JPH0955939A - Moving picture compression method/device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a moving picture compression method which can securely estimate distortion quantity by means of a simple method and to obtain the best picture with a constant bit rate by controlling code generation quantity based on data obtained by orthogonal conversion or quantization, namely, distribution information of the conversion coefficient of orthogonal conversion. SOLUTION: A data value counting means 41 counting a data value obtained as the result of quantization by a quantization means 14 or a data value obtained in the middle stage of quantization and a code quantity control means 42 controlling the generation quantity of encoding data in accordance with the counted result are provided. In this method, not only generated code quantity and the quantization coefficient but also the data values which are orthogonally converted or quantized are considered and generated code quantity is controlled. The data value counting means 41 selectively counts the data value of the DC component of a non-intra-macro block being a specified block attribute based on the discriminated result of a macro block and outputs the counted result to the code quantity control means 42.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a moving picture compression method and apparatus for compressing a digital moving picture, and in particular, orthogonally transforms each image forming a digital moving picture into a predetermined block unit and obtains the resulting orthogonal transformation. The present invention relates to a moving picture compression method and apparatus for quantizing coefficients and performing coding processing.

[0002]

2. Description of the Related Art In recent years, MPEG (Moving
Picture Experts Group) 1, followed by MPEG2
Each became an international standard. An encoder that adopts the MPEG moving image compression method to generate compression information from digital moving image data uses an image (hereinafter, also referred to as a picture) forming each screen of the moving image as a small block (macro block = 16 pixels ×). 16-pixel rectangular block), and after extracting a region (reference image region) similar to the macro block to be compressed from temporally previous and subsequent images, the spatial distance and orientation to the reference image region Is calculated and the difference information between the reference image area and the area to be compressed is calculated, and these pieces of information are calculated by the DCT (Discrete
te Cosine Transform (discrete cosine transform) Orthogonal transform, quantization and variable length coding are used for compression. When the motion vector and difference information are compressed in this way,
This is because it can be compressed much more efficiently than compressing the current image itself. Also, a picture compressed by the difference information cannot be restored without a picture to refer to, so an I-picture (Intra-Pi
cture; intra-frame coded image) is periodically provided. When this I picture is compressed, all macroblocks of the I picture are subjected to DCT orthogonal transform, quantization, and variable length coding as they are as intra blocks that do not refer to other images, and then compressed. Then, the next image is compressed by using this I picture as a reference image, and further the other images are compressed by using the already compressed image as a reference image. For a picture to be compressed using the reference image, only the picture that is temporally ahead is used as the reference image.
There are (Predictive) pictures and B (Bidirectionally-Predictive) pictures in which temporally preceding and following pictures are used as reference images. It should be noted that the P picture can be used as a reference image for other images as with the I picture, but the B picture is not used as a reference image for other images. Further, whether or not to compensate by the motion vector is determined by estimating and determining the difference in the amount of generated bits for each macroblock of the input image depending on the presence or absence of the compensation. On the other hand, in a decoder that restores (decompresses) a moving image from compressed information (bit stream), inverse processing of encoding, that is, inverse quantization, inverse DCT, synthesis with a reference image (addition) is performed to generate a moving image. Restore.

By the way, since compression and decompression by MPEG cause distortion in the process of quantization and dequantization, an image restored from compressed data (bit stream) is slightly different from the current image. If the quantization is performed finely, the distortion is reduced, but the compression efficiency is reduced. Therefore, it is necessary to appropriately control the quantization characteristic in order to obtain the best image quality at the determined bit rate.

Further, the quantization characteristic is a quantization coefficient determined for each macroblock (Voice Society of Japan, 1995 Vo.
l. 49, No. 4 (hereinafter referred to as Reference 1) Page 45 (10)
(Corresponding to the quantizer scale in the equation) and the quantization matrix determined by the order of the DCT coefficient (Page 45 of Document 1)
(Corresponding to W [w] [v] [u] in Eq. (10)) is controlled as a parameter, and usually the quantization step is set by adjusting the quantization coefficient, and the amount of generated bits is adjusted. It is supposed to do.

As a conventional quantization control method, as described in "3-2-5 Rate control and buffer control" on pages 49 to 50 of Document 1, the generated code amount of a screen encoded in the past and It has been proposed to estimate the bit allocation that optimizes the overall image quality based on the quantization coefficient and perform the quantization control. Further, as another conventional quantization control method, the 1994 Image Coding Symposium (PCSJ
94) 9th Symposium Materials, pp. 185-186,
"A study on code amount control for video coding" (hereinafter,
As described in (Reference 2), a method of quantitatively evaluating the distortion (SNR: Signal to Noise Ratio) of the restored image and performing the quantization control so that the SNR becomes the best is also proposed.

[0006]

However, in the conventional moving image compression method and apparatus described in the above-mentioned Document 1, the quantization coefficient is evaluated by encoding instead of evaluating the actual distortion amount. Since the quantity is controlled, it is difficult to make an adjustment to minimize the distortion. This is because there is a strong correlation between the quantized coefficient and the amount of distortion, but they are somewhat different. That is, the conventional moving image compression method and apparatus have a problem that the distortion amount cannot be accurately grasped by a simple method of estimating the distortion amount from the quantized coefficient, and a device for obtaining the distortion amount becomes complicated. I was holding.

Further, in another conventional example described in the above-mentioned document 2, in order to quantitatively evaluate the SNR, all the encoded images are decoded and compared with the current image, which requires a huge calculation. However, there is a problem that the cost of the compression device is increased. In addition, normal MPE
Even in the G encoder, it is necessary to create a decoded image in order to use it as a reference image, but it is not necessary to decode the B picture because only I pictures and P pictures are required as reference images.

In order to solve the above-mentioned conventional problems, the present invention realizes a moving image compression method capable of reliably estimating the distortion amount by a simple method, and obtains the best image at a constant bit rate. It is an object of the present invention to provide a low-cost moving image compression apparatus capable of performing the above.

[0009]

Means for Solving the Problems The present inventor
In the above, we paid attention to the following reasons that the quantized coefficient does not accurately reflect the distortion amount. First, in MPEG, the inverse quantization in the non-intra macroblock of the decoder is defined by the following equation. (4th of the above-mentioned literature 1
Expressions (10) and (11) in "3-2-4 DCT encoding, quantization, variable length encoding" on pages 3 to 48. However, in Document 1, the non-intra block is described as “non-intra block”. ) F_dash [v] [u] = ((2 × QF [v] [u] + k) × W [1] [v] [u] × quantizer scale) / 32 .. (1) k = Sign (QF [v] [u]) .. (2) where F_dash [v] [u]: DCT coefficient after dequantization QF [v] [u]: DCT coefficient before dequantization (that is, quantization) DCT coefficient after) W [1] [v] [u]: Quantization matrix (for non-intra macroblock) quantiser scale: Quantization coefficient Sign (): 1 if the sign of the argument is positive, and negative -1,
If it is 0, a function that returns 0 /: Truncation division For inverse quantization having such characteristics, quantization is usually performed by the following formula.

