JPH06217279A - Image data processor - Google Patents

Abstract

PURPOSE:To improve picture quality deterioration at an edge part such as a character periphery and to control the amount of data after compression to a specific value in field or frame units. CONSTITUTION:The edge part is decided from orthogonally transformed data and while a quantization weight distribution is switched, a 1st quantizing circuit 7 performs quantization on the basis of the quantization weight distribution to calculate a quantization coefficient (QM). This quantization coefficient (QM) is multiplied by the quantization weight distribution switched by the edge part decision making and a 2nd quantizing circuit 13 quantizes delayed orthogonally transformed data to generate a variable length encoding output. Thus, the quantization is performed with the quantization weight distribution matching the features of an input image, so recording and transmission of high picture quality are enabled. Further, the total amount of encoded data in one frame or one frame of the input image can be suppressed below a certain value.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image data processing apparatus for compressing image data converted into a digital signal based on two-dimensional orthogonal transformation and variable length coding and performing recording and transmission. .

[0002]

2. Description of the Related Art When recording image data on a recording medium, data compression is used to efficiently use the storage capacity of the recording medium. For example, in a digital VTR, data compression enables recording / reproduction for a longer time with the same tape usage amount. Alternatively, it is possible to record a higher quality image with the same tape usage amount and recording / reproducing time.

A typical method for compressing these image data is a combination of two-dimensional orthogonal transformation and variable length coding. First, this compression method will be briefly described.

The image data is divided into blocks of about 8 × 8 samples and the discrete cosine transform (Discreat Cosine Trans
Two-dimensional orthogonal transformation such as form: DCT) is performed. The image data converted into a two-dimensional frequency component by this orthogonal transformation is quantized with a predetermined quantization coefficient corresponding to the two-dimensional frequency component, and then horizontal / vertical components are converted from the direct current and low frequency components where energy is likely to concentrate. Both are zigzag scanned into high frequency components and encoded by a variable length code such as Huffman coding.

Energy is concentrated on DC and low frequency components in a general natural image, and it is difficult to recognize deterioration of high frequency components both horizontally and vertically due to human visual characteristics. Quantization is performed roughly as compared to the mid and low frequencies. Furthermore, zigzag scanning is performed on each quantized frequency data, and the run length indicating the number of consecutive zero data and the level of the non-zero component are encoded by a variable length code such as Huffman code to realize data compression. ing.

In these compression methods, quantized data is compressed by variable-length coding, so that the total amount of data after compression (total amount of coded data) changes depending on the contents of the image. For example, in an image that contains a lot of high-frequency components such as edge information, the energy disperses into many frequency components in addition to DC and low-frequency components, so the probability that zero data will occur consecutively decreases, and the total code The amount of data increases. If the total coded data amount changes depending on the image content in this way, it becomes unclear how many images can be recorded in a storage medium with a limited recording capacity, or it is impossible to transmit a moving image at a predetermined transmission rate. Problems such as difficulty occur. Further, in a digital VTR or the like, if an image of one field or one frame can be always compressed to a fixed total code data amount, data can be written in a predetermined position of a recording medium in a unit of field or frame,
Variable speed reproduction can be easily realized, and editing can be performed in field or frame units.

In order to eliminate the above-mentioned inconveniences in the conventional compression system, the quantization step in quantizing the orthogonally transformed data is changed to keep the total code data amount after compression to a predetermined value. It was.

[0008]

However, in the conventional compression method, since the total code data amount after compression is suppressed to a predetermined value, coarse quantization is performed for a complicated image having a sharp edge such as a character. Therefore, there is a problem that the high-frequency component included in the original image is not completely reproduced and the edge portion is likely to be greatly deteriorated.

The deterioration of image quality at these edge portions can be improved by detecting a block containing an edge and quantizing the high frequency components more finely than the normal flat portion. However, you will have to quantize the complex image more finely,
The total code data amount after compression increases. That is, there is a problem that the total coded data amount after compression changes remarkably depending on the complexity of the image content, and it becomes difficult to keep the total coded data amount after compression to a predetermined value.

It is an object of the present invention to provide an image data processing apparatus capable of improving the image quality deterioration of the edge portion and controlling the total code data amount after compression to a predetermined value in field or frame units.

[0011]

In order to solve the above problems, the present invention determines the edge portion based on the AC component of the orthogonally transformed image data and switches the quantization weight distribution.

Further, based on the quantization weight distribution switched by the above determination, the first quantization circuit quantizes the quantization coefficient (QM) that makes the total code data amount in a field or frame unit a predetermined value. It is calculated.

Further, the calculated quantized coefficient (QM) is multiplied by the quantized weight distribution switched by the edge determination, and the image data orthogonally transformed by the second quantized circuit is quantized to have a variable length. This is encoded and output.

