JP3382556B2 - Image encoding apparatus and method - Google Patents

Description

Detailed Description of the Invention

[0001]

FIELD OF THE INVENTION This invention relates to encoding image data.

[0002]

2. Description of the Related Art In the ADCT compression method, the data format is standardized by JPEG. FIG. 14 is a diagram showing a standard data format of a general ADCT compression method. As shown in FIG. 14, the data format is roughly divided into a header section 5101 and a data section 5102. Header part
The 5101 includes a quantization table unit 5111 and a Huffman table unit 5112.

FIG. 15 is a block diagram showing a configuration example of a conventional ADCT compression / expansion circuit. In FIG. 15, the color space conversion circuit
The 5201 converts the RGB data of the input NTSC color image into YCrCb data separated into a luminance signal and a color difference signal. R
Conversion from GB data to YCrCb data is performed by the matrix operation of equation (1).

Hereinafter, the data converted into YCrCb is processed separately for Y data, Cr data, and Cb data.
In particular, since human visual characteristics are relatively sensitive to luminance and insensitive to color difference, the luminance data Y and the color difference data CrCb are handled as described below in subsampling and quantum. Often different in the coded table.

The sub-sampling is to make CrCb data into a smaller number of data than Y data.
There are two modes for subsampling, one is called "4-2-2 mode" and the other is "4-1-1 mode".
Called.

FIG. 16 is a diagram showing a specific example of general sub-sampling. For example, the "4-2-2 mode" sub-sampling for Cr data as shown in FIG. 16 (a) is performed by averaging two Cr data that are horizontally adjacent to each other, and Data. For example, the data '0 in the figure (b)
0b 'is calculated by the equation (2). 00b = (00 + 01) / 2… (2)

On the other hand, in the "4-1-1 mode" sub-sampling, four Cr data which are adjacent to each other in the horizontal direction and the vertical direction are averaged to obtain one data shown in FIG. For example, the data '00c' in FIG. 7C is calculated by the equation (3). 00c = (00 + 01 + 10 + 11) / 4… (3)

The sub-sampled Y, Cr and Cb data are processed independently. In the middle of the process,
Although the coefficients of the quantization table may differ, the processing procedure is exactly the same, so the following description will be given only for the processing of Y data.

The actual image data Y output from the color space conversion circuit 5201 shown in FIG. 15 is a data unit of 8 × 8 (total of 64) of 8 lines continuous in the horizontal direction and 8 lines continuous in the vertical direction. , DCT (discrete cosine transform) circuit 5202, and converted into frequency space data.

FIG. 17 is a diagram showing a state of a general DCT. The 8 × 8 actual image data shown in FIG. 17 (a) is the DCT circuit 5202.
Is converted into the 8 × 8 frequency space data shown in FIG. The component indicated by 5401 in the same figure (b) is a DC component,
This is a parameter proportional to the average value of the 8 × 8 actual image data in FIG. In addition, the other 63 components in FIG.
It is called the C component and represents the space of different frequencies,
The number is proportional to the amplitude of that frequency. The figure
Arrows 5402 to 5404 shown in (b) indicate that the frequency becomes higher in the direction of the arrow. In particular, arrow 5402 indicates the horizontal frequency (vertical stripes in the actual image space) and arrow 5403 indicates the vertical frequency. (Horizontal stripes in the actual image space) and arrow 5404 represents the frequency in the horizontal and vertical directions (lattice stripes in the actual image space).

FIG. 18 is a diagram showing how DCT is performed on general 2 × 2 block image data. In FIG. 18, 5601 is real image data and 5602 is DCT data of the real image data 5601. Further, 5603 represents the physical meaning of the DCT data 5602. In particular, 5631 is DC data and 5632 to 5634 are AC data.
The data. The part indicated by black squares in 5603 is +1.
Is indicated by a white square (□). DC data 5602
, The DC data value '100' is 4 in the actual image data 5601.
This is the average value of the data.

FIG. 19 is a diagram showing a general 8 × 8 block DCT base image. When DCT processing of natural images, in most images, the numerical values are concentrated in the low frequency part, and only very small numerical values appear in the high frequency part. Furthermore, human visual characteristics are sensitive to the gradation of the low frequency part,
It is insensitive to the gradation of the high frequency part.

The data processed by the DCT circuit 5202 shown in FIG. 15 is divided by the quantization circuit 5204 by the data in the quantization table 5203 and rounded off.

FIG. 20 is a diagram showing a state of quantization in the conventional ADCT compression / decompression device. The DCT data shown in FIG. 20 (a) divided by the quantization table shown in FIG. 20 (b) is
It is the quantized data shown in FIG. 7C, and although the zero data increases and the compression rate increases due to the quantization, it is also a factor that makes the data irreversible at the time of expansion. The numerical values in the quantization table are set to small values in the low frequency part and large values in the high frequency part in consideration of the visual characteristics described above.

