JPH0556270A - Reverse quantizing method and image data restoring device - Google Patents

Abstract

PURPOSE:To decrease the wasteful initializing processing by considering the distribution of an effective coefficient at respective quantizing coefficient matrixes of two continuous blocks concerning a reverse quantizing method when an original image is restored from code data by an orthogonal converting system. CONSTITUTION:In a reverse quantizing method to obtain the coefficient matrix from a quantizing coefficient obtained by decoding the code data by an orthogonal converting coding system, at the quantizing coefficient matrix respectively corresponding to respective blocks and the block before them, the difference of the distribution of the effective coefficient having a value except zero is discriminated, the quantizing threshold respectively corresponding to the effective coefficient included in the quantizing coefficient matrix corresponding to respective blocks is multiplied, and the reverse quantizing is selectively performed. The obtained reverse quantizing result is respectively held in accordance with the effective coefficient address, the initializing processing to the equivalent address is performed in accordance with the difference of the distribution of the effective coefficient with the previous block and the coefficient matrix corresponding to respective blocks is restored.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a dequantization method for dequantizing a quantized coefficient obtained by decoding code data in an image data decompression device for decompressing an original image from code data, and the dequantization method. The present invention relates to an image data restoration device to which the method is applied.

As a coding method for compressing the data amount of a multi-valued image such as a halftone image or a color image without deteriorating its characteristics, an adaptive discrete cosine transform coding method utilizing two-dimensional orthogonal transform (Adaptive Discrete Cosine). Transfo
rm, hereinafter referred to as ADCT method) is widely used.

In this ADCT method, a multi-valued image is divided into blocks each having a predetermined number of pixels (for example, 8 × 8 pixels), and the image data is orthogonally transformed for each block to obtain transform coefficients (hereinafter referred to as DCT coefficients). The data amount is compressed by quantizing each element of the matrix using a corresponding visual adaptation threshold value (described later) and then performing variable length coding.

[0004]

2. Description of the Related Art FIG. 11 shows the configuration of an image data compression apparatus to which a conventional ADCT method is applied. In addition, in FIG.
The example of the block obtained by dividing a multi-valued image is shown.

A multi-valued image read by an image reading device or the like is sequentially input to the DCT transform unit 611 for each block described above, and the two-dimensional discrete cosine transform (hereinafter referred to as DCT transform) by the DCT transform unit 611. By the processing, it is converted into a matrix of 8 rows and 8 columns (hereinafter, referred to as DCT coefficient D) including DCT coefficients corresponding to the spatial frequency components. FIG. 13 shows an example of the DCT coefficient D.

Each component of the DCT coefficient D is quantized by the linear quantizer 620 using the corresponding component of the quantization threshold value Q _TH . The above-mentioned quantization threshold Q _TH is
It is obtained from the visual adaptation threshold value corresponding to each spatial frequency and the quantization control parameter SF. This visual adaptation threshold value is determined in advance based on an experimental result regarding visual sensitivity to each spatial frequency component, and is given as a quantization matrix V _TH (see FIG. 14). Further, the quantization control parameter SF is a coefficient that determines the quantization accuracy of the image, depending on the image quality required for the restored image,
It is set by the operator prior to the encoding process of the image data for one screen.

Here, the value of each component of the above-mentioned quantization matrix V _TH is, as shown in FIG. 14, the absolute value of the component corresponding to a low spatial frequency, depending on the spatial frequency characteristic of human visual sensitivity. On the contrary, the absolute value of the component corresponding to the high spatial frequency is set to a large value. Therefore, the quantized coefficient D _QU obtained by quantizing the DCT coefficient D by the linear quantizer 620 is, as shown in FIG. 15, a component at the upper left corner of the matrix indicating the DC component (hereinafter referred to as the DC component).
In many cases, only a very small number of AC components around the DC component showing low spatial frequency components are effective coefficients having a value other than zero, and most AC components are invalid coefficients having a value of zero. ..

When the quantized coefficient D _QU obtained in this way is scanned using a scanning order called zigza scan shown in FIG. 16, a one-dimensional array in which a series of effective coefficients is followed by invalid coefficients is formed. Is obtained. The one-dimensional array is converted by the encoding unit 631 into a combination of the effective coefficient (index) and the continuous length (run) of the invalid coefficient that is continuous before this index, and each combination is converted based on the code table 632. The image data is compressed by replacing each with a code corresponding to the appearance frequency and performing variable length coding.

The coded data thus obtained is restored to image data by the image data restoration device shown in FIG. Decoding unit 711 of image data restoration device
Contrary to the above-mentioned code table 632, is provided with a decoding table 712 which indicates a combination of a run and an index corresponding to the code, and decodes sequentially input codes to obtain a combination of an index and a run. Input to the inverse quantizer 720.

In the inverse quantizing unit 720, the array restoring unit 721 restores the above-mentioned one-dimensional array from the input index and run, and the respective components of this one-dimensional array are sequentially input to the multiplier 722. It Further, the quantization threshold value holding unit 723 holds the quantization threshold value Q _TH , and the corresponding quantization threshold value is sent to the multiplier 722 in synchronization with the input of the quantization coefficient to the multiplier 722. There is. In this way, the multiplier 723 dequantizes each component of the quantized coefficient D _QU using the corresponding quantization threshold, and D
The CT coefficient D is restored.