Further, the orthogonally transformed image data and the edge determination information which are input to the second quantization circuit only for the period of approximately one field or one frame required for calculating the above-mentioned quantization coefficient (QM). Is to delay.

Further, the variable length coded output is multiplexed with the edge determination information and the quantized coefficient (QM) and recorded or transmitted.

The quantized coefficient (QM) is multiplexed for each packet or sync block unit which is the minimum unit for recording and transmitting coded data.

[0017]

By determining the edge portion based on the AC component of the orthogonally transformed image data and switching the quantization weight distribution, the high frequency component of the edge portion can be quantized more finely than the normal flat portion.

Further, by increasing the amount of data by switching the quantization weight distribution, the first quantization circuit quantizes the quantization coefficient (QM) based on the quantization weight distribution switched by the edge determination. It is possible to calculate the quantization coefficient (QM) that includes and has a predetermined total code data amount in units of fields or frames.

Further, the calculated quantization coefficient (QM) is multiplied by the quantization weight distribution switched by the edge determination, and the image data orthogonally transformed by the second quantization circuit is quantized. , Compressed data that is quantized with a quantization weight distribution suitable for the edge portion and the flat portion and is controlled to have a predetermined total code data amount can be output.

Further, the orthogonally transformed image data and the edge part determination information input to the second quantization circuit are provided only for a period of approximately one field or one frame required to calculate the quantization coefficient (QM). By delaying, a quantized coefficient (QM) that provides a predetermined total code data amount is calculated using all data in the field or frame of the input image, and then the second quantized coefficient is calculated using this quantized coefficient (QM). The quantization can be performed by the digitizing circuit.

Furthermore, the block transmitted at the time of decoding is the block determined to be the edge portion by multiplexing or recording the edge portion determination information and the quantized coefficient (QM) on the variable length coded output and recording or transmitting the multiplexed information. Whether or not there is, and the value of the quantized coefficient at the time of encoding can be known without error.

Further, by multiplexing the quantized coefficient (QM) for each packet or sync block unit, which is the minimum unit for recording and transmitting, even when a transmission path error occurs during recording and transmitting, the quantized coefficient (QM) ) Error does not affect the entire screen.

[0023]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of a main part of a compression device for compressing image data in units of frames to a predetermined code amount according to an embodiment of the present invention.

In FIG. 1, reference numeral 1 is a block forming circuit which forms blocks of 8 samples in the vertical direction and 8 samples in the horizontal direction from input image data, and 2 is an orthogonal transform circuit for performing a discrete cosine transform on the block data of 8 × 8 samples, Reference numeral 3 is an edge discrimination circuit for discriminating whether or not the block includes an edge based on the AC coefficient of the image data converted into frequency components, and 4 is a first quantization weight for quantizing the block in the flat portion of the image. Distribution table, 5 is a second quantization weight distribution table for quantizing a block including an edge portion of an image, 6
Is a switching circuit for switching the quantization weight distribution table based on the edge discrimination information from the edge discrimination circuit 3, and 7 is a first for quantizing the image orthogonally transformed based on the quantization weight distribution table selected by the switching circuit 6. Quantization circuit, 8 is a coefficient calculation circuit for calculating a quantization coefficient QM for obtaining a predetermined code amount from quantized data, and 9 is a first
Delay circuit for delaying the edge discrimination information until the coefficient is calculated by the quantization circuit 7 and the coefficient calculation circuit 8
0 is a switching circuit for switching the quantization weight distribution table based on the delayed edge discrimination information, 11 is a multiplication circuit for multiplying the quantization weight distribution selected by the switching circuit 10 by the quantization coefficient QM, and 12 is a quantization coefficient QM. A delay circuit for delaying the image data that has been orthogonally transformed until is calculated, and a second delay circuit 13 that quantizes the delayed data based on the quantization weight multiplied by the quantization coefficient QM in the multiplication circuit 11. Quantization circuit, 14 is a variable length coding circuit for variable length coding by zigzag scanning the quantized data, 1
A multiplexing circuit 5 multiplexes the coded data, the edge discrimination information, and the quantized coefficient QM and outputs the multiplexed data.

Next, the operation of this embodiment will be described. The image data to be compressed is input to the blocking circuit 1 to form a block of 8 × 8 samples. When the input image signal is input in the form of raster scan on a line-by-line basis, it may be configured such that eight lines or more are written in the memory inside the blocking circuit 1 and then sequentially read out in block units. Further, when a 2: 1 interlaced video signal is input, it is possible to compose the video signal of two fields into an image of one frame in the memory inside the blocking circuit 1 and then sequentially read out in block units. Good. Alternatively, in the still image portion of the input image, the frame may be combined as described above to form a block, and in the moving image area, an 8 × 8 block may be formed from signals of eight lines adjacent in the field.