QF [v] [u] = ((F [v] [u] × 16 + W [1] [v] [u] / 2) / W [1] [v] [u]) / quantiser scale) / 32 .. (3) where F [v] [u]: DCT coefficient before quantization /: Truncation division Using these conversion formulas, the DCT coefficient F [v] [u] is quantized The DCT coefficient QF [v] [u] before dequantization is obtained, and this QF [v] [u] is dequantized to obtain the DCT coefficient F after dequantization.
Get _dash [v] [u]. The difference between the DCT coefficient F [v] [u] before quantization and the DCT coefficient F_dash [v] [u] after dequantization is
It is distortion due to quantization and inverse quantization.

In this equation, the point different from the inverse quantization by mere multiplication is that there is a term of Sign (QF [v] [u]), and the presence of this term results in QF as shown in FIG. [v]
The reverse such that the range of F [u] [v] where [u] is 0 is wider than the range of F [u] [v] where QF [v] [u] is 1 or more or -1 or less. Quantization is performed (hereinafter, this working area is called a dead zone). Therefore, a wider range is quantized as 0, and bit reduction can be performed efficiently. Note that in FIG. 6, W [1] [v] [u] = 16, quantizer_sca
It shows the relationship between F [v] [u] and F_dash [v] [u] under the condition of le = 8, and the lines indicated by multiple black circles and solid lines are F [v] [u] and F.
The lines represented by dotted lines in the relationship of _dash [v] [u] represent the conversion equations with zero distortion. That is, the difference between the solid line and the dotted line represents the quantization distortion. Also, the level when F_dash = 0 is Q
F [v] [u] = 0, F_dash = ± 12 corresponds to QF [v] [u] ± 1, and F_dash = ± 20 corresponds to QF [v] [u] ± 2. In the range of QF [v] [u] = 0, there are 15 widths from −7 to 7 in the value of F, while other values have a width of 8. Since the width of QF [v] [u] = 0 is wider than that of other values, the quantization error when QF [v] [u] = 0 is the quantum of other values. It will be larger than the error. In addition, in the case of QF [v] [u] = 0, a maximum of 7 quantization distortions indicated as deviations from the dotted line (transformation formula of 0 distortion) occur, whereas in other cases, Only a maximum error of 4 has occurred. Thus, even with the same quantized coefficient, the quantized coefficient does not accurately reflect the distortion amount due to the phenomenon that the quantized error differs when the value of QF [v] [u] is different.

Therefore, in the present invention, in order to accurately grasp the distortion amount and improve the accuracy of the code amount control, not only the quantized coefficient but also the quadrature transform coefficient information to be quantized is grasped and used for the code amount control. To reflect. Then, when there are few macroblocks that become non-zero after quantization, the distortion is larger, so control is performed so that the quantization coefficient becomes small, and when there are many macroblocks that become non-zero after quantization, Control so that the conversion coefficient becomes large. By doing so, the distortion becomes uniform, and a good restored image can be obtained.
In particular, this effect is prominent for a moving image in which there is a large difference in the non-zero coefficient ratio between the P picture and the B picture.

For example, as shown in FIG. 7, the brightness of the screen is
For moving images (fade images) that change slowly and continuously
In the P picture, is it a reference image (I picture)?
Difference D₁Is larger than a normal image,
The number of numbers increases. On the other hand, B picture is the previous and next images.
The difference D from the reference image is used because the average of ₂
Is not very large, and the non-zero coefficient does not increase so much.
No. That is, non-zero between P picture and B picture
There is a big difference in the ratio of the coefficients. In such a moving image
On the other hand, judging from the quantized coefficient as in the method of Reference 1,
If you control so that the bit allocation is appropriate, P
Difference in the amount of distortion between the picture and the B picture,
The amount of distortion increases with the repetition period of P picture and B picture.
A so-called flicker phenomenon occurs, and it is very unsightly to recover.
It becomes the original image.

On the other hand, for example, the code amount control may be performed based on the number of blocks and the quantization coefficient in which the DC component (corresponding to the average value of the pixel data in the block) becomes non-zero after quantization. For example, the P-picture with many non-zero coefficients has a larger quantization step width than that of the conventional method, and the B-picture with less non-zero coefficients has a smaller quantization step width than that of the conventional method so that the distortion is uniformized. It becomes possible to obtain a restored image. In particular, distortions of DC (direct current) components and low-frequency components are conspicuous. Therefore, by grasping the number of blocks in which these components are non-zero, the code amount and the image quality can be controlled more effectively. it can.

Further, for code amount control, the distortion of the DCT coefficient may be obtained instead of the distortion of the restored image. Therefore, as the second invention, the distortion of the DCT coefficient (from the dotted line of the black circle in FIG. 6). Deviation) is calculated and reflected in the code amount control. This makes it possible to achieve the same effect as the code amount control described in Document 2 with a simpler device.

The inventor of the present invention has reached the following solving means from the above viewpoints. That is, the invention according to claim 1 orthogonally transforms a moving image or a predictive difference image of the moving image in units of a predetermined block, and quantizes an orthogonal transform coefficient obtained as a result of the orthogonal transform to obtain a code of the image. A moving image compression method for generating encoded data, comprising the step of controlling the generation amount of the encoded data based on the data obtained by the orthogonal transformation or the quantization thereof. In this case, by reflecting the information of the orthogonal transform space (information of the spatial frequency domain) instead of the real space in the code amount control, it is possible to more accurately grasp the quantization distortion and improve the accuracy of the code amount control.
Further, as described in claim 2, it is preferable that the generation amount of the encoded data is controlled according to the number of blocks in which the DC component of the orthogonal transform coefficient becomes non-zero after the quantization. If the code amount is controlled based on the number of orthogonal transform coefficients that become non-zero after quantization and the quantized coefficient, the quantization step width can be made larger in the P picture having many non-zero coefficients than before. A B-picture with few zero coefficients has a smaller quantization step width than that of the conventional method, and the distortion (S
This is because a restored image with uniform NR) can be easily obtained.

According to a third aspect of the present invention, the moving image or the predictive difference image of the moving image is orthogonally transformed in a predetermined block unit, and the orthogonal transformation coefficient obtained as a result of the orthogonal transformation is quantized, A moving image compression method for generating encoded data of an image, the step of calculating a difference between an orthogonal transform coefficient obtained by dequantizing the quantized orthogonal transform coefficient and the quantized orthogonal transform coefficient. And a step of controlling the generation amount of the encoded data based on the sum of the differences. In this case, it is possible to accurately estimate the distortion without calculating the distortion of the image in the real space by calculating the difference between before and after the quantization of the orthogonal transformation coefficient and counting the sum to calculate a good image. Can be obtained.