Next, the DCT coefficient D is input to the inverse DCT conversion unit 731, and the inverse DCT conversion unit 731 performs the inverse D.
The image data of the corresponding block is restored by performing the CT conversion process.

[0012]

By the way, in the above-mentioned conventional method, the inverse quantization processing is performed by multiplying all the quantization coefficients by the quantization threshold value. However, the multiplication process of the invalid coefficient having a value of zero and the quantization threshold is a useless process in which the multiplication result is known in advance. That is, conventionally, 64 times of multiplication processing is performed for the dequantization processing of the quantized coefficient D _QU of one block, regardless of whether it is useless processing. It took a long time to process.

The present applicant has already filed Japanese Patent Application No. 1-316708 "Image Data Restoration Method" as a technique for speeding up inverse quantization processing by eliminating such wasteful processing. .. FIG. 18 shows a schematic configuration of an inverse quantization processing unit according to this technique.

According to this technique, the address calculation unit 724 obtains the address of the corresponding effective coefficient from the value of the input run, and the quantization threshold holding unit 723 causes the multiplier 722 to obtain the quantization threshold value corresponding to this address. By sending out and multiplying this quantization threshold value and index by the multiplier 722, only effective coefficients are selectively dequantized.
In this case, the multiplier 72 corresponds to the above-mentioned address.
By holding the multiplication result by 2 in the DCT coefficient holding unit 725, the DCT coefficient D of 8 rows and 8 columns is restored.

In this technique, in the inverse quantization processing of each block, the inverse quantization result is written only in the address corresponding to the effective coefficient of the DCT coefficient holding unit 725, so that
Before performing the inverse quantization process for the next block, it is necessary to perform an initialization process for clearing the DCT coefficient obtained by the inverse quantization process for the previous block. For example, the DCT coefficient holding unit 7
The initial value “0” may be written in all 64 addresses of 25.

However, since the content of the address corresponding to the invalid coefficient of the previous block is zero, writing zero again by the initialization process is a wasteful process. Further, the address corresponding to the effective coefficient of the quantized coefficient D _QU of the next block is updated by writing the applicable effective coefficient at the time of the inverse quantization process of the next block.
It is not necessary to initialize these addresses in advance.

That is, regardless of whether the distributions of the effective coefficients in the quantized coefficient D _QU corresponding to two consecutive blocks match or do not match, the initialization processing is performed for all 64 addresses. Due to the time required for the unnecessary initialization processing, the time shortening effect of the inverse quantization processing due to the reduction of unnecessary multiplication processing is lost.

The present invention reduces wasteful multiplication processing and, at the same time, makes it possible to reduce wasteful initialization processing in consideration of the distribution of effective coefficients in the quantization coefficient matrix of each of two consecutive blocks. An object is to provide a method and an image data restoration device.

[0019]

FIG. 1 shows the principle of the inverse quantization method of the present invention. According to the invention of claim 1, the original image is orthogonally transformed for each block including a plurality of pixels,
Decode code data when restoring an original image from code data by an orthogonal transform coding method that encodes a quantized coefficient matrix obtained by quantizing each component of a coefficient matrix composed of transform coefficients with a corresponding quantization threshold In a dequantization method that obtains a coefficient matrix from the quantized coefficients obtained by, the difference between the distributions of effective coefficients having non-zero values in the quantized coefficient matrix corresponding to each block and the block before it is determined. Then, the effective coefficient included in the quantized coefficient matrix corresponding to each block is multiplied by the corresponding quantization threshold to selectively dequantize, and the effective coefficient is displayed according to the effective coefficient address indicating the position of the effective coefficient in the quantized coefficient matrix. Then, the obtained dequantization results are held respectively, and the initialization processing for the corresponding address is performed according to the difference in the distribution of effective coefficients from the previous block. Te, characterized in that to restore the coefficient matrix corresponding to each block.

FIG. 2 shows the configuration of the image data restoration device according to the second and third aspects. According to a second aspect of the present invention, a quantized coefficient matrix obtained by orthogonally transforming an original image for each block including a plurality of pixels and quantizing each component of a coefficient matrix including transform coefficients with a corresponding quantization threshold is obtained. Decoding means 101 for continuously inputting coded data by an orthogonal transform coding method for coding, decoding this coded data, and sequentially outputting decoded data representing a quantized coefficient matrix corresponding to each block, and decoding An address calculating unit that calculates, based on the data, an effective coefficient address indicating a position in the quantized coefficient matrix of an effective coefficient having a value other than zero included in the quantized coefficient matrix of each block, and outputs it as the current block address. 111, an inverse quantization unit 112 for inverse quantization by multiplying an effective coefficient by a quantization threshold value indicated by a corresponding current block address, and an inverse quantization Means 1
Transform coefficient storage means 113 for storing the result of dequantization by 12 as a transform coefficient in the current block address, and the current block address are sequentially held, and the effective coefficient address held in the processing of the previous block is the previous block address. Based on the previous block address and the current block address, the current block invalid address that corresponds to the invalid coefficient in each block and the valid coefficient in the previous block is determined. Means 115, an initialization means 116 for writing an initial value to the storage location of the transform coefficient storage means 113 indicated by the current block invalid address for initialization, and a coefficient matrix composed of the transform coefficients stored in the transform coefficient storage means 113. Inverse orthogonal transform means 102 for performing inverse orthogonal transform to restore image data is provided. Characterized in that was.