The image data having the block structure as described above is input to the orthogonal transform circuit 2 and converted into 8 × 8 frequency components by discrete cosine transform. The image data converted into the frequency component is input to the edge determination circuit 3, the first quantization circuit 7 and the delay circuit 12.

The edge discriminating circuit 3 discriminates the edge portion or the flat portion based on the AC coefficient of the image data converted into the frequency component, and the edge discriminating information is sent to the switching circuit 6 and the delay circuit 9. Output. Specifically, the AC coefficient of the frequency component of the image data (“6 excluding the DC term
The determination may be made based on whether or not the sum of absolute values of “3” data) exceeds a predetermined value. Alternatively, the determination may be made using the sum of absolute values of the components excluding the DC coefficient and the extremely low frequency component. In addition, in order to prevent malfunction due to noise superimposed on the input image, the determination may be performed by the sum of absolute values after coring processing in which an AC coefficient that does not reach a predetermined level is regarded as zero.

The switching circuit 6 selects the first quantization weight distribution table 4 for the flat portion when the orthogonally transformed block is the flat portion, based on the edge discrimination information from the edge discrimination circuit 3, and selects the edge. In the case of a block including a copy, the second quantization weight distribution table 5 for edges is selected and input to the first quantization circuit 7.

An example of the coefficient values of the first quantization weight distribution table 4 for the flat portion and the second quantization weight distribution table for the edge is shown in FIGS. 2 (a) and 2 (b), respectively. Shown in. In (a) and (b) of FIG. 2, the upper left corner of the figure is the quantization coefficient for the DC component, the right direction corresponds to the high frequency component of the horizontal frequency, and the down direction corresponds to the high frequency component of the vertical frequency. Is a conversion factor. Further, the larger the quantization coefficient, the coarser the frequency component is quantized. In FIG. 2A showing the first quantization weight distribution table 4 for the flat portion, in consideration of the degree of concentration of energy in a general image and the visual characteristics, the larger the horizontal / vertical frequency is, the larger the frequency becomes. Set to be a coefficient. 2B showing the second quantization weight distribution table 5 for the edge part.
Then, although the coefficient in the middle and low frequencies is the same as that for the flat portion, in order to improve the reproducibility of the high frequency component, the coefficient is set so as not to exceed the predetermined value even if the horizontal / vertical frequency increases.

The first quantization circuit 7 shown in FIG. 1 quantizes the image data converted into frequency components by the orthogonal transformation circuit 2 in accordance with the quantization weight distribution table selected by the switching circuit 6. This quantization process is configured to divide the data of each frequency component by the corresponding quantization table. When the flat part quantization weight distribution of FIG. 2A is selected by edge discrimination, the DC component of the image data is divided by 3 and the integerized operation result is used as the quantized output, and horizontal / vertical Both highest frequency components are divided by 20 and quantized.

The data quantized by the first quantization circuit 7 based on the edge discrimination information is input to the coefficient calculation circuit 8. At this time, the quantized data is zigzag scanned from the DC and low frequency components where the energy is likely to be concentrated to the horizontal and vertical high frequency components. FIG. 3 shows the scanning order of this zigzag scanning, in which data is sequentially scanned and output from the DC component "0" at the upper left to the highest frequency component "63" in the horizontal / vertical direction at the lower right in the numerical order.

In the coefficient calculation circuit 8, the first quantization circuit 7
At, the code amount after variable length coding is calculated for the data quantized by the predetermined quantization weight distribution, and it is determined whether or not the target total code data amount TD of one frame is less than or equal to TD. 1
When the total coded data amount of the frame exceeds the target value, the total coded data amount is controlled by further dividing (quantizing) each quantized value by the quantization coefficient QM.

The method of calculating the quantized coefficient QM will be described below. First, from the data output from the first quantization circuit 7 in the scan order of the zigzag scan, the code lengths at the time of variable length coding are cumulatively added to obtain the total code data amount Td ₍₁₎ of one frame of the input image. calculate. At the same time, in the coefficient calculation circuit 8, the quantized output from the first quantization circuit 7 is further divided by a power of 2 such as 2, 4, 8, 16, 32, ... In this case, the total code data amount Td ₍₂₎ , Td ₍₄₎ , T of one frame of the input image
d ₍₈₎ , Td ₍₁₆₎ , Td ₍₃₂₎ , ... Are calculated in parallel. The total code data amount Td ₍₁₎ when the output from the first quantization circuit 7 is subjected to variable length coding as it is is the target 1
When it is smaller than the total code data amount TD of the frame, that is, when Td ₍₁₎ ≤TD, the quantization coefficient QM is 1
Is output.