According to a fourth aspect of the present invention, an orthogonal transform means for orthogonally transforming a moving image or a predictive difference image of the moving image in a predetermined block unit, and a quantum for quantizing an orthogonal transform coefficient obtained as a result of the orthogonal transform. And a data value obtained as a result of the quantization by the quantizing means in a moving picture compression apparatus for generating encoded data of the image by the encoding process including the orthogonal transformation and the quantization. Data value counting means for counting the data values obtained in the middle of the step, and code amount control means for controlling the generation amount of the encoded data according to the counting result of the data value counting means are provided. According to a fifth aspect of the present invention, the data value counting means has means for determining the block attribute of the image of the block unit for the data value, and the block Preferably selectively count data values corresponding to a specific block attribute on the basis of the sex determination result. This is because it is possible to select only the non-intra blocks that have an effect by reflecting the counting result in the control of the code generation amount and efficiently process the non-intra blocks. In addition, claim 6
As described above, the data value counting means has means for determining the conversion order of the orthogonal transformation for the data value, and based on the result of the determination of the conversion order, the data value corresponding to the specific conversion order is obtained. Alternatively, the data value counting means may have means for determining whether or not the orthogonal transformation coefficient is zero for the data value. However, a specific data value corresponding to the non-zero value may be selectively counted based on the determination result. Blocks in which the DC (direct current) component is non-zero, or distortion of low-frequency components are also noticeable, so by grasping the number of blocks in which the components are non-zero, the code amount and image quality can be effectively controlled. Because it can be done.

Further, in the invention described in claim 8, an orthogonal transform means for orthogonally transforming a moving image or a predictive difference image of the moving image in a predetermined block unit, and an orthogonal coefficient obtained as a result of the orthogonal transform are quantized. In a moving picture compression apparatus, which comprises a quantizing means and which generates coded data of the image by an encoding process including the orthogonal transformation and quantization, an inverse quantizing means for inversely quantizing the output of the quantizing means. And an orthogonal transform coefficient difference calculating means for calculating the difference between the orthogonal transform coefficient dequantized by the inverse quantizing means and the orthogonal transform coefficient before quantization, and a data value for counting the output of the orthogonal transform coefficient difference calculating means. The counting means and the code amount control means for controlling the generation amount of the encoded data based on the counting result of the data value counting means are provided, and as described in claim 9. , The data value counting means has means for determining the conversion order of the orthogonal transformation for the data value, and selectively counts the data value corresponding to a specific conversion order based on the result of the determination of the conversion order. Is desirable.

[0020]

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings. As shown in the block diagram of FIG. 1, the moving image compression apparatus of the present embodiment is
Orthogonal transform means for orthogonally transforming a moving image or a predictive difference image of the moving image in a predetermined block unit, for example, DCT (Discre
te Cosine Transform) means 13
And a quantizing unit 14 for quantizing an orthogonal transform coefficient obtained as a result of the orthogonal transform, for example, a DCT coefficient, and generates encoded data of the image by an encoding process including the orthogonal transform and quantization. It is supposed to do.
The moving picture compression apparatus further includes a data value counting means 41 for counting the data value obtained as a result of the quantization by the quantization means 14 or the data value obtained in the middle of the quantization.
And a code amount control unit 42 for controlling the generation amount of the encoded data according to the counting result of the data value counting unit 41. The data value counting means 41 is the quantizing means 14
The orthogonal transformation coefficient information, that is, the information of the orthogonal transformation space, is grasped by counting the data value obtained as a result of the quantization by or the data value obtained in the middle of the quantization. Further, the code amount control means 42 grasps the orthogonal transform coefficient information from the result of counting by the data value counting means 41, and controls the generation amount of the encoded data in the block unit according to the result.

As a mode of carrying out the moving picture compression method of the present invention using this apparatus, first, the moving picture or the prediction difference image of the moving picture is orthogonally transformed into a predetermined block unit by the orthogonal transformation means 13. , A method of generating coded data of the image by quantizing the orthogonal transformation coefficient obtained as a result of the orthogonal transformation by the quantizing means 14, and the orthogonal transformation coefficient or data obtained by the quantization thereof, for example, the orthogonal transformation. A step of controlling the quantization means 14 by the code amount control means 42 based on the information from the data value counting means 41 for counting the data after the quantization of the transform coefficient to control the generation amount of the encoded data. It will be.

In this way, when the orthogonal transform coefficient information is reflected in the code amount control, if there are few blocks in which the orthogonal transform coefficient becomes non-zero after quantization and the quantization distortion becomes large, the quantization coefficient becomes small. On the other hand, if there are many macroblocks that become non-zero after quantization and the quantization distortion is small, if the quantization coefficient is controlled to be large, the distortion will be uniform and a good restored image will be obtained. Obtainable.

In the code amount control step, the generation amount of the encoded data can be controlled according to the number of blocks in which the DC component (DC component) of the orthogonal transform coefficient is non-zero. In that case, in particular, A significant effect can be obtained for a moving picture in which the ratio of non-zero coefficients is greatly different between the P picture and the B picture. The data value counting unit 41 may include a unit 51 for discriminating the block attribute of the image of the block unit for the data value. In that case, a specific block attribute is determined based on the discrimination result of the block attribute. Corresponding data values can be selectively counted. Further, the data value counting means 41
May have a means 52 for discriminating the transformation order of the orthogonal transformation with respect to the data value and selectively counting data values corresponding to a particular transformation order based on the discrimination result of the transformation order. Well, further, a specific data value corresponding to the non-zero value is selectively counted on the basis of the determination result of the means 53 for determining whether or not the orthogonal transformation coefficient is zero value for the data value. You can also do it.

The "orthogonal transform" used in the present invention is preferably a discrete cosine transform (DCT), but is not limited to this. Further, the “orthogonal transform coefficient” obtained as a result of the orthogonal transform may include not only the DC component but also other orders, and for example, it is preferable that the orthogonal transform coefficients of orders up to the first order are included.

[0025]

1 to 3 are diagrams showing detailed examples of a moving image compression apparatus according to this embodiment, and FIG. 1 is a block diagram of this example. In FIG. 1, reference numeral 1 is an image signal input terminal for inputting a digital moving image signal via a frame memory or the like (not shown), and the image signal input terminal 1 stores images (pictures) constituting each screen of the moving image. Data including brightness and color difference is input. For the input image data, each screen of the moving image is, for example, 16 pixels × 16
Blocks are divided into rectangular blocks of pixels (hereinafter referred to as macroblocks), and the blocks are sent to the difference image generation means 12 and the motion vector search means 31 in macroblock units. The motion vector search unit 31 compares the image of the macroblock to be encoded (encoded) with the reference image (I or P picture already encoded) stored in the restored image recording unit 24, and compares To search the image area most similar to the input macroblock and output the resulting motion vector to the reference area extracting means 32. The reference area extracting unit 32, on the basis of the motion vector obtained as a result of the search by the motion vector searching unit 31, selects an area similar to the macroblock to be compressed from the reference images in the restored image recording unit 24 as the reference area. To extract.

The differential image generating means 12 is based on the data from the blocking means 11 and the reference area extracting means 32, and the spatial frequency of the differential image, which has been motion-compensated using the spatial frequency distribution of the input image and the motion vector in advance. , A macroblock attribute that reduces the amount of generated bits is selected (whether or not motion compensation is performed), and either an input image or a motion-compensated difference image is output using a motion vector. . The attributes of the selected macroblock are an intra macroblock that encodes the current image without performing motion compensation, a non-intra macroblock that encodes a difference image by performing motion compensation, and one frame. There is a field encoding method in which the image of is divided into two fields and encoded.