According to a third aspect of the present invention, in the image data restoration device according to the second aspect, the conversion coefficient storage means 113 is
It is characterized in that a plurality of areas having a capacity corresponding to a coefficient matrix for one block are switched to store a coefficient matrix corresponding to consecutive blocks.

FIG. 3 shows the configuration of the image data restoration device according to claims 4 and 5. According to a fourth aspect of the present invention, in the image data restoration device according to the second aspect, the DC block represented by only the DC component is detected based on the number of effective coefficients included in the quantization coefficient matrix corresponding to each block. Then
The present invention is characterized by including a block detection means 121 for providing the detection result to the inverse orthogonal transformation processing by the inverse orthogonal transformation means 102.

According to a fifth aspect of the present invention, in the image data restoration device according to the second aspect, an invalid block that does not include an effective coefficient is determined based on the number of effective coefficients included in the quantized coefficient matrix corresponding to each block. The present invention is characterized by including block detection means 131 for detecting and providing the detection result to the inverse orthogonal transformation processing by the inverse orthogonal transformation means 102.

[0024]

According to the invention of claim 1, only the effective coefficients of the quantized coefficient matrix corresponding to each block are selectively dequantized and held at the corresponding effective coefficient addresses, and consecutive 2
The initialization process can be performed while considering the distribution of effective coefficients in the quantized coefficient matrix of one block. As a result, it is possible to reduce wasteful initialization processing when the valid coefficient address matches with that of the previous block or when the address of the previous block corresponds to the invalid coefficient.

According to a second aspect of the present invention, the address calculation means 1 is based on the decoded data obtained by the decoding means 101.
The effective coefficient included in the quantized coefficient matrix of each block is selectively dequantized by 11 and the inverse quantization means 112, and stored in the transform coefficient storage means 113 in accordance with the effective coefficient address. Further, the address holding means 114 and the discrimination means 115 determine the invalid address of the current block, and the initialization means 116 writes the initial value to the storage location indicated by the invalid address of the current block of the conversion coefficient storage means 113. The coefficient matrix corresponding to each block can be restored and used for the inverse orthogonal transform process by the inverse orthogonal transform unit 102.

This makes it possible to omit wasteful initialization processing in consideration of the distribution of effective coefficients in the quantized coefficient matrix of each of two consecutive blocks. According to the third aspect of the present invention, the process of storing the transform coefficient in any of the plurality of regions of the transform coefficient storage unit 113 and the inverse orthogonal transform process for the coefficient matrix stored in the other region can be performed in parallel. it can.

According to a fourth aspect of the invention, the block detecting means 12 is provided.
Since the DC block is detected by 1, the inverse orthogonal transforming means 102 can perform the processing adapted to the DC block according to the detection result by the block detecting means 121.

According to the invention of claim 5, the block detecting means 13 is provided.
Since the invalid block is detected by 1, the inverse orthogonal transforming unit 102 can perform the processing adapted to the invalid block according to the detection result of the block detecting unit 131.

[0029]

Embodiments of the present invention will now be described in detail with reference to the drawings. FIG. 4 shows the configuration of an embodiment of the image data restoration device of claim 2.

In FIG. 4, the decoding unit 711 and the decoding table 71 are shown.
2 corresponds to the decoding means 101, and is obtained by the decoding unit 711 searching the combination of the corresponding index and run from the decoding table 712 according to the input of each code, as in the conventional case. The search result is transmitted to the inverse quantization unit 210 as decoded data.

In the dequantizer 210, the above-mentioned decoded data is separated into a run and an index by the demultiplexer 201 and sent to the address calculation means 111 and the dequantization means 112, respectively.

The address calculating means 111 is composed of an integrating circuit 211 and a block detecting section 212,
The integrating circuit 211 is configured to sequentially integrate the value of the run by adding the numerical value "1". A run is input to the block detection unit 212, and a detection signal indicating that the end of the block has been detected is output according to the input of the run " _REOB " indicating the end of the block, and the integration circuit 2
The integrated value of 11 is reset to the initial value "0".

In this way, the scanning order by zigzag scanning (see FIG. 16) is obtained as the integration result corresponding to each index. Since this scanning order indicates the position occupied by the effective coefficient corresponding to each index in the quantized coefficient D _QU of 8 rows and 8 columns, this integration circuit 2
The scanning order obtained by 11 may be sent as the current block address.

In FIG. 4, the inverse quantization means 112 is composed of a multiplier 221 and a quantization threshold holding unit 222, and this quantization threshold holding unit 222 corresponds to the scanning order by zigzag scanning. , A quantization threshold value corresponding to each DCT coefficient is held. Therefore, in accordance with the input of the current block address obtained by the address calculation means 111, the quantization threshold holding unit 222 sends the corresponding quantization threshold to the multiplier 221, and the multiplier 221.
However, by multiplying the input index by this quantization threshold, the effective coefficient indicated by the index can be dequantized to obtain the DCT coefficient.

In this case, the demultiplexer 201
Thus, the run and the index are separated in advance and are input to the address calculating means 111 and the dequantizing means 112, respectively. However, the address calculating means 111 and the dequantizing means 112 respectively execute the run and the dequantized data from the decoded data. It may be configured by including means for extracting the index.