When Td ₍₁₎ > TD, the minimum code amount Td _(m) exceeding the target total code data amount TD and the maximum code amount not exceeding TD are selected from the above calculation results. Quantization coefficient QM is calculated by the linear approximation of the first order using the code amount Td _(n) of QM = m + (Td _(m) -TD ₎ * ₍ nm) / (Td _(m) -Td _(n) ) ... where Td _(m) >TD> Td _(n) , n> m≥1 Alternatively, the same is true even if it is calculated by the following formula. QM = n- (TD-Td _(n) ) ・ (nm) / (Td _(m) -Td _(n) ) ... where Td _(m) >TD> Td _(n) , n> m≥ 1 The quantized coefficient Q calculated by the coefficient calculation circuit 8 as described above.
M is input to the multiplication circuit 11 and the multiplexing circuit 15.

Quantization coefficient Q in these coefficient calculation circuits 8
To calculate M, it is necessary to obtain the total coded data amount when variable-length coding the data obtained by dividing the input image data of all one frame by each numerical value by cumulative addition. Processing time is required. Therefore, in the present embodiment, the image data converted into the frequency component by the orthogonal transform circuit 2 is delayed by about 1 frame using the delay circuit 12 and the edge determination information is delayed by about 1 frame using the delay circuit 9, respectively, and a predetermined target is obtained. After the quantization coefficient QM for realizing the code amount is calculated, the second quantization circuit 13 is used for quantization.

The edge discrimination information delayed by the delay circuit 9 indicates whether or not the image data converted into the frequency component input to the second quantization circuit 13 is a block including an edge. It is configured. As a result, similar to the operation of the first quantization circuit 7, the switching circuit 10 selects the first quantization weight distribution table 4 in the normal flat portion and the second quantization weight distribution table 5 in the edge portion.
Is selected.

In the second quantizing circuit 13, the image data converted into frequency components is a value obtained by multiplying the coefficient corresponding to each frequency of the quantizing weight distribution table selected by the switching circuit 10 by the quantizing coefficient QM. Quantize by. For example, when QM = 5.8, the DC component of the block in the flat portion is 5.8 × 3 = 17.
It is quantized by 4. The quantization coefficient QM is a value that has to be further divided in order to control the data quantized by the first quantization circuit 7 to a target code amount. Therefore, the orthogonally transformed data is quantized (divided) by the product of the coefficient value of the quantization weight distribution table and the quantized coefficient QM in the second quantization circuit 13, and encoded by the variable length coding circuit 14. As a result, the data can be controlled to the target code amount.

Further, the quantization weight distribution tables of the flat portion and the edge portion are set equal in advance between the compressed image decoding apparatus and the compression apparatus of the present embodiment,
It is not necessary to transmit these pieces of information, and the quantized coefficient QM may be recorded or transmitted by time-multiplexing the quantized coefficient QM at least for each frame and the edge determination information for each block by the multiplexing circuit 15. Since the quantization coefficient QM affects the entire screen when an error occurs due to a transmission path error, it may be multiplexed for each packet or synchronization block which is the minimum unit when recording and transmitting the compressed code data. With this configuration, even when a transmission path error occurs in the quantized coefficient QM, the influence on the decoded image can be minimized.

As described above, according to the present embodiment, since the quantization weight distribution suitable for the edge or flat portion of the input image is adaptively switched and used, it is possible to reduce the occurrence of distortion around the character. it can.

Further, since the edge portion is quantized finely by switching the quantization weight distribution and the code amount changes, the first quantum including the adaptive switching of the quantization weight distribution. Since the quantization coefficient QM for realizing the target predetermined code amount is determined for the converted output, the total code data amount of one frame can be controlled to the target value.

Since the quantized coefficient QM is determined to control the total code data amount of one frame to the target value for each frame, the quantized coefficient QM does not change within one frame. The inside of the screen can be quantized with a substantially constant quantization characteristic, and local deterioration does not occur.

Next, the configuration and operation of the coefficient calculation circuit 8 shown in FIG. 1 will be described in detail. FIG. 4 is a block diagram showing the configuration of the coefficient calculation circuit 8. In the figure, 16
Is an input terminal of the quantized data from the first quantization circuit 7 shown in FIG. 1, and 17 (17-1 to 17-5) is the number of consecutive zeros from the quantized data input by the zigzag scan. A data amount calculation circuit for obtaining the code length of the Huffman code by combination with a level of data other than zero and cumulatively adding it, and 18 is a data amount calculation circuit 17-1 to 17-5.
Data amount calculated in Td ₍₁₎ , Td ₍₂₎ , Td ₍₄₎ , T
This is a quantization coefficient calculation circuit for calculating the quantization coefficient QM based on d ₍₈₎ and Td ₍₁₆₎ .