More specifically, for example, when the input image contains many components with high spatial frequencies, while the difference image has only components with low spatial frequencies, it is better to orthogonally transform the difference images. Since it can be converted to a small amount of data,
It is more efficient to encode the input macroblock as a non-intra macroblock. Therefore, the difference image generation unit 12 has a value corresponding to an AC component (AC component) when the input image is encoded as an intra block as it is and a value corresponding to the AC component when encoded as a non-intra block that is the difference image. And are respectively calculated, and an advantageous macroblock attribute is determined according to the comparison result of both values. This is because the generated bit amount can be estimated and judged by using the general tendency that the generated code amount increases when the AC component (AC component) of the DCT coefficient obtained by the DCT conversion increases. The difference image generation means 12 performs a subtraction process between the input image and the reference image according to the determination result as needed, generates a difference image and outputs it to the DCT means 13. The input image to the blocking means 11 is an I picture (Intra-Pictur).
e; intra-frame coded image), all macroblocks in the I picture become intra blocks, and the data from the blocking means 11 remains as is in the DCT means 1.
3 and the input image to the blocking means 11 is a P picture (Predictive-Picture; inter-frame forward prediction coded image) or B picture (Bidirectio).
nally predictive-Picture), the attribute of the macroblock to be quantized as described above is determined for each macroblock.

The DCT means 13 is the difference image generating means 12
After dividing the prediction error signal for each macroblock (input signal for intra macroblocks, signal of the difference image for non-intra macroblocks) into blocks of, for example, 8 × 8 pixels, a known two-dimensional D
A CT operation is performed to convert the input image into a plurality of DCT coefficients from a low frequency term including a direct current component (DC component) and an alternating current component (AC component) to a high frequency term. By this DCT transformation, the input image is decomposed into a plurality of stepwise image components that gradually express fineness, from the DC component (average value image) that is the first low-frequency term to the AC component of the high-frequency term. Will be. In addition, in the natural image, the pixel values (for example, luminance or color difference) randomly distributed before the DCT conversion are D
After the CT conversion, the high-frequency terms are concentrated, so that it is possible to effectively compress information by removing the high-frequency terms.

The DCT coefficient output from the DCT means 13 is quantized by the quantizing means 14 according to the macroblock attribute (intra macroblock, non-intra macroblock) by a quantizing matrix and macro depending on the frequency. Quantization is performed using a quantization coefficient that is determined for each block. The quantizing means 14 includes a DC component (DC component) and an AC component (A) of the DCT coefficient obtained by the DCT transform.
C component) is independently divided by a divisor called a quantization step, the remainder is rounded to remove high frequency terms, and each DCT coefficient is quantized. The quantization step width of the quantization performed by the quantization means 14 is controlled by the code amount control means 42, which will be described later, whereby the generated code amount (generated bit amount) for each macroblock can be controlled.

The data quantized by the quantizing means 14 is variable-length coded by the variable-length coding means 15 and then outputted from the output terminal 2 while averaging the bit rate in the buffer 16, which is not shown in the figure. Is transmitted to the external decoder via the transmission line of. That is, since the amount of information generated in an image signal varies depending on the complexity of the image and the intensity of movement,
A transmission buffer 16 is provided to absorb this fluctuation and transmit at a substantially constant transmission rate. Further, a code amount counting means 17 (described later) for counting the generated code amount is provided so that the quantization characteristic can be controlled based on the occupation rate of the buffer.

On the other hand, the data quantized by the quantizing means 14 is also input to the inverse quantizing means 21, and the inverse quantizing means 21 dequantizes and the inverse DCT means 22 inverse D.
It is restored by the CT conversion, that is, by the same processing as the decoder. The restored image, for example, the restored I-picture image is stored in the restored image recording means 24 as a local decoded image and becomes a reference image of the image to be encoded next. This processing is performed in order to suppress the difference between the image of the I picture restored by the decoder and the current image, and when another image is transmitted using the I picture as a reference image, the I image restored by the decoder is used. This is because the same image as the picture image must be created on the decoder side. Note that the P picture, like the I picture, is an inverse quantization means 21 and an inverse DCT means 22.
Is processed in the restored image generating means 23 and synthesized (added) with the reference image from the reference area extracting means 32 used at the time of encoding, so that the restored image is restored in the same manner as the decoder processing and stored in the restored image storing means 24. Saved.

By the way, in order to keep the bit rate of the data (bit stream) encoded (orthogonal transformation and quantization) at the target rate in the encoding process and to obtain a better restored image quality, the conventional technique is used. As described above, it is necessary to properly control the quantization step width. Therefore, in this embodiment, the code amount counting means 17 for counting the generated code amount and the quantization coefficient counting means 18 for counting the quantized coefficient are provided, and the code amount control means 42 includes both means 17, The quantizing means 1 determines the allocated bit amount of the image to be encoded next based on the generated code amount and the average quantization coefficient input as information from 18.
It is possible to control the generated code amount by changing the quantization step width of No. 4.

In the case of such control, the number of generated codes is counted by the conventional moving image compression apparatus described in Document 1, and the image to be encoded next is assigned based on the generated code amount and the average quantization coefficient. The processing content is almost the same as that of determining the bit amount. Expressions (18) to (25) on page 49 of Document 1 express this control. In this case, it is assumed that the average of the quantized coefficients reflects the image quality. The quantization coefficient is controlled so as to have a constant ratio (1: Kp: Kb).

On the other hand, in this embodiment, not only the generated code amount and the quantized coefficient but also the orthogonally transformed and quantized (ie, encoded) data value including the information of the DCT space is taken into consideration. The quantity control is performed, and the data value distribution counting means 41 is provided for that purpose.
FIG. 2 is a detailed diagram of the data value distribution counting means 41 in this embodiment. In FIG. 2, reference numeral 51 identifies the attribute of the macroblock with respect to the data value input from the quantizing means 14, and the determination result of the block attribute. It is a non-intra macroblock extracting means for extracting data corresponding to the non-intra macroblock based on. Further, 52 is a DC component extracting means for extracting DC component data representing an average value of pixel data (for example, luminance) of each macroblock among the data from the non-intra macroblock extracting means 51, and 53 is a DC component extracting means 52. Is a non-zero coefficient extracting means for extracting the non-zero coefficient which is the data of the non-zero component from the data from 1. The non-zero coefficient extracted by the non-zero coefficient extracting means 53 is counted by the counting means 54. That is, the data value counting means 41 selectively counts the data value of the DC component of the non-intra macroblock having the specific block attribute based on the determination result of the macroblock attribute, and the count result is the code amount control means. It outputs to 42. The reason for extracting the DC component of the non-intra macroblock in this way is that the present invention is effective for the non-intra macroblock.
This is because in EG, the number of non-zero coefficients of non-intra macroblocks is a measure of quantization distortion, and the DC component is the most noticeable frequency component.