FIG. 21 shows a quantization table recommended by JPEG. FIG. 21 (a) is for Y data and FIG. 21 (b) is for CrCb data.

The quantized data output from the quantization circuit 5204 shown in FIG. 15 is input to the Huffman coding circuit 5205. The Huffman encoding circuit 5205 converts the input quantized data to DC.
Separate into component and AC component, and encode independently.

There is only one DC component in each block, and its value is often large, but since it is an average density, it has a correlation with the DC components of adjacent blocks. Therefore, the coding of the DC component is performed by obtaining the difference between the DC component of the block to be coded and the DC component of the block immediately preceding the block, and coding the difference data having a high appearance frequency with a short bit length. Assign

FIG. 22 is a diagram showing an example of a general zigzag scan. As shown in FIG. 22, for the AC component, the data is rearranged by the zigzag scan so that the zero data is continuous. After that, a run-length histogram of zero data is taken, and a code having a short bit length is assigned to data having a high appearance frequency.

The data encoded and compressed by the Huffman encoding circuit 5205 is processed by the format encoder 5206 in FIG.
The data is formatted in the standard format as shown in, and transmitted from the communication interface (I / F) 5207 or stored in the memory of the communication I / F 5207.

Decompression of data is performed by receiving the data received by the communication I / F 5207 or extracting the data from the memory of the communication I / F 5207.
The format decoder 5208 decodes it, and the Huffman decoding circuit 5210 decodes it. The inverse quantization circuit 5211 multiplies the decoded data by the quantization table 5209 read by the format decoder 5208. For example, the data shown in FIG. 20 (c) is inversely quantized as shown in FIG. 23 by the quantization table shown in FIG. 20 (b). The dequantized data is
Inverse DCT circuit 5212 reverse DCT, color space conversion circuit 5213 YCrC
b → RGB conversion.

[0021]

However, the above technique has the following problems.

That is, the ADCT compression method can improve the compression rate by increasing the value of the quantization table. However, since the compression rate is irreversible due to the rounding error in the quantization, the improvement of the compression rate is achieved by expanding the expanded image. There is a problem that deteriorates. Note that data changes due to compression and decompression are shown in Fig. 20.
DCT data before quantization shown in (a) and after dequantization shown in FIG.
It is clear when compared to the DCT data.

The main image deteriorations include block distortion and mosquito noise. Block distortion means that the AC component is lost in the 8 × 8 block that is the processing unit of ADCT, the joint with the adjacent block becomes conspicuous, and in extreme cases
It is a deterioration in which only the DC component remains and the resolution becomes 1/8. Mosquito noise is the noise that occurs around the outline of a character, and the image edge originally shown in Figure 24 (a) is due to ADCT due to the necessary AC component cutoff and quantization error. This is the deterioration that results in the image edge shown in b).

When attempting to correct the above deterioration at the time of expansion, correction information of the original image such as whether a high frequency component is included is required. However, when only the encoded data of the image is received, the above-mentioned correction information does not exist, and thus it is impossible to efficiently correct the decoded image. As a method for solving this, it is conceivable to attach correction information in addition to the coded data of the image, but in this case, there is a problem that the overall data amount becomes large.

As described above, it is impossible to transmit the additional information required by a person who handles the image data later without increasing the overall data amount of the image data.

The present invention is intended to solve the above-mentioned problem, and the additional information required by a person who later handles image data is stored in the image data while suppressing an increase in the overall amount of the image data. The main purpose is to provide a technique for efficiently holding the material.

More specifically, when the image is held as a frequency component, it is intended to suppress an increase in the overall amount of data and also hold additional information so as not to deteriorate the image as much as possible. And

In particular, when the image data is converted into frequency components, quantized, and entropy-encoded, it is an object of the present invention to prevent the reduction of the compression rate of the image data due to the data operation for holding the additional information as much as possible. .

[0029]

The present invention has the following structure as one means for achieving the above object.

The image processing apparatus according to the present invention comprises a converting means for converting image data into a data string of frequency components, a quantizing means for quantizing the data string obtained by the converting means, and an edge component from the data string. The first code that extracts one data corresponding to and identifies that data,
And, by the second code indicating the approximate value before the quantization of the data, in the quantized data string obtained by the quantization means, a replacement means for replacing a part of the quantized data corresponding to a high frequency component, Encoding means for entropy-encoding the quantized data string after replacement, the replacement means, when there is no data corresponding to the edge component in the data string of the frequency component, the first and second It is characterized in that both codes of are set to zero.

