MA00 MA01 MA23 MA32 MA33 MC01 MC22 MC24 ME01 NN09 NN28 PP04 RC09 RC28 SS01 SS06 SS11 UA33 5J064 AA02 AA04 BA09 BA16 BC01 BC16 BD02

Claims (20)

[Claims] 1. An image information conversion device which takes in image compression information encoded by orthogonal transformation and motion compensation, further reduces the amount of code, and outputs the reduced amount of code, wherein the amount of code assigned to a header portion of the image compression information is Detecting means for detecting, based on the code amount allocated to the header portion detected by the detecting means, a first calculating means for calculating a code amount to be allocated to the orthogonal transform coefficient of the image compression information; A second calculator for calculating a target code amount based on a code amount to be assigned to the orthogonal transform coefficient, calculated by the first calculator; and the target code calculated by the second calculator. Control means for controlling a code amount at the time of requantization of the image compression information based on the amount. 2. The apparatus according to claim 1, wherein the orthogonal transform is a discrete cosine transform. 3. When the target code amount for the header portion is T _header and the target code amount for the orthogonal transform coefficients is T _coef , the second calculating means sets the target code amount for each frame of the image compression information. The code amount T is given by The image information conversion device according to claim 1, wherein the calculation is performed as: Wherein said second calculating means, a code amount of the header unit detected by said detection means, said T _header
The image information conversion device according to claim 3, wherein: 5. A method according to claim 1, further comprising: calculating an average quantization scale for each frame of the captured image compression information;
Calculating means for calculating the complexity for each frame, wherein the second calculating means calculates the target code amount T based on the complexity calculated by the calculating means.
The image information conversion device according to claim 3, wherein _coef is obtained. 6. The image processing apparatus according to claim 3, further comprising: a determination unit configured to determine a pseudo GOP structure from an interval between I pictures in the image information obtained by decoding the image compression information.
2. The image information conversion device according to claim 1. 7. The method according to claim 1, wherein the first calculating unit sets N as the number of frames included in the pseudo GOP,
e is the code amount of the image compression information to be output, and picture_rat
Let e be a value representing how many frames of the output image are displayed per second, and when re-encoding the first frame of the pseudo GOP, the allocated code amount R for the uncoded frame in the pseudo GOP Is given by In the input image compression information, the total R _{head of the} code amount for the header of the frame included in the pseudo GOP is expressed as And the allocated code amount R _coef for the orthogonal transform coefficient of the _uncoded frame in the pseudo GOP is given by: The image information conversion device according to claim 3, wherein the image information is calculated as: 8. The method according to claim 1, wherein, after each frame is re-encoded, when the amount of generated code of the frame allocated to the orthogonal transform coefficient is S _coef , The code amount R _coef is given by: The image information conversion apparatus according to claim 7, wherein the image information is updated as: 9. The method according to claim 6, wherein the second calculating unit uses the allocated code amount R _coef to calculate the target code amount T for each frame.
The image information conversion device according to claim 8, wherein _coef is calculated. 10. The image information conversion apparatus according to claim 3, wherein said second calculating means calculates said target code amount T _coef using values of parameters Kp and Kb according to an image. apparatus. 11. The apparatus according to claim 10, wherein the second calculating means adaptively calculates the value of the parameter using the complexity of each frame of the input image compression information. Image information conversion device. 12. The second arithmetic means assigns the complexity of each frame to an average quantization scale in each frame in the input image compression information and the orthogonal transform coefficient in the frame. The image information conversion device according to claim 11, wherein the calculation is performed using the obtained code amount. 13. Xi, Xp, Xb are each represented by a pseudo GO
When the complexity in the I picture, P picture, and B picture constituting P is set and 1 / (1 + m) is a predetermined value, the second calculating means sets the parameters Kp and Kb as follows: (Equation 7) The image information conversion apparatus according to claim 11, wherein the calculation is performed by: 14. An average quantization scale code in the frame of the input image compression information is Q, and a portion of the total code amount allocated to the orthogonal transform coefficient is S.
_{When coef} is set, the second arithmetic means calculates the complexity X as 14. The image information conversion device according to claim 13, wherein the calculation is performed by: 15. The method according to claim 15, wherein the second calculating means is configured to calculate the 1 / (1
The image information conversion apparatus according to claim 13, wherein the value of (+ m) is set to 0.6 to 1.2. 16. The method according to claim 16, wherein the second calculating means is configured to calculate the 1 / (1
The image information conversion apparatus according to claim 15, wherein the value of (+ m) is set to 1.0. 17. The apparatus according to claim 13, wherein said second calculating means executes an exponential calculation when obtaining said parameters Kp and Kb by referring to a preset table. Image information conversion device. 18. The apparatus according to claim 18, wherein said second calculating means sets Kp = 1.0, Kb = 1. 4
The image information conversion apparatus according to claim 13, wherein the value is used. 19. An image information conversion method of an image information conversion device which takes in image compression information encoded by orthogonal transformation and motion compensation, further reduces the amount of code, and outputs the encoded information. A detecting step of detecting the allocated code amount; and calculating a code amount to be allocated to the orthogonal transform coefficient of the image compression information based on the code amount allocated to the header portion detected by the processing of the detecting step. A second calculation step of calculating a target code amount based on a code amount to be assigned to the orthogonal transform coefficient, which is calculated by the processing of the first calculation step; A control step of controlling a code amount at the time of requantization of the image compression information based on the target code amount calculated by the processing of the calculation step. Picture information converting method comprising Mukoto. 20. A program for controlling an image information conversion apparatus which takes in image compression information encoded by orthogonal transformation and motion compensation and further outputs a code amount after reducing the code amount. A detecting step of detecting the obtained code amount, and calculating a code amount to be allocated to the orthogonal transform coefficient of the image compression information based on the code amount allocated to the header portion detected by the processing of the detecting step. A first calculation step, a second calculation step of calculating a target code amount based on the code amount to be assigned to the orthogonal transform coefficient, calculated by the processing of the first calculation step, and the second calculation A control step of controlling a code amount at the time of requantization of the image compression information based on the target code amount calculated by the process of step; Recording medium from which a computer readable program is recorded, which comprises.