FIG. 3 is a detailed diagram of the code amount control means 42 in this embodiment. As shown in the figure, the code amount control means 42 includes Xi, Xp, Xb calculation means 61 and Xi storage means 6.
2, Xp storage means 63, Xb storage means 64, Ti, Tp, Tb
It has a calculating means 65, a dji, djp, djb calculating means 66 and a Q calculating means 67. Xi, Xp, Xb calculation means 6
1 is the code amount counting means 17 and the quantization coefficient counting means 1
8 and the values Xi, Xp and Xb representing the complexity of the encoded image based on the information from the non-zero coefficient counting means 54.
Are calculated by the following equations.

Xi = Si × Qi .. (4) Xp = Sp × Qp × (1-α × (N_nonintra_dc_nonzero_p-N_nonintra_dc_nonzero_b)) .. (5) Xb = Sb × Qb × (1 + α × (N_nonintra_dc_nonzero_p-N_nonintra_dc_non )) .. (6)

Here, Xi: a value representing the complexity of an image encoded as an I picture. Xp: A value representing the complexity of an image encoded as a P picture. Xb: A value representing the complexity of an image encoded as a B picture. Si: generated bit amount of an image encoded as an I picture. Sp: generated bit amount of an image encoded as a P picture. Sb: generated bit amount of an image encoded as a B picture. Qi: Average quantization coefficient of an image encoded as an I picture. Qp: Average quantization coefficient of the image encoded as a P picture. Qb: Average quantization coefficient of an image encoded as a B picture.

N_nonintra_dc_nonzero_p: Ratio of non-zero coefficient of DC component of non-intra macroblock of image encoded as P picture. Count value of the counting means 54. N_nonintra_dc_nonzero_b: Ratio of non-zero coefficient of non-intra macroblock DC component of image encoded as B picture. Count value of the counting means 54. α: A constant (0 that indicates the influence of the non-zero coefficient ratio on the image quality)
≤α).

The Xi storage means 62 and the Xp storage means 63 are also provided.
The Xb storage means 64 stores the values Xi, Xp, and Xb representing the complexity of the image calculated by the Xi, Xp, and Xb calculation means 61, respectively. At the start of encoding, the Xi storage means 62, Appropriate initial values are set in the Xp storage means 63 and the Xb storage means 64, respectively. Further, the Ti, Tp, Tb calculation means 65 allocates each bit when the next screen is encoded as an I, P and B picture based on the data from the Xi storage means 62, Xp storage means 63 and Xb storage means 64. The quantities Ti, Tp, Tb are calculated by the following equations. Note that the following formula is a predetermined picture aggregate (GOP = Group of pictu) consisting of one screen of I pictures and one or more screens of P and B pictures.
This is an example in which the total bit amount of (res) is determined and the bit amount assigned to each picture is determined and encoded.

Ti = max {R / 1 (1+ (NpXp) / (XiKp) + (NbXb) / (XiKb)), bit rate / (8 × picture rate)} .. (7) Tp = max {R / 1 (Np + (NbKbXb) / (KbXp)), bit rate / (8 × picture rate) .. (8) Tb = max (R / 1 (Nb + (NpKbXp) / (KpXb)), bit rate / (8 × picture rate) .. (9)

Here, Ti: allocated bit amount when the next screen is encoded as an I picture. Tp: Allocation bit amount when the next screen is encoded as a P picture. Tb: Allocated bit amount when the next screen is encoded as a B picture. R: Amount of bits left in GOP. Np: The number of P pictures that have not been encoded in the GOP. Nb: The number of B pictures that have not been encoded in the GOP.

Kp: P picture quantized scale code ratio based on the I picture quantized scale code. Kb: Ratio of B picture quantization scale code with reference to I picture quantization scale code. max {a, b}: A function that returns the larger value of a and b.

When the allocated bit amount is calculated in this way, each screen is contained in the allocated bit amount, so that the quantization coefficient of each macroblock is appropriately controlled. Dj in the above formula introduced for that purpose
The parameters i, djp, and djb indicate the remaining capacity of the virtual buffer, and are calculated by the following formula by the dji, djp, and djb calculating means 66 (corresponding to formulas (28) to (30) in Literature 1). .

Dji = d0i + Bj-1−Ti × (j-1) / MBcnt .. (10) djp = d0p + Bj-1−Tp × (j-1) / MBcnt .. (11) djii = d0i + Bj- 1-Tb x (j-1) / MBcnt .. (12) where, dji: remaining amount of virtual buffer for I picture djp: remaining amount of virtual buffer for P picture djb: remaining amount of virtual buffer for B picture

D0p: The remaining amount of the P picture virtual buffer at the start of the first macroblock encoding. The initial value is 0. d0b: Remaining amount of the B picture virtual buffer at the start of the first macroblock encoding. The initial value is 0. Bj-1: Amount of bits required for the encoded macroblock on the screen. j-1: Number of macroblocks encoded in the screen. MBcnt: Number of macroblocks on the screen.

Using the above-mentioned dji, djp, and djb, the quantized coefficient Q of the macroblock to be encoded next is calculated by the Q calculation means 67 according to the following equation ((31) in Document 1).
~ (32) formula). Qj = dj × 31 / (2 × bit rate / picture rate) .. (13) where, Qj: quantization coefficient (quantise) of the j-th macroblock
r scale code). dj: Virtual buffer remaining amount according to the picture type, that is, dji, djp or djb.

As described above, the quantization coefficient set in the macro block unit is used to perform appropriate quantization by the quantization means 14, and the quantization coefficient used for the quantization is sent to the quantization coefficient counting means 18. Is counted. Then, the coefficient value of the quantization coefficient counting means 18 is used when calculating the allocated bit amount of the next screen.

As described above, when the code amount control is performed in consideration of the frequency of non-zero coefficients, for example, N_nonintra_d
In the case of a faded image in which c_nonzero_p is large and N_nonintra_dc_nonzero_b is small, the value Xp representing the complexity of the image encoded as the P picture is smaller than that in the conventional code amount control (described in Reference 1), and B The value Xb representing the complexity of the image encoded as a picture becomes large. Such changes in the values Xp and Xb reduce the allocated bit amount Tp when the next screen is encoded as a P picture, and the allocated bit amount T when the next screen is encoded as a B picture.
play a role in increasing b. Therefore, the remaining amount djp of the P picture virtual buffer is small, and the remaining amount djb of the B picture virtual buffer is large. As a result,
The quantized coefficient of the P picture becomes large, and the quantized coefficient of the B picture becomes small. This control is performed by the above-mentioned dead zone effect, that is, the quantized DCT coefficient QF.
The width of [v] [u] = 0 becomes wider than that of other values (for this reason, when the quantization error when the DCT coefficient QF [v] [u] = 0 after quantization is another value) Becomes larger than the quantization error of, and even if the quantization coefficient is the same, if the value of the quantized DCT coefficient QF [v] [u] is different, the quantization error is different and the quantization coefficient is distorted. The effect of canceling out the phenomenon that the P picture and the B picture are distorted in a faded image or the like can be prevented from increasing. As a result, the conventional problem that the restored image becomes unsightly due to the flicker phenomenon in which the distortion amount increases between the P picture and the B picture and the distortion amount increases in the repetition period of the P picture and the B picture is solved. Therefore, a good restored image can be obtained. Moreover, the moving picture compression apparatus of the present embodiment counts the data after the quantization of the DCT coefficient by using the data value counting means 41, so that the distortion of the actual restored image (SN
Since a simple process of estimating the distortion amount without calculating R) and reflecting the information in the code amount control is adopted, it is necessary to provide a complicated arithmetic circuit for calculating the distortion of the restored image. And can be a low-cost device.