Further, a converting means for converting the image data into a data string of frequency components, a quantizing means for quantizing the data string obtained by the converting means, and a value of the data string of a predetermined value or more, and , A first code that identifies one piece of data that satisfies the condition that the value after quantization is zero, and a generation means that generates a second code other than zero that indicates the approximate value of that data before quantization A part of the quantized data string obtained by the quantizing part, which corresponds to a high-frequency component, is replaced with the first and second codes, and a quantized data string after replacement. Encoding means for entropy encoding, the replacement means, in the data string of the frequency component, when the data that satisfies the condition does not exist, the replacement of the first and second codes. Characterized by a zero.

The image processing method according to the present invention converts image data into a data string of frequency components, quantizes the data string obtained by the conversion, and extracts one data corresponding to an edge component from the data string. A first code that identifies the data and a second code that indicates the approximate value of the data before quantization, which corresponds to a high-frequency component in the quantized data string obtained by the quantization. If the data corresponding to the edge component does not exist in the data string of the frequency component in the extraction, each step has a step of entropy-encoding the quantized data string after replacement, The first and second codes are both zeroed.

Further, the image data is converted into a data string of frequency components, the data string obtained by the conversion is quantized, and the value in the data string is equal to or more than a predetermined value and the quantized value is zero. A first code that identifies one piece of data that satisfies a certain condition, and a quantized data string that is obtained by the above quantization that generates a second code other than zero that indicates the approximate value of that data before quantization. Among them, a part of the quantized data corresponding to a high frequency component is replaced with the first and second codes, and each step of entropy-encoding the quantized data string after the replacement is included, and in the generation, When there is no data satisfying the condition, both the first code and the second code are set to zero.

[0034]

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment of the present invention will be described in detail below with reference to the drawings. In the following description, an example in which 8 × 8 blocks are processed as one unit will be described, but the present invention is not limited to this, and blocks of arbitrary size can be processed as a unit.

[0035]

First Embodiment FIG. 1 is a block diagram showing a configuration example of a first embodiment according to the present invention.

A color space conversion circuit a 101 converts RGB color space data of an NTSC color image into YCrCb color space data.
A DCT circuit 102 converts the YCrCb color space data output by the color space conversion circuit 101 into frequency space data in 8 × 8 block units. A quantization table a 103 is composed of a rewritable memory.

Reference numeral 104 denotes a quantization circuit, which is a quantization table a103.
The output of the DCT circuit 102 is quantized based on An edge information encoder 105 inputs the output of the DCT circuit 102 and the output of the quantization circuit 104 and creates edge information. 106
Is a synthesizing circuit a, in which the output of the quantizing circuit 104 is zigzag scanned and rearranged, and the edge information encoder 105
And the output of.

Reference numeral 107 denotes a Huffman coding circuit, which is a synthesizing circuit a10.
The output of 6 is separated into DC component and AC component and Huffman coding is performed. A format encoder 108 converts the output of the Huffman encoding circuit 107 and the quantization table a103 into the ADCT standard format. A communication I / F 109 transmits the output of the format encoder 108 to an external device or stores it in an internal memory.

Reference numeral 110 denotes a format decoder, which is a communication I / F 10
9 received from external device or communication I / F 109
Decodes ADCT standard format data retrieved from the internal memory of the. 111 is a quantization table b, which is composed of a rewritable memory and stores the quantization table decoded by the format decoder 110. A Huffman decoding circuit 112 decodes the code data decoded by the format decoder 110.

Reference numeral 113 denotes an inverse quantization circuit, which is a quantization table b11.
Based on 1, the quantized data decoded by the Huffman decoding circuit 112 is dequantized. 114 is an edge information decoder,
The edge information decoded by the Huffman decoding circuit 112 is decoded. Reference numeral 115 denotes a synthesizing circuit b, which synthesizes the output of the inverse quantization circuit 113 by inverse zigzag scanning and returning it to an 8 × 8 block, and the output of the edge information decoder 114.

An inverse DCT circuit 116 inversely DCTs the DCT data output from the combining circuit b115. 117 is a color space conversion circuit b
Then, the YCrCb color space data output from the inverse DCT circuit is converted to RG
Convert to B color space data.

In this embodiment, the color space conversion circuits 101 and 117, the DCT circuit 102 and the inverse DCT circuit 116, the quantization circuit 104 and the inverse quantization circuit 115, and the Huffman encoding circuit 107 shown in FIG.
Since a known circuit is used for the Huffman decoding circuit 112, detailed description thereof will be omitted.

Next, a method of extracting edge information will be described. FIG. 2 is a diagram for explaining the frequency domain of a general DCT.