The quantized data from the quantization circuit 7 is input to the terminal 16 of FIG. 4 in the form of an 11-bit digital signal in the form of a zigzag scan for each block. If the input data is zero, the data amount calculation circuit 17 counts the number of consecutive zeros (zero run) by the counter provided inside, and if the data other than zero is input, this data is calculated. The code length of the Huffman code generated using the ROM table is calculated from the zero run and the data level until the occurrence of, and the code length data is sequentially cumulatively added.

When the 11-bit data from the terminal 16 is directly input to the first data amount calculation circuit 17-1 and is quantized by the first quantization circuit 7 with a predetermined quantization weight distribution. The total code data amount Td _{(1) of} is calculated. Second
The data amount calculation circuit 17-2 of
Of the bit data, 10-bit data from the MSB (1 bit shift to the lower order, LSB 1-bit truncation) is input, and the total code data amount Td ₍₂ when the data from the first quantization circuit 7 is divided by _{2 )} Is calculated. Further, the MSB 9-bit data (lower 2 bits shift) of the 11-bit data is input to the third data amount calculation circuit 17-3, and the data from the first quantization circuit 7 is converted into 4 bits.
The total code data amount Td ₍₄₎ when divided by is calculated. Furthermore, the fourth and fifth data amount calculation circuits 17-4, 17-
Similarly for 5, the total code data amount Td when dividing by 8 using MSB 8bit data (lower 3bit shift)
₍₈₎ Calculate the total code data amount Td ₍₁₆₎ when divided by 16 using 7-bit MSB data (lower 4-bit shift).

In the configuration example of the coefficient calculation circuit shown in FIG. 4, n bits are shifted downward and n on the LSB side is used.
Although the bit is rounded down for processing, it may be rounded for rounding. By doing so, there is an effect that the quantization coefficient QM can be calculated accurately.

These calculated data amounts Td ₍₁₎ , Td
₍₂₎ , Td ₍₄₎ , Td ₍₈₎ and Td ₍₁₆₎ are input to the quantized coefficient calculation circuit 18, and the quantized coefficient QM is calculated according to the above equation or equation.

The process of calculating these quantized coefficients QM is shown in FIG.
Will be explained. In the graph of FIG. 5, the horizontal axis represents the quantization coefficient and the vertical axis represents the total code data amount. According to the calculation results of the data amount calculation circuits 17-1 to 17-5, the total code data amount of the quantization coefficients 1, 2, 4, 8, 16 is Td _{(1) to} Td.
_It can be plotted as ₍₁₆₎ . In this operation example, the minimum code amount that exceeds the target total code data amount TD is Td ₍₄₎
Therefore, the maximum code amount that does not exceed TD is Td ₍₈₎ .
As shown in FIG. 5, by linear approximation using these two points,
Quantization coefficient QM for controlling the total code data amount to TD
You can decide. Since the quantized coefficient QM calculated in this way is a linear approximation of the original characteristic shown by the broken line in FIG. 5, an error actually occurs. However, it is generally known from the rate / distortion theory that the quantized coefficient versus total coded data amount characteristic of FIG. 5 is also downwardly convex, and is actually obtained by using the quantized coefficient QM determined by linear approximation. The total code data amount Td _(QM) does not exceed at least the target total code data amount TD, and the input image can be controlled to a predetermined code data amount.

Although the quantization coefficient QM is calculated by the five data amount calculation circuits 17 in the configuration example of the coefficient calculation circuit shown in FIG. 4, the present invention is not limited to this, and the quantization amount QM may be calculated by two or more data amount calculation circuits. You may comprise. When the configuration is made up of two data amount calculation circuits, even for the most complicated image assumed as an input image, the first bit shift amount that is less than or equal to the target total code data amount Even for an easy image, two bit shift amounts of the second bit shift amount that are equal to or more than the target total code data amount are obtained in advance, and the data amount is calculated after performing these two bit shift processes. One data amount calculation circuit may be provided and the quantization coefficient QM may be calculated. By setting in this way, in a normal image, the data amount after the first bit shift is smaller than the target total code data amount, and the data amount after the second bit shift is the target total code amount. Since it is larger than the data amount, the quantization coefficient QM can be calculated from the ratio of the total code data amount and the target code amount of both. Further, when three or more data amount calculation circuits are provided, it is possible to add a data amount calculation circuit for calculating the data amount after performing the bit shift processing between the first and second bit shift amounts. Good. By providing three or more data amount calculation circuits in this way, the quantization coefficient QM can be calculated more accurately.

As described above, in the present embodiment, by selecting the quantization weight distribution suitable for the edge portion or the flat portion of the input image, it is possible to reduce the image quality deterioration, and moreover, it is possible to reduce the deterioration of the image quality in units of fields or frames. Since the total amount of coded data after compression can be controlled to be equal to or less than the value of, it is possible to transmit a moving image at a constant bit rate with less deterioration of the edge portion and the like. Further, in the digital VTR or the like, by recording the data compressed in the unit of field or frame at a predetermined position of the recording medium, it is possible to easily improve the image quality at the time of variable speed reproduction or easily realize the editing in the unit of field or frame.