Further, in FIG. 4, the DCT coefficient obtained by the above-mentioned inverse quantizing means 112 is inputted to one of the input ports of the multiplexer 231, and the numerical value "" is inputted to the other input port of the multiplexer 231. "0" has been entered. That is, the multiplexer 231 operates according to the switching instruction described later, and the multiplier 2 described above is used.
The output of 21 and the numerical value "0" are selected and sent to the buffer 232.

The buffer 232 corresponds to the conversion coefficient storage means 113, and is configured to store the data input via the multiplexer 231 at the designated address.

Further, in FIG. 4, the memory 241 and the counter 242 form an address holding means 114, and the counter 242 increments the count value by "1" each time an effective coefficient address is input to the memory 241. It is configured to add. Further, the memory 241 uses the counter 2
The above-mentioned effective coefficient address is held at the address indicated by the count value of 42.

For example, as shown in FIG. 5B, the address corresponding to the effective coefficient shown by hatching in FIG.
The addresses are held in the addresses of the memory 241 corresponding to the input order.

The effective coefficient address held in the memory 241 in this manner is transferred to the memory 251 shown in FIG. 4 in accordance with the detection signal from the block detection unit 212 described above, and is stored as the previous block address by the memory 251. It is configured to be retained. At this time, the count value of the counter 242 is transferred to the register 254 shown in FIG. 4 and held by the register 254.

The memory 251, the comparator 252, the discrimination processing section 253, and the register 254 described above are used as the discrimination means 115.
The comparator 252 compares the previous block address held in the memory 251 with the current block address from the address calculation means 111, and the discrimination processing unit 253 will be described later based on the comparison result. It is configured to perform distribution discrimination processing.

The discrimination result by the discriminating means 115 is input to the multiplexer 231 and is used for the input port selection processing by the multiplexer 231. The previous block address output from the memory 251 described above is also input to the selector 255 together with the current block address from the address calculation unit 111, and the selector 255 responds to the determination result by the determination unit 115 described above. , The previous block address or the current block address is selected, and the buffer 23
2 is designated as a write address.

That is, the multiplexer 231 and the selector 255 operate in accordance with the discrimination result by the discriminating means 115 to write the numerical value "0" as an initial value in the address indicated by the discrimination result. The configuration of realizing the function of the initializing means 116 is made up of 231 and the selector 255.

Next, regarding the dequantization operation by the dequantization unit 210 shown in FIG. 4, the decoded data representing the quantized coefficient D _QU shown in FIGS. 5A and 5C are continuously input. A case will be described in detail as an example.

In this case, in the memory 251 described above,
The effective coefficient address in the quantized coefficient D _QU shown in FIG. 5A has already been held as the previous block address. The address calculation means 111 calculates the effective coefficient address as described above in response to the input of the combination of the index and the run of the decoded data representing the quantized coefficient D _QU shown in FIG. The addresses are sequentially input to the memory 241 and the comparator 252.

FIG. 6 is a flow chart showing the inverse quantization operation. In response to the input of the current block address (step 301), the discrimination processing unit 253 designates the address of the memory 251 and instructs the output of the previous block address. For example, the determination processing unit 253 may be configured to control the pointer designating the address of the memory 251. This pointer may be set to the head address of the memory 251 when the dequantization process of each block is started, and the pointer may be sequentially advanced.

Next, the comparator 252 compares the current block address (current AD) with the previous block address (previous AD). Based on the comparison result, first, the current block and the previous block are set to the same position. It is determined whether or not effective coefficients are distributed (step 302).

For example, since the current block address and the previous block address that are input first are both the address "0" corresponding to the DC component, the comparator 252 causes the current block address and the previous block address to match. The result of the comparison to the effect is obtained. In this case, it is determined that the effective coefficient is distributed corresponding to the corresponding address in both the current block and the previous block, and step 30
The positive determination in 2 may be made.

In this case, the discrimination processing section 253 instructs the multiplexer 231 to select the output of the multiplier 221 and the selector 255 to select the current block address. As a result, the DCT coefficient obtained by the multiplier 221 is stored in the current block address of the buffer 232 (for example, the address corresponding to the DC component) (step 303). At this time, the discrimination processing unit 25
3 updates the pointer designating the address of the memory 251 (step 304), instructs the memory 251 to output the next effective coefficient address, and then processes the next decoded data.

Here, if the value of the pointer before update matches the content of the register 254 in step 304, it means that there is no effective coefficient in the previous block after the current block address input in step 301. ing. Therefore, after that, according to the input of the current block address, the process of storing the output of the multiplier 221 in the corresponding address may be performed.

On the other hand, in the case of a negative judgment in step 302, it is judged that there is a difference in the distribution of effective coefficients between the current block and the previous block, and it is judged whether or not the current block address is smaller than the previous block address ( Step 30
5).

If the current block address is smaller than the previous block address, it means that this current block address corresponds to the invalid coefficient in the previous block. Therefore, in the case of the affirmative determination in step 305, the determination processing unit 253 determines that the multiplexer 231 and the selector 25 are the same as in step 303 described above.
5 to store the DCT coefficient in the current block address of the buffer 232 (step 306) and process the next decoded data.

On the other hand, if the current block address is larger than the previous block address, this indicates that this previous block address corresponds to the invalid coefficient in the current block. That is, this previous block address is a current block invalid address.