The human visual characteristics are sensitive to the gradation of the area A (low frequency portion) in FIG. 2, but the area B (high frequency portion) in FIG.
Is insensitive to the gradation of, and at a pixel density of 200 dpi or more, region C (super high frequency part) in the figure looks like a flat level. For example, even a small character of about 6 points has a line width of about 2 pixels at 400 dpi.
The components are concentrated in area B in FIG.

Therefore, unless enlargement processing is performed,
The area C (super high frequency part) of 2 is not important, and the information necessary for image restoration is included in areas A and B in the figure, and the data is '0' even though it has a large amount of information. This is the area that will be converted to. In the present embodiment, as shown in FIG. 3, the edge information is extracted in the priority order of the zigzag scan order excluding the DC component, but in the low frequency part, the quantization table value may be small, which is important. Since information is hardly lost, edge information may be extracted, for example, in the priority order shown in FIG.

In this embodiment, a code code [0] representing the frequency space of the edge and a code code [1] representing the amplitude of the edge are used.
Two codes are used as edge information, and a priority number as shown in FIG. 3 or 4 is set to code [0] indicating frequency information.

FIG. 5 and FIG. 6 are views showing an example of the method for creating the edge information code, where the axis x represents the amplitude, T the threshold value, and H the step width.

In this embodiment, in an arbitrary 8 × 8 block,
Whether 8 edges or not is included in the 8 × 8 block
Judgment is made based on whether there is an absolute value of AC component larger than T.
In FIG. 5, in the code [1] indicating the amplitude information, the step width H is set to those having a threshold value T or more and a threshold value −T or less,
Assign the corresponding numbers. That is, when the amplitude x is decoded as described below, for example, the intermediate value T + H / 2 indicated by a black circle in FIG. 5 is set as the amplitude x. Code [1] = 2 if T + 2H <x <T + 3H, code [1] = 1 if T + H <x <T + 2H, code [1] if T <x <T + H. 1] = 0, code [1] = -1 if-(T + H) <x <-T, code [1] = if-(T + 2H) <x <-(T + H) -2,-(T + 3H) <x <-(T + 2H) If code [1] = -3

Also, the amplitude x at an arbitrary position of the 8 × 8 block
Can be said to be less than or equal to half the value QT in the quantization table corresponding to that position and greater than the threshold value T, as shown in equation (4). T <| x | <QT / 2… (4)

Therefore, it suffices to know whether the amplitude is positive or negative. As shown in FIG. 6, the code [1] indicating the amplitude information has a number '0' for the threshold value T or more,
The number '1' may be assigned to the threshold value −T or less, and the amplitude x is decoded according to the equation (5), for example.

FIG. 7 is a flowchart showing an example of edge information extraction processing in the edge information encoder 105.

In FIG. 7, first, in step S201, edge information code [0] and code [1] are initialized to '0', and step S2
In 02, set the parameter i indicating the processing position of 8 × 8 block to '1'
To

In subsequent steps S203 to S205, each AC of 8 × 8 blocks is arranged in the order of priority shown in FIG. 3 (or FIG. 4).
Edge information is extracted from the components, and the edge information extraction processing is terminated when the edge information is extracted. Step S203
Then, in the DCT data output from the DCT circuit 102, the absolute value | x _i | of the amplitude of the AC component of the priority order corresponding to the parameter i and the threshold value
Compare with T, and if | x _i |> T, proceed to step S204,
If | x _i | ≦ T, the process proceeds to step S206. Then, in step S204, of the quantized data output by the quantization circuit 104,
Quantized value y of AC component of priority corresponding to parameter i
_{If i} is y _i ≠ 0, the process proceeds to step S206, and if y _i = 0, the process proceeds to step S205.

Next, in step S205, as described above and in FIG.
Edge information cod by the method described in (or Figure 6)
e [0], code [1] are created, the edge information data is output in step S208, and then the edge information extraction processing ends.

In step S206, the parameter i is incremented, and it is determined in step S207 whether all AC components have been checked. That is, in step S207, i
If <64, the process returns to step S203, and if i ≧ 64, the process proceeds to step S208. In step S208, the edge information data is output, and then the edge information extraction process ends. In this case, edge information code [0] = 0, code [1] = 0 is output.

Next, the synthesizing circuit a106 zigzag scans the quantized data of 8 × 8 blocks output from the quantizing circuit 104 in the order shown in FIG. 22, for example, and rearranges them into a series of data. The edge information data output by the edge information encoder 105 is replaced with the last two pieces of data in the series of data.