As described above, according to the present invention, a moving image in which the distortion amount (SNR) can be easily controlled by a simple method of reflecting the information of the DCT space in the code amount control can be easily performed. (EN) Provided is a low-cost moving image compression apparatus that can realize an image compression method and can obtain the best image at a constant bit rate, and can implement the method to obtain the best image at a constant bit rate. be able to.

[0050]

BEST MODE FOR CARRYING OUT THE INVENTION Next, a preferred embodiment of the second invention will be described. As shown in the block diagram of FIG. 4, the moving image compression apparatus according to the present embodiment is a DCT as an orthogonal transform unit that performs orthogonal transform, for example, DCT transform, on a moving image or a predicted difference image of the moving image in a predetermined block unit.
Means 13 and quantizing means 14 for quantizing the orthogonal transform coefficient obtained as a result of the orthogonal transform, eg DCT,
The coded data of the image is generated by the coding process including the orthogonal transformation and the quantization. The moving picture compression apparatus further includes an inverse quantizer 22 for inversely quantizing the output of the quantizer 14, and an inverse quantizer 22.
Quantization error calculating means 73 as an orthogonal transform coefficient difference calculating means for calculating the difference between the inverse transform quantized orthogonal transform coefficient and the unquantized orthogonal transform coefficient, and the output of the quantization error calculating means 73 are counted. The data value counting means 71 and the code amount control means 72 for controlling the generation amount of the encoded data in the block unit based on the counting result of the data value counting means 71. Further, the data value counting means 71 has means 52 for discriminating the transformation order of the orthogonal transformation for the data value, and selectively selects the data value corresponding to a particular transformation order based on the discrimination result of the transformation order. It may be one that counts.

As a mode for carrying out the moving picture compression method of the present invention using this apparatus, the prediction difference image of the moving picture is orthogonally transformed in a predetermined block unit, and the orthogonal transformation obtained as a result of the orthogonal transformation is carried out. Quantize the coefficients to generate coded data for the image. Further, a step of calculating a difference between the orthogonal transform coefficient obtained by dequantizing the orthogonal transform coefficient after the quantization and the orthogonal transform coefficient before the quantization,
Controlling the generation amount of the encoded data in the block unit based on the sum of the differences.

The "orthogonal transformation" here is preferably discrete cosine transformation (DCT), but is not limited to this. Further, the “orthogonal transform coefficient” obtained as a result of the orthogonal transform may include not only the DC component but also other orders, and for example, it is preferable that the orthogonal transform coefficients of orders up to the first order are included.

[0053]

EXAMPLE FIG. 4 and FIG. 5 are views showing an example of the moving image compression apparatus of this embodiment. The moving image compression apparatus of the second embodiment is different from the above-described embodiment (first embodiment) in the quantization error calculation means 73, which is an orthogonal transform coefficient difference calculation means, and the configuration related thereto, but most of the other. Are almost the same. Therefore, the same or corresponding components as those of the first embodiment are designated by the same reference numerals, and only different components will be described here.

As shown in FIG. 4, the quantization error calculating means 7
3 is for calculating the quantization error of the DCT coefficient.
The difference between the DCT coefficient before quantization output from the T means 13 and the DCT coefficient after inverse quantization output from the inverse quantization means 21 (data before inverse DCT conversion) is calculated as a quantization error, The calculated value of the quantization error is output to the data value counting means 71.

As shown in FIG. 5, the data value counting means 71 is a DC which represents the average value of the pixel data (for example, brightness) of each macroblock among the data from the error calculating means 73.
It has a DC component extracting means 82 for extracting the component data and a counting means 84 for counting the data from the DC component extracting means 82. That is, the data value counting means 71
Of the DCT coefficient, only the DC component of the quantization error is extracted and counted, and the counting result is sent to the code amount control means 72.

The code amount control means 72 is constructed in substantially the same manner as the code amount control means 42 of the first embodiment, and has Xi, Xp, X.
Xi, Xp, Xb calculation means similar to the b calculation means 61, Xi storage means 62, Xp storage means 63, Xb storage means 64, Ti, T
It includes p, Tb calculating means 65, dji, djp, djb calculating means 66 and Q calculating means 67. And that X
The i, Xp, Xb calculating means calculates the values Xi, Xp and Xb representing the complexity of the encoded image based on the count value of the quantization error DC component from the data value counting means 71 by the following equations.

Xi = Si × Qi × (1+ β × dc_distortion_i) .. (14) Xp = Sp × Qp × (1+ β × dc_distortion_p) .. (15) Xb = Sb × Qb × (1+ β × dc_distortion_b) .. (16)

Here, Xi: a value indicating the complexity of an image encoded as an I picture. Xp: A value representing the complexity of an image encoded as a P picture. Xb: A value representing the complexity of an image encoded as a B picture. Si: generated bit amount of an image encoded as an I picture. Sp: generated bit amount of an image encoded as a P picture. Sb: generated bit amount of an image encoded as a B picture.

Qi: Average quantization coefficient of an image encoded as an I picture. Qp: Average quantization coefficient of the image encoded as a P picture. Qb: Average quantization coefficient of an image encoded as a B picture. β: A constant (0 ≦ β) indicating the influence of the total sum of distortions of DC components on image quality. dc_distortion_i: Sum of distortion of DC component of I picture. dc_distortion_p: Sum of distortion of DC component of P picture. dc_distortion_b: Sum of distortion of DC component of B picture.

Then, the calculated values Xi, Xp and Xb
Based on the above, the same processing as in the first embodiment is executed to change the quantization step width of the quantization by the quantization means 14 and control the generated code amount.

In this way, when the code amount control is performed in consideration of the total sum of distortions of DC components, for example, the total sum dc_distortion_p of distortions of DC components of P picture is small,
Sum of distortion of DC component of B picture dc_distortion_b
In the case of a large faded image, the value Xp representing the complexity of the image encoded as a P picture is smaller than that in the conventional code amount control (described in Reference 1), and it is encoded as a B picture. The value Xb representing the image complexity becomes larger. Such a value X
The change in p and Xb serves to reduce the allocated bit amount Tp when the next screen is encoded as a P picture and to increase the allocated bit amount Tb when the next screen is encoded as a B picture. The remaining amount djp of the virtual buffer is small and the remaining amount djb of the virtual buffer for B picture is large. As a result, the quantization coefficient of the P picture is large and the quantization coefficient of the B picture is small. This control acts to cancel the adverse effects of the dead zone effect described above, so that it is possible to prevent an increase in the distortion difference between the P picture and the B picture with respect to a faded image or the like, and the restored image is displayed due to the flicker phenomenon. It is possible to solve the conventional problem of being unsightly and obtain a good restored image. Moreover, the moving picture compression apparatus of the present embodiment grasps the quantization error of each DCT coefficient by using the quantization error calculation means 73 and reflects the information in the code amount control. Distortion can be accurately estimated, and a better restored image can be obtained.