For example, the second input current block address "3" is determined to be larger than the second previous block address "2", and a negative determination is made in step 305. In this case, the determination processing unit 253 determines that the current block invalid address is detected, and the multiplexer 2
31 is instructed to select the numerical value “0”, and the selector 255 is instructed to select the previous block address. As a result, the numerical value “0” is stored in the corresponding address of the buffer 232 (for example, the previous block address “2”), and the effective coefficient written during the inverse quantization process of the previous block is cleared (step 307). ).

By the way, in the case of the negative judgment in the above step 305, there is a possibility that the current blockless invalid address exists in addition to the invalid coefficient indicated by the previous block address.

Therefore, the discrimination processing unit 253 updates the pointer after the end of step 307 described above (step 3
08), after instructing the memory 252 to output the next previous block address, the process returns to step 302 and the process of detecting the difference in distribution of effective coefficients between the current block and the previous block may be performed. As a result, the current block invalid address can be detected without exception.

In step 308 described above,
If the pointer before update matches the final address set in the register 254, the process of storing the DCT coefficient in the current block address is performed in the same manner as step 303 instead of the pointer update process. According to the input of the current block address, the process of storing the output of the multiplier 221 in the corresponding address may be performed.

Steps 301 to 308 described above
By repeating the above, the DCT coefficient D of the portion preceding the effective coefficient address corresponding to the effective coefficient included in the quantized coefficient D _QU of the current block is sequentially restored.

Further, the last previous block address and the last current block address are compared in accordance with the detection signal from the block detection unit 212, and the discrimination processing unit 253 performs the following processing according to the comparison result. As a result, the portion of the DCT coefficient D after the last current block address is restored.

When it is determined that the last previous block address is larger than the last current block address, the effective coefficient address in the previous block exists after the last current block address. In this case, the determination processing unit 253
However, the contents of the memory 251 may be sequentially read until the contents of the register 254 and the pointer match, and the initialization process for each of these addresses of the buffer 232 may be performed in the same manner as in step 307 described above. In other cases,
Since it is processed in the above-mentioned step 304 or step 308, the process may be ended as it is.

In this way, the inverse quantization result is stored in the current block address of the buffer 232, and the current block invalid address of the buffer 232 is selectively initialized, so that the DCT coefficient D of 8 rows and 8 columns is obtained. Can be restored.

As a result, unnecessary multiplication processing between the invalid coefficient and the quantization threshold is reduced, and unnecessary initialization processing for the address corresponding to the invalid coefficient in the previous block and the effective coefficient address in the current block is reduced. It becomes possible to do.

Here, as shown in FIGS. 5 (a) and 5 (c), in the quantized coefficient D _QU corresponding to two consecutive blocks, the effective coefficient distributions are often similar to each other. ,
In step 305 described above, very few are detected as the current block invalid address. For example, in the quantization coefficient D _QU shown in FIG. 5 (c), only three shaded addresses are detected as the current block invalid address requiring initialization processing. Therefore, the time required for the inverse quantization process of the quantized coefficient D _QU is the time required for the multiplication process for the seven effective coefficients and the time required for the write process of the initial value “0” to the above three addresses. Of the quantized coefficients for one block D _QU
It is possible to significantly reduce the time required for the inverse quantization process of.

Further, as described above, in the process of writing the DCT coefficient obtained by the multiplier 221 into the buffer 232, the process of writing the initial value "0" to the address detected in step 306 is performed, so that one block It is not necessary to perform the initialization process again after the end of the inverse quantization process.

Therefore, in response to the detection signal from the block detection unit 212, after the discrimination processing unit 253 has performed the above-mentioned processing, it is indicated that the restoration processing of the DCT coefficient D is completed by the inverse DC.
The configuration may be such that the T conversion unit 731 is notified and the start of the inverse DCT conversion process is instructed.

In response to this start instruction, the inverse DCT converter 7
31 is the DC stored in the buffer 232 as in the conventional case.
By performing the inverse DCT conversion process on the T coefficient, 1
The block of image data is restored. Further, by repeating the above-described inverse quantization processing and inverse DCT conversion processing for each block, image data for one screen is restored.

It should be noted that after all the effective coefficients of the current block are stored in the current block address of the buffer 232, the initialization process for the current block invalid address of the buffer 232 may be performed. For example, the memories 241 and 251 having 64 addresses are provided, and information indicating whether or not each address is a valid coefficient is stored for the quantization coefficient D _QU of each of the current block and the previous block, and each of them is valid. A map representing the distribution of coefficients may be created, the current block invalid address may be obtained from the comparison result of this map, and initial processing may be performed.

Further, since the inverse quantization process of each block and the inverse DCT transform process of each block can be processed independently, the inverse quantization process and the inverse DCT transform process may be processed in parallel. Good.

FIG. 7 shows the configuration of an embodiment of the dequantization unit of the image data restoration device of the third aspect. In FIG. 7, the conversion coefficient storage unit 113 has two buffers 233a and 233.
b, selector 234, and multiplexer (MPX) 235
And a pipeline control unit 236, and the output of the multiplexer (MPX) 231 described above is input to the buffers 233a and 233b, respectively. Further, in the conversion coefficient storage means 113, the pipeline control unit 236 causes the buffer 2
33a, 233b, selector 234, and multiplexer 2
35 to control the write operation and the read operation for the buffers 233a and 233b. Hereinafter, the buffers 233a and 233b are generically referred to as the buffer 233.