As described above, the AC component such as the character edge appears in the area B (high frequency portion) in FIG. 2, and a large value appears in the area C (super high frequency portion) in FIG. 2 in a normal image. There is no. When simulated with an image (4096 blocks) that was read with relatively high accuracy, the 63rd and 64th data had 16 quantization data values of "+1" or "-1" after being quantized at a quantization table value of "20". , The quantization table value is '50'
Then all became '0'. The quantization table values recommended by JPEG are “103” and “99” as shown in FIG. 21, and the 63rd and 64th data are normally set to “0” by quantization. Therefore, even if the synthesizing circuit a106 replaces the last two data of the series of data, that is, the 63rd and 64th data with the edge information data, the image is not deteriorated.

On the other hand, the replacement of data by the synthesis circuit a106 may reduce the zero run length. However, in a natural image, generally, the ratio of edges occupied in the image is small, and as described above, the edge information when no edge is extracted is code [0] = 0, code [1] = 0. If it is set so that the replacement of the data by the synthesizing circuit a106 affects the compression efficiency to a minimum.

There is also a problem of data compatibility with a conventional device having a standard ADCT format decoding function. For example, when communicating with a conventional device, edge information data code [0] = 0 By setting code [1] = 0, data compatibility can be maintained.

Next, the output of the synthesizing circuit a106 is coded by the Huffman coding circuit 107, and the format encoder 108
Sent to. The format encoder 108 stores the output of the Huffman encoding circuit 107 and the quantization table a103 in FIG.
Converted to the ADCT standard format shown in 4, but the quantization table unit 5111 corresponding to the data replaced with the edge information
The data of is 0.

Next, the case of expanding the data received from the external device by the communication I / F 109 or taken out from the internal memory of the communication I / F 109 will be described.

The ADCT standard format data output from the communication I / F 109 is decoded by the format decoder 110, and the data fetched from the quantization table unit 5111 shown in FIG. 14 is stored in the quantization table b111. Also,
The data extracted from the Huffman table unit 5112 is input to the Huffman decoding circuit 112 and decoded.

Next, the inverse quantization circuit 113 inversely quantizes the quantized data output from the Huffman decoding circuit 112 based on the quantization table b111, but as described above, it is replaced with the edge information. Since the quantization table data corresponding to the data is '0', the 63rd and 64th data of the series of data output by the inverse quantization circuit 113 is '0'. on the other hand,
The edge information decoder 114 inputs the quantized data output from the Huffman decoding circuit 112, and decodes the parameter i representing the frequency of the edge information and the amplitude x _i from the last two pieces of data. Next, the synthesis circuit b115, the inverse quantization circuit 113
The data input from the, for example, rearranged in the order shown in FIG. 22, to configure 8 × 8 block, further based on the edge information input from the edge information decoder 114, designated by the parameter i, 8 × shown in Figure 3 (or Figure 4)
Replace the data in the specified block of 8 blocks with the amplitude x _i .

Next, the inverse DCT circuit 116 inverse DCTs the DCT data input from the synthesizing circuit b115 and outputs YCrCb data. Subsequently, the color space conversion circuit b117 converts the YCrCb data input from the inverse DCT circuit 116 into RGB data and outputs the RGB data.

Next, the effect of the correction according to the present embodiment will be shown in FIG.
Description will be made regarding the case where the DCT data shown in 20 (a) is quantized by the quantization table shown in FIG. 20 (b). In the following description, it is assumed that the threshold value T = 10, the priority order shown in FIG. 4 and the code creation method shown in FIG. 6 are used.

20A and 20C, the block satisfying the following two conditions is the block indicated by 2d in FIGS. 20A and 20C, that is, the priority order shown in FIG. It is a block of 8 ', and the DCT data of the block is '10 .3'. Therefore, the edge information data is code [0] = 8, code [1] = 0
become. Quantized data = 0 | DCT data |> T = 10

FIG. 8 is a diagram showing an example of the quantized data and edge information data synthesized by the synthesis circuit a106. That is, the data of the last two blocks 7g, 7h of the quantized data in FIG. 20 (c) is replaced with the edge information data,
Block 7g is edge information data code [1] = 0
h is replaced with edge information data code [0] = 8.

FIG. 9 is a diagram showing an example of the DCT data corrected by the synthesizing circuit b115, in which the data of the block 2d is replaced based on the edge information data. In the conventional device, compression and expansion are performed in the following order. DCT data in Figure 20 (a) ↓ Quantized data in Figure 20 (c) ↓ DCT data in Figure 23

In this embodiment, compression / decompression processing is performed in the following order. DCT data in Figure 20 (a) ↓ Composite data in Figure 8 ↓ Corrected DCT data in Figure 9

Block of original image DCT data shown in FIG. 20 (a)
2d is '10 .3 ', the block 2d of the decompression result by the conventional device shown in FIG. 23 is'0', the block 2d of the decompression result according to the present embodiment shown in FIG. 9 is' 11 ', and according to the present embodiment,
It is clear that the edge information is corrected. Note that the figure
The data of the block 2d shown in 9 is decoded by the edge information decoder 114 according to the expression (5) as shown in the expression (6). x = (10 + 24/2) / 2 = 11… (6)

In the above description and the drawings, the example in which the number of blocks for extracting edge information is one has been described, but the present embodiment is not limited to this, and edge information is extracted from a plurality of blocks in order of priority. May be extracted and the data from the end of the quantized data may be replaced with a plurality of edge information data.