Next, another embodiment of the present invention will be described with reference to the block diagram of FIG. FIG. 6 is an embodiment of a compression device for compressing image data in a predetermined code amount in frame units, as in the embodiment shown in FIG.

In FIG. 6, 1 is a block forming circuit which forms blocks of 8 samples in the vertical direction and 8 samples in the horizontal direction from the input image data, 2 is an orthogonal transform circuit for performing discrete cosine transform on the block data of 8 × 8 samples, Reference numeral 3 is an edge discrimination circuit for discriminating whether or not the block includes an edge based on the AC coefficient of the image data converted into frequency components, and 4 is a first quantization weight for quantizing the block in the flat portion of the image. Distribution table, 5 is a second quantization weight distribution table for quantizing a block including an edge portion of an image, 6
Is a switching circuit for switching the quantization weight distribution table based on the edge discrimination information from the edge discrimination circuit 3, and 7 is a first for quantizing the image orthogonally transformed based on the quantization weight distribution table selected by the switching circuit 6. , A coefficient calculation circuit 8 for calculating a quantization coefficient QM for obtaining a predetermined code amount from the quantized data, and 9 until the coefficients are calculated by the quantization circuit 7 and the coefficient calculation circuit 8. , A delay circuit that delays the edge discrimination information, a delay circuit that delays the quantized data until the quantized coefficient QM is calculated, and a quantizer 13 that quantizes the delayed data based on the quantized coefficient QM. Second quantizer circuit, 1
Reference numeral 4 is a variable-length coding circuit that zigzag-scans the quantized data to perform variable-length coding, and 15 is a multiplexing circuit that multiplexes the coded data, the edge discrimination information, and the quantized coefficient QM and outputs the multiplexed data. .

In the embodiment shown in FIG. 1, the output of the orthogonal transformation circuit 2 is delayed by the delay circuit 12 and the second quantization circuit 13 determines the quantization weight distribution selected by the quantization coefficient QM and the edge information. It is configured to be quantized by the product of coefficients. On the other hand, in the configuration of this embodiment of FIG. 6, the output of the orthogonal transform circuit 2 is quantized by the quantized weight distribution coefficient selected by the edge information in the first quantizer circuit 7, and then further quantized. It is configured to be quantized again by the coefficient QM. In the configuration of FIG. 1, the multiplication circuit 1 multiplies the product of the quantization weight distribution coefficient selected by the edge information and the quantization coefficient QM for controlling the predetermined total code data amount.
1 is obtained in advance and is quantized at once by the second quantization circuit 13, whereas in the configuration of the present embodiment shown in FIG. Quantization coefficient Q for performing quantization with the selected quantization weight distribution coefficient and further controlling to a predetermined total code data amount
The basic operation is the same as that of the above-described embodiment of FIG. 1, although the second quantization circuit 13 performs quantization depending on M.

With the configuration shown in FIG. 6, since the data input to the second quantization circuit 13 has already been quantized based on the edge information, the quantization coefficient for controlling the total code data amount. It is sufficient to quantize only with QM. Therefore, the multiplication circuit 11 and the switching circuit 1 shown in FIG.
0 is not necessary and has the effect of reducing the circuit scale. The data further delayed by the delay circuit 12 is the first quantization circuit 7
Since the data is quantized by, the required number of bits of data can be reduced, and the memory capacity of the delay circuit 12 can be reduced.

On the other hand, the configuration shown in FIG. 1 is different from the configuration shown in FIG.
By sufficiently increasing the calculation accuracy of 1, there is an effect of preventing the accumulation of calculation errors due to quantization.

In the configuration shown in FIG. 1, the image data orthogonally transformed by the orthogonal transformation circuit 2 is input to the second quantization circuit 13 after being delayed by the delay circuit 12, but the image data is orthogonally transformed. Instead of delaying, a second orthogonal transformation circuit may be provided to perform orthogonal transformation, and the result may be input to the second quantization circuit 13. The required number of bits (about 12 bits) of the orthogonally transformed data is the required number of bits of the original image (8bi
t), the memory capacity of the delay circuit 12 is reduced.

Next, in the multiplexing circuit 15 shown in FIGS. 1 and 6, how the encoded data of the image and the additional information such as the edge discrimination information and the quantization coefficient QM are multiplexed will be described with reference to FIG. Will be explained.

FIG. 7 shows compressed image data of length 204
This is a data structure when divided into 8-bit packets for transmission. 7A is a packet at the head of the frame, and FIG.
(B) is the next packet following (a).