In this case, the discrimination processing unit 253 described above is used.
Is configured to send a notification to the effect that the process of restoring the DCT coefficient D has ended to the pipeline control unit 236, and the pipeline control unit 236, in response to this notification, writes the DCT coefficient D to the buffer 233. Is set to the read valid state, the other buffer 233 is set to the write valid state, and the selector 234 is instructed to switch the output port. In response to this, the selector 234 inputs the address from the selector 255 to the buffer 233 which is newly set to the write valid state, and the DCT coefficient writing process is performed.

At this time, the pipeline control unit 23
6 instructs the multiplexer 235 to select the input port corresponding to the buffer 233 in the read valid state, and instructs the inverse DCT conversion unit 731 to start the inverse DCT conversion process. The decoding unit 711 may be instructed to start the decoding processing of the next block in response to the end notification of.

In this way, since it is possible to perform the inverse quantization process of the next block in parallel with the inverse DCT transform process, it is possible to further speed up the restoration process for the entire image data.

By the way, the applicant of the present invention has proposed a technique for reducing the time required to restore the entire image in Japanese Patent Application No.
No. 290304 “Image data restoration system” has already been filed, and by implementing this technique together with the inverse quantization method of the present invention, it is possible to further speed up the restoration process of image data.

In Japanese Patent Application No. 1-290304 "Image Data Restoration Method", when the effective coefficient contained in the quantized coefficient D _QU or DCT coefficient D is only the DC component, the block produces an effective AC component. Instead of executing the inverse DCT conversion process by determining that it is a DC block that does not have DC
By multiplying the components by a constant and obtaining the image data of the corresponding block, the restoration processing of the entire image is speeded up.

FIG. 8 shows the configuration of an embodiment of the image data restoration device of claim 4. 8, the image data restoration device has a configuration in which a comparator 261 corresponding to the block detection means 121 is added to the inverse quantization unit 210 shown in FIG. Further, in FIG. 8, the inverse DCT conversion unit 731, the shift circuit 262, and the multiplexer 263 form an inverse orthogonal conversion unit 102.

The above-mentioned comparator 261 is configured to compare the count value of the counter 242 of the address holding means 114 with the threshold value "1" according to the input of the detection signal from the block detection unit 212, and the multiplexer. 263 is configured to operate according to the comparison result.

Here, when the comparator 261 determines that the quantization coefficient D _QU includes one effective coefficient, this effective coefficient corresponds to the DC component of the DCT coefficient D. That is, the block corresponding to this quantized coefficient D _QU is a DC block that does not have a valid AC component, and the image data of this block is all DC described above.
Since it is a value obtained by dividing the component by the constant “8”, the D stored in the buffer 232 by the shift circuit 262 described above.
By shifting the C component to the right by 3 bits, the image data of this block can be obtained.

That is, the multiplexer 263 switches between the output of the inverse DCT conversion section 731 and the output of the shift circuit 262 according to the output of the comparator 261, so that the inverse DCT conversion section 731 in the inverse orthogonal transformation means 102. The inverse DCT conversion process can be replaced with the shift process by the shift circuit 262.

As described above, the inverse quantizing unit 210 is provided with the block detecting means 121 to detect the DC block, so that the inverse orthogonal transforming means 102 is adapted to the DC block having no effective AC component. The processing described above can be performed, and the time required for the restoration processing of the image data can be shortened as a whole.

In this case, since only the DC component of the DCT coefficient D is used for the image data restoration process, the address corresponding to the AC component of the previous block is initialized to 8
It is not necessary to recover the DCT coefficient D of row 8 column. Therefore,
When a DC block is detected, the initialization process for the address corresponding to the AC component of the block preceding this DC block is skipped, and then the inverse quantization process of the block containing the AC component is performed. The inverse quantization process may be performed as described above using the address of the corresponding AC component as the previous block address.

For example, in accordance with the detection result indicating that the DC block is detected, the transfer operation from the memory 241 to the memory 251 and the transfer operation of the count value of the counter 242 to the register 254 are stopped, and the block preceding the DC block is stopped. The effective coefficient address and the number of the effective coefficient address in the above are stored in the memory 251 and the register 254, respectively. In this case, the effective coefficient address held up to the address of the memory 251 indicated by the contents of the register 254 is the previous block when the inverse quantization process of the quantization coefficient D _QU of the block having the effective AC component is performed. The DCT coefficient D used as an address and corresponding to the corresponding block is restored as described above.

As a result, the number of times the initialization process is executed can be suppressed to the minimum limit, so that the time required for the inverse quantization process can be further shortened. Further, the dequantization method of the present invention may be applied to an image data restoration device of a hierarchical restoration system.