Furthermore, instead of obtaining the distortion of the restored image, the distortion of the DCT coefficient is calculated and reflected in the code amount control. Therefore, the same effect as the code amount control as described in Reference 2 is obtained. Can be realized with a simpler device.
It should be noted that it is preferable to select, as the device, either the first embodiment, which has a simple configuration, or the moving picture compression device, which can accurately estimate the distortion, according to the second embodiment. The method may include both of the steps performed in these examples. For example, two methods can be used properly according to the type of moving image.

Further, as in the above-mentioned embodiments, not only the non-zero coefficient of the DC component is reflected in the code amount control, but
It is possible to perform more effective code amount control by adaptively reflecting the coefficients of other orders according to the human visual characteristics. Further, even when another code amount control method is adopted instead of the code amount control method as described in Document 1, it is possible to obtain the same effect by reflecting the frequency of non-zero counting in the code amount control. It is possible.

For example, for the equations Tp and Tb, the following equation reflecting the number of non-zero coefficients can be adopted. Tp = max (R / (Np + (NbKpXb) / (KbXp)) × (1-α × (N_nonintra_dc_nonzero_p-N_nonintra_dc_nonzero-b)), bit_rate / (8 × picture_rate) .. (8 ') Tb = max (R / (Nb + (NpKbXp) / (KpXb)) × (1 + α × (N_nonintra_dc_nonzero_p-N_nonintra_dc_nonzero-b)), bit rate / (8 × picture_rate) .. (9 ')

Here, Xi: a value representing the complexity of an image encoded as an I picture. Xp: A value representing the complexity of an image encoded as a P picture. Xb: A value representing the complexity of an image encoded as a B picture.

N_nonintra_dc_nonzero_p: Proportion of non-zero coefficient of non-intra macroblock DC component of image encoded as P picture. Count value of the counting means 54. N_nonintra_dc_nonzero_b: Ratio of non-zero coefficient of non-intra macroblock DC component of image encoded as B picture. Count value of the counting means 54. α: A constant (0 that indicates the influence of the non-zero coefficient ratio on the image quality)
≤α). R: Amount of bits left in GOP.

Np: The number of P pictures that have not been encoded in the GOP. Nb: The number of B pictures that have not been encoded in the GOP. Kp: a constant indicating the ratio of the P picture quantization scale code with respect to the I picture quantization scale code. Kb: A constant indicating the ratio of the B picture quantization scale code with respect to the I picture quantization scale code.

Max {a, b}: A function that returns the larger value of a and b. In this case, the calculated allocation bit amounts Tp and Tb
Substitute in (10) to (13) to obtain the quantized coefficient. By doing this, for example, N_nonintra_dc_nonzero_p is large,
N_nonintra_dc_nonzero_b is the conventional code amount control (such as the one described in Reference 1) in the case of a small fade image.
Compared with, the allocated bit amount Tp when the next screen is encoded as a P picture is reduced, and the next screen is
The allocated bit amount Tb in the case of encoding as a picture can be increased, the quantization coefficient Qjp of the P picture becomes larger, and the quantization coefficient Qjb of the B picture becomes smaller. As a result, B picture distortion is reduced. Therefore, the distortion of the B picture and the distortion of the P picture whose distortion amount decreases due to the dead zone effect described above are equalized, and a good restored image without a flicker phenomenon can be obtained.

Further, as an example in which the ratio of non-zero coefficients is reflected in the rate control by changing the quantized coefficient Q,
The following formula can be given. Qjp = djp × 31 / (2 × bit rate / picture rate) × (1 + α (N_nonintra_dc_nonzero_p-N_nonintra_dc_nonzero_b)) .. (13) Qjb = djp × 31 / (2 × bit rate / picture rate) × (1- α (N_nonintra_dc_nonzero_p-N_nonintra_dc_nonzero_b)) .. (14)

Here, Qjp: the quantization coefficient of the j-th macroblock of the P picture. Qjb: Quantization coefficient of the jth macroblock of B picture. djp: A value indicating the virtual buffer remaining amount for each macroblock of the P picture. djb: A value indicating the virtual buffer remaining amount for each macroblock of the B picture. α: A constant (0 that indicates the influence of the non-zero coefficient ratio on the image quality)
≤α). With this, for example, N_nonintra_dc_nonzero_p
Is large and N_nonintra_dc_nonzero_b is small, in the case of a faded image, compared to the conventional code amount control (described in Reference 1), the quantization coefficient Qjp of the P picture
Can be made large and the quantization coefficient Qjb of the B picture can be made small, and the distortion of the B picture can be reduced. Therefore, the distortion of the B picture and the distortion of the P picture whose distortion amount decreases due to the dead zone effect described above are equalized, and a good restored image without a flicker phenomenon can be obtained.

Although MPEG is taken as an example in each of the above-mentioned embodiments, moving image compression methods other than MPEG may be used.
If the inverse quantization steps have different biases, based on the same technical idea as the above-described embodiment, the distribution of the quantized values according to the biases is obtained and reflected in the code amount control, and the inverse It is possible to realize code amount control according to the quantization step.

[0072]

According to the first aspect of the present invention, the code generation amount is controlled based on the data obtained by the orthogonal transform or the quantization thereof, that is, the distribution information of the transform coefficient of the orthogonal transform. , By reflecting the information of the orthogonal transform space (information of the spatial frequency domain) instead of the real space in the code amount control, the distortion amount can be estimated without calculating the distortion amount of the restored image, and the best image can be obtained at a constant bit rate. Can be obtained.

According to the second aspect of the present invention, the code generation amount is controlled according to the number of blocks in which the DC component of the orthogonal transform coefficient becomes non-zero after the quantization. By using the number of non-zero coefficients of the macroblock as a measure of quantization distortion, it is possible to obtain a particularly good image of the image of the frequency component that is most noticeable. Claim 3
According to the invention described in, the difference between the orthogonal transform coefficient obtained by dequantizing the orthogonal transform coefficient after quantization and the orthogonal transform coefficient before quantization is calculated, and the code amount based on the sum of the differences. Is controlled, the distortion can be accurately estimated without calculating the distortion of the image in the real space, and a good image can be obtained.

According to a fourth aspect of the present invention, data value counting means for counting the data value obtained as a result of the quantization by the quantization means or the data value obtained in the middle of the quantization,
Since the code amount control unit for controlling the generation amount of the encoded data in block units according to the counting result of the data value counting unit is provided, the coefficient information of the orthogonal transform space, not the real space, is reflected in the code amount control. Accordingly, it is possible to provide a low-cost moving image compression apparatus capable of estimating the distortion amount and obtaining the best image at a constant bit rate without calculating the distortion amount of the restored image.

According to the invention described in claim 5, there is provided means for discriminating the block attribute of the image in block units, and the data value corresponding to the specific block attribute is selectively selected based on the discrimination result of the block attribute. Since the counting is performed, it is possible to select only the non-intra blocks that have an effect by reflecting the counting result in the control of the code generation amount, and it is possible to perform efficient processing.