A block address (12 bits), a start address (11 bits), and a quantized coefficient QM (7 bits) are transmitted to the first 30 bits of the packet, followed by edge discrimination information W _n (for each block). 1 bit), variable length code data VLC _n (length is indefinite) is transmitted, and data exceeding the packet length is transferred to the next packet. Note that W _n and VLC _n are edge discrimination information and variable length code data of the nth block.

The block address is additional information indicating the position within the frame of the image block first transmitted in this packet. Since (a) of FIG. 7 is the packet at the beginning of the frame, this block address is “0”.
(B), the remaining VL of the variable-length code data of block 2
"3" indicating a block to be newly transmitted following C ₂ '
Becomes

The start address is additional information indicating the start position in the packet of the valid data of the image block transmitted first in this packet. In the packet of FIG. 7A, since the data of W ₀ and VLC ₀ are transmitted from the 30th bit, the start address is “30”, and in FIG. 7B, the variable-length code data of block 2 is transmitted. Rest V
Since W ₃ and VLC ₃ of the block 3 are newly started from the 32nd bit subsequent to LC ₂ ′, the start address becomes “32”.

The inverse quantization characteristic of the data transmitted with the above-mentioned data structure is determined by the quantization circuit QM and the edge discrimination information W _n by the decoding circuit, and the variable length code data VLC _n is decoded, The compressed and transmitted image can be decoded.

Further, even when a specific packet is lost due to a transmission path error or the like, the block indicated by the block address can be sequentially decoded from the position of the start address transmitted at the beginning of the packet, and the influence of the transmission path error or the like can be obtained. Has the effect of minimizing the

The data structure of this example shown in FIG.
Block address, start address, quantization coefficient QM
The order and the number of bits are not limited to this, as long as they are multiplexed at a predetermined position in the packet,
Even when other additional information is multiplexed, it is in accordance with the gist of the present invention.

In the digital VTR or the like, the block address, start address, quantization coefficient QM, and edge discrimination information W are included in the sync block divided by a specific sync pattern, as in the data structure in the packet in the above transmission example. _It may be configured to record _n and variable length code data VLC _n . With this configuration, it is possible to obtain a partially reproduced image by knowing the block address and the start position of the data from the data at the head of the synchronous block that is intermittently reproduced even during variable speed reproduction.
As described above, according to the present invention, there is an effect that a favorable variable speed reproduction image can be obtained relatively easily.

Further, in the above-described embodiment, the edge discrimination information W _n and the variable length code data VLC _n are sequentially recorded and transmitted in block units. For example, four blocks are collectively W ₁ , W ₂ , W. ₃ , W ₄ , VLC ₁ , VL
C _2, VLC _3, may be recorded transmission as VLC _4.
Alternatively, the luminance signal Y and the two types of color difference signals R and B are combined into YW ₁ , YW ₂ , YW ₃ , YW ₄ , RW ₁ , BW ₁ , and YVL.
C ₁ , YVLC ₂ , YVLC ₃ , YVLC ₄ , RVLC ₁ ,
Recording and transmission may be performed collectively as in BVLC ₁ . When processing multiple blocks together as described above, configure the blocks that are close to each other on the screen to be in the same group, and perform the differential processing of the DC coefficient of orthogonal transformation within the group and transmit it. May be. By doing so, the level of the DC coefficient can be reduced by utilizing the correlation between the adjacent blocks, which has the effect of increasing the data compression efficiency.

[0066]

As described above, according to the present invention, the quantization weight distribution suitable for the edge portion and the flat portion of the input image is used to reduce the distortion that is likely to occur around the characters, There is an effect that it becomes possible to perform recording transmission with higher image quality.

Further, by calculating a quantization coefficient (QM) including an increase in data due to switching of the quantization weight distribution and having a predetermined total code data amount in field or frame units, one frame of the input image or 1 The total coded data amount of the field can always be kept below a certain value, and moving images can be transmitted at a certain transmission rate. In digital VTRs, etc.
The input image can always be recorded in a predetermined recording position in units of one frame or one field, and there is an effect that the image quality during variable speed reproduction is improved and editing can be performed in units of fields or frames.

Further, since this quantized coefficient QM is determined to control the total coded data amount of one frame or one field to the target value for each one frame or one field, the quantized coefficient QM is within one screen. It does not change, and the inside of the screen can be quantized with a substantially constant quantization characteristic, which has the effect of preventing local deterioration.

Further, the block transmitted at the time of decoding is the block determined to be the edge portion by multiplexing or recording the edge determination information and the quantized coefficient (QM) on the variable length coded output and recording or transmitting the multiplexed information. Whether or not there is, and the value of the quantized coefficient at the time of encoding can be known without error, and the decoding process can be simplified. In addition, the quantization coefficient (Q
M) may be multiplexed and recorded and transmitted at least in each field or frame, and the edge determination information may only need to transmit 1 or 2 bits of information for each block, which has the effect of improving the efficiency of recording and transmission. .