FIG. 9 shows the configuration of an embodiment of a hierarchical restoration type image data restoration device to which the inverse quantization method of the present invention is applied.
9, the image data restoration device is configured by adding a hierarchical restoration processing unit 270 to the image data restoration device shown in FIG. 4, and includes the decoding unit 711 and the dequantization unit 2 described above.
10 and the inverse DCT transform unit 731 restore the image in each layer from the input code data, and based on the obtained image and the image in the previous layer, the layer restoration processing unit 27.
0 has a configuration to restore an image including all information up to the current layer. The layer restoration processing unit 270 holds, for example, image data of the previous layer, adds this image data and the image data obtained by the inverse DCT conversion unit 731, and outputs all information up to the current layer. It may be configured to obtain image data including

In the case of performing hierarchical restoration, as a code for one block in each hierarchy, a code obtained by selectively encoding only a part of the components of the quantized coefficient D _QU in 8 rows and 8 columns is input. For example, in the spectral selection method, first, only the DC component of each block is selectively encoded and transmitted, and then the second and third components are subjected to zigzag scanning.
The components corresponding to higher spatial frequencies are selectively coded and transmitted in sequence, such as selecting the component to be scanned next. Therefore, in each layer, the distributions of the effective coefficients in the quantized coefficients D _QU of the respective blocks almost overlap each other, so that the effective coefficient address in the previous block and the invalid coefficient in the current block corresponding to the invalid coefficient are invalid. The address is very small.

Therefore, when the dequantization method of the present invention is applied to the image data decompression device of the hierarchical decompression system, the time required for the dequantization process of the quantization coefficient D _QU corresponding to each block is shortened. Especially big.

The applicant has already applied for a patent application No. 2-180473 as a technique for shortening the time required to restore the entire image data in the hierarchical restoration.
This technique detects an invalid block in which the quantized coefficient D _QU does not include a valid coefficient, omits the inverse DCT transform process and the image data update process for this invalid block, and speeds up the restoration process of the entire image data. It is intended to

FIG. 10 shows the configuration of an embodiment of the image data restoration device of claim 5. 10, the image data restoration device has a configuration in which a comparator 271 corresponding to the block detection unit 131 is added to the image data restoration device shown in FIG. 9, and the inverse DCT conversion unit 731 and the hierarchy restoration processing unit are included. The inverse orthogonal transform means 102 is constituted by 270.

Further, the comparator 271 described above responds to the input of the detection signal from the block detection unit 212, and the count value of the counter 242 of the address holding means 114 and the threshold value "0".
_Are compared, and when they match, an invalid block signal indicating that the quantized coefficient D _QU of the corresponding block does not include an effective coefficient is output.

For example, in the spectral selection method, since the components corresponding to the high spatial frequencies are sequentially transmitted as described above, in the high order hierarchy, all the selected components may be invalid coefficients.

In such a case, as a code for one block, "E" indicating that the code for one block is completed
Only the "OB" code is input, and in response thereto, the invalid block signal is output by the comparator 271 described above.

In this case, all the components of the quantized coefficient D _QU corresponding to the corresponding block are invalid coefficients, and therefore,
Since all the components of the corresponding DCT coefficient D are also invalid coefficients, the inverse quantization process and the inverse DC for this block are performed.
The T conversion process is a useless calculation process. Further, in this case, since new information regarding the image of the current layer is not input, it is not necessary to update the image including the information up to the previous layer by the layer restoration processing unit 270 described above.

For example, in response to the above-mentioned invalid block signal, the discrimination processing unit 253 replaces the notification that the restoration processing of the DCT coefficient D is completed, and indicates that the corresponding block is an invalid DCT conversion unit. You may notify 731. Further, at this time, similarly to the processing of the DC block described above, from the memory 241 and the counter 242 to the memory 2
The transfer processing to 51 and the register 254 may be stopped, the contents of the memory 251 and the register 254 may be held, and used for the inverse quantization processing of the next block.

The inverse DCT transform unit 731 skips the inverse DCT transform process in response to the notification that the block is an invalid block, and relays this notification to the hierarchy restoration processing unit 270. The restoration processing unit 270 may be configured to skip the layer restoration process described above and directly output the image data of the previous layer.

As described above, the inverse quantizing unit 210 is provided with the block detecting means 131 to detect an invalid block, so that the inverse orthogonal transforming means 102 adapts to an invalid block which does not include valid information. The processing described above can be performed, and the time required for the restoration processing of the image data can be shortened as a whole.

[0095]

As described above, according to the present invention, when the effective coefficient is selectively dequantized to restore the coefficient matrix, the effective coefficient distribution in the quantized coefficient matrix corresponding to two consecutive blocks is calculated. By performing the initialization process while considering it, it is possible to reduce wasteful initialization process for the address corresponding to the invalid coefficient of the previous block and the address corresponding to the valid coefficient of the current block. The time required for processing can be shortened, and the speed of image data restoration processing can be increased.

Further, by performing the inverse quantization process and the initialization process and the inverse DCT transform process in parallel, the processing time per block can be shortened and the restoration process can be performed at a higher speed.

Further, by detecting the DC block and the invalid block and performing the processing adapted to these blocks in the inverse orthogonal transform means, the time required for the inverse quantization processing of each block is shortened and the entire image is restored. The time required for processing can be significantly reduced.

[Brief description of drawings]

FIG. 1 is a diagram showing the principle of an inverse quantization method according to the present invention.

FIG. 2 is a diagram showing a configuration of an image data restoration device according to claims 2 and 3;

FIG. 3 is a diagram showing a configuration of an image data restoration device according to claims 4 and 5;

FIG. 4 is a diagram showing a configuration of an embodiment of the image data restoration device of claim 2;

FIG. 5 is an explanatory diagram of effective coefficient addresses.

FIG. 6 is a flowchart showing an inverse quantization process.

FIG. 7 is a diagram showing a main part configuration of an embodiment of an inverse quantization unit of the image data restoration device according to the present invention.