In the above description and FIGS. 5 and 6, an example of the method of creating the edge information code has been described.
The present embodiment is not limited to this, and other edge information code creation methods may be used. For example, DCT
If the data is sorted in descending order and the edge information is extracted from one or a plurality of blocks in which the quantized data is '0' in descending order of DCT data, more accurate correction is possible.

As described above, according to the present embodiment, the frequency information and the amplitude information of the frequency are added to the compressed image data without deteriorating the compression rate and without deviating from the ADCT standard format. It is possible to include edge information of the original image, which is a combination of the above, and to perform sufficient correction when decompressing the compressed image data, so that deterioration of image quality due to image data compression / decompression can be improved.

Further, according to the present embodiment, by setting the edge information data code [0] = 0, code [1] = 0, the standard ADC
It is also possible to maintain data compatibility with a conventional device having a T format decoding function.

In the above description and FIG. 1,
Although the configuration example in which the DCT data is corrected by the synthesizing circuit b115 based on the edge information decoded by the edge information decoder 114 has been described, the same result can be obtained by the configuration example shown in FIG. 10, for example. That is, in the configuration example shown in FIG. 10, an image processing circuit 301 is provided in the subsequent stage of the color space conversion circuit b117 instead of the synthesizing circuit b115, and the image processing circuit 301 corrects the RGB data based on the edge information. Is.

[0076]

Second Embodiment FIG. 11 is a block diagram showing a configuration example of a second embodiment according to the present invention.

In the second embodiment shown in FIG. 11, the quantization table a103 and the edge information encoder 105 are removed from the configuration example of the first embodiment shown in FIG. This is a configuration in which a quantization table c202, a quantization table d203, and a selector a204 are added. Also,
In the second embodiment, in its decompression unit, the quantization table b111 and the edge information decoder 114 are removed from the configuration example of the first embodiment shown in FIG. 1, and the quantization table e214, the quantization table f215, and the selector b216 are added. This is the added configuration. No. 2
The other configurations of the embodiment are substantially the same as those of the first embodiment, and the same reference numerals are given to omit detailed description.

In the compression unit of FIG. 11, an edge determination circuit 201 checks the DCT data of the 8 × 8 block output from the DCT circuit 102, and the 8 × 8 block shows a halftone image showing a gradual gradation change. , Or an edge portion such as a character is determined. Note that the determination result is selector a2
04 and the synthesis circuit a107, 8x8 block, for example
It is replaced with the end data (64th data) of the series of data obtained by the zigzag scanning shown in 22.

Numerals 202 and 203 are respectively quantization tables, which are composed of rewritable memories, a quantization table c202 is a quantization table corresponding to a halftone image and the like, and a quantization table d203 is a quantization table corresponding to characters and the like. It is a conversion table. A selector a 204 selects either the quantization table 202 or 203 based on the determination result of the edge determination circuit 201 and sends the selected quantization table to the quantization circuit 104.

Next, in the decompression unit of FIG. 11, reference numerals 214 and 215 denote quantization tables, which are rewritable memories and store the quantization tables decoded by the format decoder 110. The quantization table e214 stores a quantization table corresponding to a halftone image or the like, and the quantization table f215 stores a quantization table corresponding to a character or the like.

216 is a selector b, which is a Huffman decoding circuit 112.
Either the quantization table 214 or 215 is selected based on the end data (64th data) of the series of data that has been decoded by, and the selected quantization table is dequantized by the inverse quantization circuit.
Send to 113.

Next, the edge determination method will be described.

As described in the first embodiment, the human visual characteristics are sensitive to the gradation of the area A (low frequency part) in FIG. 2, but are sensitive to the gradation of the area B (high frequency part) in FIG. Is insensitive,
Furthermore, when the pixel density is 200 dpi or more, the area C in the figure
(Ultra high frequency part) looks like a flat level. For example, even a small character of about 6 points has a line width of about 2 pixels at 400 dpi, so the AC component thereof is concentrated in the area B of FIG. Therefore, in the present embodiment, all 63 blocks of the AC components may be subjected to the edge determination, but may be limited to the blocks in the area indicated by the black squares in FIG.

FIG. 13 is a flow chart showing an example of edge judgment processing in the edge judgment circuit 201. In the present embodiment, whether or not an edge is included in an arbitrary 8 × 8 block is determined by whether or not the 8 × 8 block has an absolute value of an AC component larger than the threshold value T.