According to the sixth aspect of the present invention, the data value counting means has means for discriminating the transformation order of the orthogonal transformation coefficient, and corresponds to a specific transformation order based on the discrimination result of the transformation order. Since data values are selectively counted, it is possible to control the code amount and image quality more effectively by grasping the number of blocks in which the DC component and low frequency component of the orthogonal transform coefficient obtained by orthogonal transform are non-zero. It can be carried out.

According to the seventh aspect of the invention, the data value counting means determines whether or not the orthogonal transformation coefficient is zero value for the data value, and based on the determination result, the non-zero value is obtained. Since the corresponding specific data value is selectively counted, the number of blocks in which the noticeable DC (direct current) component or the low-frequency component becomes non-zero is grasped, so that the code amount is more effective. , The image quality can be controlled.

According to the eighth aspect of the present invention, the inverse quantization means for inversely quantizing the output of the quantization means, the orthogonal transform coefficient inversely quantized by the inverse quantization means, and the orthogonal transform before quantization. Orthogonal transform coefficient difference calculating means for calculating the difference with the coefficient, data value counting means for counting the output of the orthogonal transform coefficient difference calculating means, and the block unit of the encoded data based on the counting result of the data value counting means. Since the code amount control means for controlling the generation amount of is provided, the distortion can be accurately estimated without calculating the distortion of the image in the real space, and a good image can be obtained.

According to the invention described in claim 9, the data value counting means discriminates the transformation order of the orthogonal transformation, and selectively selects the data value corresponding to the particular transformation order based on the discrimination result of the transformation order. Since the number of blocks in which the DC component and the low frequency component of the orthogonal transform coefficient obtained by the orthogonal transform are non-zero is grasped, the code amount and the image quality can be controlled more effectively.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an example of an embodiment of a moving image compression apparatus according to the present invention.

FIG. 2 is a block diagram showing a configuration of a data value counting means of the first embodiment.

FIG. 3 is a block diagram showing a configuration of a code amount control means of the first embodiment.

FIG. 4 is a block diagram showing another embodiment of the moving picture compression apparatus according to the present invention.

FIG. 5 is a block diagram showing a configuration of a data value counting means in another embodiment.

FIG. 6 is a graph showing the relationship between the distortion due to quantization and the quantization count.

FIG. 7 is a graph showing a difference in distortion amount between a fade image and P and B pictures.

[Explanation of symbols]

1 Image Signal Input Terminal 2 Bit Stream Output Terminal 11 Blocking Means 12 Difference Image Generating Means 13 DCT Means (Orthogonal Transforming Means) 14 Quantizing Means 15 Variable Length Encoding Means 16 Buffers 17 Code Amount Counting Means 18 Quantization Coefficient Counting Means 21 inverse quantization means 22 inverse DCT means 23 restored image generation means 24 restored image storage means 31 motion vector search means 32 reference area extraction means 41,71 data value distribution counting means 42,72 code amount control means 51 discriminating block attributes Means 52 Means for discriminating the transformation order of orthogonal transformation 53 Means for discriminating whether or not the orthogonal transformation coefficient is zero value 72 Code amount control means 73 DCT coefficient quantization error means (orthogonal transformation coefficient difference calculating means)

 ─────────────────────────────────────────────────── ─── Continued Front Page (72) Inventor Toshifumi Sakaguchi 4-36-19 Yoyogi, Shibuya-ku, Tokyo Within Graphics Communication Laboratories, Inc. (72) Inventor Yusumi Wataya 4-chome Yoyogi, Shibuya-ku, Tokyo 36th-19th Stock Company, Graphics Communication Laboratories

Claims (9)

[Claims] 1. A moving image in which a moving image or a predictive difference image of a moving image is orthogonally transformed in a predetermined block unit, and an orthogonal transformation coefficient obtained as a result of the orthogonal transformation is quantized to generate coded data of the image. An image compression method, comprising the step of controlling the generation amount of the encoded data based on the data obtained by the orthogonal transformation or the quantization thereof. 2. The moving image according to claim 1, wherein the generation amount of the encoded data is controlled according to the number of blocks in which the DC component of the orthogonal transform coefficient is non-zero after the quantization. Compression method. 3. A moving image for generating coded data of the image by orthogonally transforming a moving image or a predictive difference image of the moving image in units of a predetermined block, and quantizing an orthogonal transform coefficient obtained as a result of the orthogonal transform. An image compression method, a step of calculating a difference between the orthogonal transform coefficient obtained by dequantizing the orthogonal transform coefficient after the quantization and the orthogonal transform coefficient before the quantization, and based on the sum of the differences. And a step of controlling the generation amount of the encoded data. 4. An orthogonal transformation means (13) for orthogonally transforming a moving image or a predictive difference image of the moving image in a predetermined block unit.
And a quantizer (14) for quantizing an orthogonal transform coefficient obtained as a result of the orthogonal transform, and a moving image for generating encoded data of the image by an encoding process including the orthogonal transform and quantization. In the compression device, a data value counting means (41) for counting a data value obtained as a result of the quantization by the quantization means (14) or a data value obtained in the middle of the quantization, and the data value counting means (41 ) Code amount control means (42) for controlling the generation amount of the encoded data according to the counting result of
And a moving picture compression apparatus. 5. The data value counting means (41) has means (51) for discriminating a block attribute of an image of the block unit with respect to the data value, and a specific block based on the discrimination result of the block attribute. 5. A data value corresponding to an attribute is selectively counted, wherein the data value is counted.
The moving image compression apparatus described in 1. 6. The data value counting means (41) has means (52) for discriminating a transformation order of the orthogonal transformation with respect to the data value, and based on the discrimination result of the transformation order, a specific transformation order is determined. The moving picture compression apparatus according to claim 3, wherein the corresponding data values are selectively counted. 7. The data value counting means (41) has means (53) for judging whether or not the orthogonal transformation coefficient is zero value for the data value, and the non-zero value is based on the judgment result. 7. The moving picture compression apparatus according to claim 4, wherein a specific data value corresponding to is selectively counted. 8. An orthogonal transformation means (13) for orthogonally transforming a moving image or a predictive difference image of the moving image in a predetermined block unit.
And a quantizer (14) for quantizing an orthogonal transform coefficient obtained as a result of the orthogonal transform, and a moving image for generating encoded data of the image by an encoding process including the orthogonal transform and the quantization. In a compression device, an inverse quantization means (22) for inversely quantizing the output of the quantization means (14), an orthogonal transform coefficient inversely quantized by the inverse quantization means (22), and an orthogonal transform before quantization. Orthogonal transform coefficient difference calculating means (73) for calculating the difference with the coefficient, data value counting means (71) for counting the output of the orthogonal transform coefficient difference calculating means (73), and counting by the data value counting means (71) Code amount control means (72) for controlling the generation amount of the encoded data based on the result
And a moving picture compression apparatus. 9. The data value counting means (71) has means (82) for discriminating the transformation order of the orthogonal transformation with respect to the data value, and based on the discrimination result of the transformation order, a specific transformation order is determined. 9. The moving picture compression apparatus according to claim 8, wherein the corresponding data value is selectively counted.