Further, by multiplexing the quantized coefficient QM for each packet or synchronization block, which is the minimum unit for recording and transmitting compressed code data, even when a channel error occurs in the quantized coefficient QM. The influence on the decoded image can be minimized.

[Brief description of drawings]

FIG. 1 is a block diagram showing an image data compression device according to an embodiment of the present invention.

FIG. 2 is an explanatory diagram showing an example of quantized weight distribution coefficients of first and second weight distribution tables 4 and 5 shown in FIG.

FIG. 3 is an explanatory diagram showing an output order of data output by zigzag scanning from the quantization circuits 7 and 13 shown in FIG.

4 is a configuration diagram showing an example of a specific configuration of a coefficient calculation circuit 8 shown in FIG.

5 is a graph diagram for explaining the operation of the coefficient calculation circuit 8 shown in FIG.

FIG. 6 is a block diagram showing an image data compression device according to another embodiment of the present invention.

7 is an explanatory diagram showing an example of a data configuration in a packet transmitted by the multiplexing circuit 15 shown in FIGS. 1 and 6. FIG.

[Description of Codes] 1 Blocking circuit 2 Orthogonal transformation circuit 3 Edge discrimination circuit 4 First quantization weight distribution table 5 Second quantization weight distribution table 6 Switching circuit 7 First quantization circuit 8 Coefficient calculation circuit 9 Delay circuit 10 Switching circuit 11 Multiplication circuit 12 Delay circuit 13 Second quantization circuit 14 Variable length coding circuit 15 Multiplexing circuit 17 Data amount calculation circuit 18 Quantization coefficient calculation circuit

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Kazuhiko Yoshizawa, 292 Yoshida-cho, Totsuka-ku, Yokohama-shi, Kanagawa Inside the Hitachi Media Visual Media Laboratory (72) Inventor, Miyoko Yoshikoshi, 292 Yoshida-cho, Totsuka-ku, Yokohama, Kanagawa Hitachi, Ltd. Visual Media Research Laboratory (72) Inventor, Yuzumi Wataya, 292 Yoshida-cho, Totsuka-ku, Yokohama, Kanagawa Prefecture

Claims (4)

[Claims] 1. An image data processing device for processing input image data based on orthogonal transform and variable length coding, means for dividing the input image data into predetermined blocks, and image data divided into the blocks. Orthogonally transforming the frequency data into frequency data, determining the edge portion based on the frequency data and generating discrimination information, and switching the quantization weight distribution based on the discrimination information to quantize the frequency data. First quantizing means, means for generating a quantized coefficient (QM) for compressing data quantized by the first quantizing means to a predetermined data amount for each frame or field, and the frequency Means for delaying the data and the discrimination information for a predetermined period, and the delayed discrimination data and the quantized coefficient (QM) for the delayed frequency data. Second quantizing means for quantizing based on the discrimination data; means for coding the data quantized by the second quantizing means to generate code data; and the discrimination information and the quantization coefficient (QM) in the code data. ) And a means for outputting the multiplexed data, and an image data processing apparatus. 2. The means according to claim 1, wherein the means for discriminating an edge portion based on the frequency data and generating discrimination information is means for generating an absolute value of an AC coefficient of the frequency data, and a predetermined value based on the absolute value. An image data processing apparatus characterized by including a means for extracting a coefficient equal to or higher than a level, a means for adding the extracted coefficients for each block, and a means for generating discrimination information based on the added value. . 3. The means for generating a quantized coefficient (QM) for compressing a predetermined amount of data for each frame or field from the data quantized by the first quantizing means according to claim 1. Means for bit-shifting the data quantized by the first quantizing means with at least two different bit shift amounts, and calculating the respective code data amounts; and a data amount that is the target of the code data amount from the bit shift amount. Means for obtaining the maximum bit shift amount exceeding the above, a means for obtaining the minimum bit shift amount from which the code data amount does not exceed the target data amount from the above bit shift amount, the above maximum bit shift amount and the above minimum bit amount And a means for generating a quantized coefficient (QM) based on the shift amount, each coded data amount, and the target data amount. Image data processing apparatus according to symptoms. 4. The minimum unit according to claim 1, wherein the means for multiplexing the discrimination information and the quantization coefficient (QM) on the code data and outputting the code data has a predetermined length for recording and transmitting the code data. Means for multiplexing the quantized coefficient (QM) for each minimum unit of recording / transmission of the code data, and for each minimum unit of recording / transmission of the code data of the block first recorded / transmitted in this minimum unit. And a means for multiplexing the start position of the code data of the block which is first recorded and transmitted in the minimum unit for each minimum unit of the record transmission of the code data. Image data processing device.