In FIG. 13, first, in step S301, the edge information code is initialized to '0', and in step S302, 8 × 8.
Set the parameter i indicating the block processing position to "1".

In the following steps S303 to S304, the edge information determination of each AC component of the 8 × 8 block is performed in the order of removing the DC component from the zigzag scan shown in FIG. 22, and the edge information determination is performed when the edge information is detected. To finish. Step S3
In 03, the parameter i of the DCT data output by the DCT circuit 102
The absolute value of the amplitude of the AC component of the priority order corresponding to | x _i | and the threshold value T are compared, and if | x _i |> T, proceed to step S304, and if | x _i | ≦ T. It proceeds to step S305.

Then, in step S304, the edge determination result
After setting the code = 1 and outputting the edge determination result in step S307, the edge determination processing ends. In this case, the edge determination result code = 1 is output.

In step S305, the parameter i is incremented, and it is determined in step S206 whether all AC components have been checked. That is, in step S206, i
If <64, the process returns to step S303, and if i ≧ 64, the process proceeds to step S307. In step S307, the edge information data is output, and then the edge information extraction process ends. In this case, the edge determination result code = 0 is output.

As described above, the edge determination circuit 201 outputs the edge determination result, and the selector a204 uses the quantization table c202 when the edge determination result code = 0 and the quantization table c202 when the edge determination result code = 1. The quantization table d203 is selected and the selected quantization table is output. Also, the selector b216
Selects the quantization table e214 when the edge determination result decoded by the Huffman decoding circuit 112 is code = 0, and selects the quantization table f215 when code = 1 and outputs the selected quantization table. .

In the above description and the drawings, an example in which the edge determination result is included in the compressed image data has been described, but the present embodiment is not limited to this. For example, a table of a quantization table used for quantization. A number or the like may be included in the compressed image data.

Also, in the above description and figures, DCT
Although the example in which the edge determination is performed using the data has been described, the present embodiment is not limited to this.
Edge determination can also be performed using the previous image data.

As described above, according to the present embodiment, in addition to the same effects as the first embodiment, by using the quantization table according to the type of image data, it is possible to optimize the type of image data. Since it can be quantized / dequantized, it can compress / decompress image data with little deterioration in image quality.

The present invention may be applied to a system composed of a plurality of devices or an apparatus composed of one device. Further, it goes without saying that the present invention can be applied to the case where it is achieved by supplying a program to a system or an apparatus.

[0094]

As described above, according to the present invention,
It is possible to provide a technique for efficiently holding the additional information required by a person who later handles image data in the image data while suppressing an increase in the overall data amount of the image data. Specifically, when the image is held as a frequency component, it is possible to hold the additional information while suppressing an increase in the entire data amount and reducing the image as much as possible.

In particular, when there is no data corresponding to an edge component in the data string of frequency components, in order to indicate that, both the first and second codes (additional information) are set to zero and the high frequency of that data string is set. It is embedded in the quantized data corresponding to the component. Therefore, frequency conversion of the image data,
When quantization and entropy coding are performed, it is possible to prevent a reduction in the compression rate of image data due to a data operation for holding additional information as much as possible.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration example of an embodiment according to the present invention,

FIG. 2 is a diagram for explaining the frequency domain of DCT,

FIG. 3 is a diagram showing an example of edge information extraction priority order according to the present embodiment;

FIG. 4 is a diagram showing a second example of edge information extraction priority order;

FIG. 5 is a diagram showing an example of a method of creating an edge information code according to the present embodiment,

FIG. 6 is a diagram showing a second example of a method for creating an edge information code,

FIG. 7 is a flowchart showing an example of edge information extraction processing in the edge information encoder of the present embodiment,

FIG. 8 is a diagram showing an example of quantized data and edge information data synthesized by a synthesizing circuit a,

FIG. 9 is a diagram showing an example of DCT data corrected by a synthesis circuit b,

FIG. 10 is a block diagram showing a second configuration example of the present embodiment,

FIG. 11 is a block diagram showing a configuration example of a second embodiment according to the present invention,

FIG. 12 is a diagram showing an example of an edge determination target block according to the second embodiment,

FIG. 13 is a flowchart showing an example of edge information determination processing in the edge determination circuit of the second embodiment,

FIG. 14 is a diagram showing a standard data format of the ADCT compression method,

FIG. 15 is a block diagram showing a configuration example of an ADCT compression / expansion circuit,

FIG. 16 is a diagram showing a specific example of subsampling,

FIG. 17 is a diagram showing a state of DCT,

FIG. 18 is a diagram showing DCT of 2 × 2 block image data;

FIG. 19 is a diagram showing a DCT base image of 8 × 8 blocks;

20 is a diagram showing a state of quantization in the ADCT compression / decompression circuit shown in FIG. 15,

FIG. 21 is a diagram showing a quantization table recommended by JPEG;

FIG. 22 is a diagram showing an example of a zigzag scan;

23 is a diagram showing a result of dequantization in the ADCT compression / decompression circuit shown in FIG. 15;

FIG. 24 is a diagram showing an example of mosquito noise.

─────────────────────────────────────────────────── ─── Continuation of the front page (56) Reference Japanese Patent Laid-Open No. 4-87460 (JP, A) Yasuhiro Nakamura, Ryoichi Hayasako, Koshio Matsui, Image and Character Composition Method Using Direct Transformation, IPSJ 36th National Congress (Showa 1988), Japan, Information Processing Society of Japan, March 16, 1984, Proceedings (II) 3W-9, page 1841-1842 Yasuhiro Nakamura, Kushio Matsui, Discrete Synthetic coding of grayscale image and text data using orthogonal transformation, IEICE Transactions, Japan, The Institute of Electronics, Information and Communication Engineers, March 25, 1989, Vol. 72-D II, No. 3, pp. 363-368 (58) Fields investigated (Int.Cl. ⁷ , DB name) H04N 1/41

Claims (8)

(57) [Claims] 1. A conversion unit for converting image data into a data string of frequency components , and a quantum for quantizing the data string obtained by the conversion unit.
And one data corresponding to an edge component from the data sequence
A first code to extract and identify the data, and
And a second code indicating the approximate value of the data before quantization.
Quantized data obtained by the quantization means
Some quantized data corresponding to high frequency components in the sequence
And replacing means for replacing, the quantized data sequence after substitution possess an encoding <br/> means for entropy encoding, said replacement means, the edge in the data string of the frequency components
If there is no data corresponding to the
An image coding apparatus characterized in that both the second code and the second code are set to zero . 2. The image data is converted into a data string of frequency components.
Then, the data string obtained by the conversion is quantized, and one data corresponding to an edge component is obtained from the data string.
A first code to extract and identify the data, and
And a second code indicating the approximate value of the data before quantization.
Of the quantized data sequence obtained by the quantization
Of these, some quantized data corresponding to high frequency components
In other words, each step for entropy coding the quantized data sequence after replacement
In the data string of the frequency component in the extraction.
If there is no data corresponding to the
And the second code are both zeroed
Image coding method. 3. Converting image data into a data string of frequency components
Conversion means for converting, and the amount to quantize the data string obtained by the conversion means
In the child substituting means and the data string, the value is equal to or greater than a predetermined value, and the amount
Identifies one data that satisfies the condition that the value after child formation is zero
The first code to do and an overview of the data before quantization.
Generating means for generating a non-zero second code indicating an approximate value
And a quantized data string obtained by the quantizing means.
Then, a part of the quantized data corresponding to the high frequency component is
Replacement means for replacing with the first and second codes, and encoding for entropy coding the quantized data string after replacement
And a means for replacing, in the data string of the frequency component,
If there is no data that meets the above conditions,
An image coding apparatus characterized in that both the second code and the second code are set to zero . 4. The data of the frequency component is generated by the generation means.
In the data sequence, the high-frequency component data is converted from the low-frequency component data.
The above conditions are judged in the order of the general direction to the computer, and the above conditions are satisfied.
Only for the first data
The image encoding device according to claim 3 , wherein the image encoding device generates a code . 5. The image data comprises 8 × 8 pixels,
The transform is an orthogonal transform, and the entropy coding is Huffman coding. 1, 3, and 4
An image coding apparatus described in any one of 1. 6. The image data is converted into a data string of frequency components.
The data string obtained by the conversion is quantized , and the value of the data string is equal to or more than a predetermined value
Identifies one data that satisfies the condition that the value after child formation is zero
The first code to do and an overview of the data before quantization.
Of the quantized data strings obtained by generating the second code other than zero, which indicates an approximate value, and the quantization,
Part of the quantized data corresponding to the frequency component is
And the second code, and each step for entropy-encoding the quantized data string after replacement
And there is no data that meets the above conditions in the generation.
If not, both the first and second codes are zeroed.
Image coding method, characterized in that that. 7. A low data string of the frequency component
Outline of going from high frequency component data to high frequency component data
The conditions are determined in order, and the first data that satisfies the conditions is
The first and second codes are generated only for
The image encoding according to claim 6, characterized in that
Method. 8. The image data consists of 8 × 8 pixels,
The transform is an orthogonal transform, and the entropy coding is a Hough transform.
Man coding, characterized in that it is Man coding.
The image coding method described in any one of 1.