FIG. 8 is a diagram showing a configuration of a main part of an embodiment of an inverse quantization unit of the image data restoration device according to claim 4;

FIG. 9 is a diagram showing a main configuration of an embodiment of an image data restoration device of a hierarchical restoration system to which the inverse quantization method of the present invention is applied.

FIG. 10 is a diagram showing a main configuration of an embodiment of an inverse quantization unit of the image data restoration device of claim 5;

FIG. 11 is a configuration diagram of a conventional image data compression device.

FIG. 12 is a diagram showing an example of a block.

FIG. 13 is a diagram showing an example of a DCT coefficient D.

FIG. 14 is a diagram showing a quantization matrix.

FIG. 15 is a diagram showing a quantization coefficient D _QU .

FIG. 16 is an explanatory diagram of zigzag scanning.

FIG. 17 is a configuration diagram of a conventional image data restoration device.

FIG. 18 is a schematic configuration diagram of a conventional inverse quantization processing unit.

[Explanation of symbols]

 101 Decoding Means 102 Inverse Orthogonal Transforming Means 111 Address Calculating Means 112 Inverse Quantizing Means 113 Transform Coefficient Storing Means 114 Address Holding Means 115 Discriminating Means 116 Initializing Means 121, 131 Block Detecting Means 201 Demultiplexer (DMPX) 211 Integrating Circuits 212 Blocks Detection unit 221,722 Multiplier 222,723 Quantization threshold holding unit 231,235,263 Multiplexer (MPX) 232,233 Buffer 234,255 Selector 236 Pipeline control unit 241,251 Memory 242 Counter 252,261,271 Comparator 253 Discrimination processing unit 254 Register 262 Shift circuit 270 Hierarchical restoration processing unit 611 DCT conversion unit 620 Linear quantization unit 631 Encoding unit 632 Code table 711 Decoding unit 712 Decoding table 20 inverse quantization unit 721 sequences restorer 724 address calculating unit 725 DCT coefficient holding unit 731 inverse DCT unit

Claims (5)

[Claims] 1. A quantized coefficient matrix obtained by orthogonally transforming an original image for each block composed of a plurality of pixels and quantizing each component of a coefficient matrix composed of transform coefficients with a corresponding quantization threshold value. In the dequantization method for obtaining the coefficient matrix from the quantized coefficients obtained by decoding the code data when restoring the original image from the code data by the orthogonal transform coding method, each block and the block before it In the quantized coefficient matrix corresponding to each, the difference in the distribution of the effective coefficient having a value other than zero is determined, and the effective coefficient included in the quantized coefficient matrix corresponding to each block is multiplied by the corresponding quantization threshold. And selectively dequantize, and hold the obtained dequantization results according to the effective coefficient address indicating the position of the effective coefficient in the quantized coefficient matrix. In both cases, the inverse quantization method is characterized in that the coefficient matrix corresponding to each block is restored by performing initialization processing on the corresponding address according to the difference in distribution of effective coefficients from the previous block. 2. A quantized coefficient matrix obtained by orthogonally transforming an original image for each block made up of a plurality of pixels and quantizing each component of a coefficient matrix made up of transform coefficients with a corresponding quantization threshold value. Decoding means (101) for continuously inputting coded data by the orthogonal transform coding method, decoding the coded data, and sequentially outputting decoded data representing a quantized coefficient matrix corresponding to each block, An address to be output as the current block address, which is obtained based on the decoded data, and which obtains the effective coefficient address indicating the position in the quantized coefficient matrix of the effective coefficient having a value other than zero included in the quantized coefficient matrix of each block. Inverse quantization means for performing inverse quantization by calculating means (111) and multiplying the effective coefficient by a quantization threshold value indicated by the corresponding current block address. 1
12), a transformation coefficient storage means (113) for storing the dequantization result by the dequantization means (112) in the current block address as a transformation coefficient, and the current block address being sequentially held, and Address holding means (1) that outputs the effective coefficient address held when processing the block as the previous block address
14), and a discrimination means (115) for discriminating the current block invalid address corresponding to the invalid coefficient in each block and corresponding to the valid coefficient in the previous block based on the previous block address and the current block address. ), An initialization unit (116) for initializing by writing an initial value to a storage location of the conversion coefficient storage unit (113) indicated by the current block invalid address, and stored in the conversion coefficient storage unit (113). An image data restoration device, comprising: an inverse orthogonal transformation means (102) that restores image data by performing an inverse orthogonal transformation on a coefficient matrix made up of the transformed coefficients. 3. The image data restoration device according to claim 2, wherein the transform coefficient storage means (113) switches a plurality of areas having a capacity corresponding to a coefficient matrix for one block to correspond to consecutive blocks. An image data restoration device having a configuration for storing a coefficient matrix for 4. The image data restoration device according to claim 2, wherein a DC block represented by only a DC component is detected based on the number of effective coefficients included in the quantization coefficient matrix corresponding to each block, The result of this detection is the inverse orthogonal transformation means (102).
Block detecting means (12
An image data restoration device comprising 1). 5. The image data restoration device according to claim 2, wherein an invalid block containing no effective coefficient is detected based on the number of effective coefficients included in the quantization coefficient matrix corresponding to each block, and this detection is performed. Block detection means (131) for providing the result to the inverse orthogonal transformation processing by the inverse orthogonal transformation means (102)
An image data restoration device comprising: