JP3392949B2 - Image compression device and image decompression device - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention
The present invention relates to an image compressing device and an image decompressing device for recording and reproducing on a recording medium according to the EG algorithm.

[0002]

2. Description of the Related Art A standardized algorithm for encoding high-resolution images and exchanging information via a communication transmission line is JPE.
Recommended by G (Joint Photographic Expert Group). Algorithm recommended by this JPEG,
That is, the color still image handled in the JPEG algorithm can be composed of a plurality of components. For example, when a color still image is used by the luminance signal Y and the color difference signals Cr and Cb, only three components are needed. The component is not used.

[0003]

On the other hand, in the baseline process of the JPEG algorithm, in order to perform a large amount of information compression, the original image data is first decomposed into the components on the spatial frequency axis by the two-dimensional DCT transform, and , Quantizes each data represented on the spatial frequency axis, and further encodes each quantized data. Therefore, the two-dimensional DCT
Since a part of the information contained in the original image is lost due to the information compression by the conversion and the quantization, the receiving side cannot restore the complete original image.

The present invention provides an image compression apparatus which makes it possible to reproduce an image as close to the original image as possible, and an image expansion apparatus for reproducing a still image from the data obtained by this image compression apparatus. Is intended.

[0005]

An image compression apparatus according to the present invention performs an orthogonal transformation on original image data composed of at least one image component, and for each component,
Orthogonal transformation means for obtaining an orthogonal transformation coefficient corresponding to a predetermined spatial frequency, quantization means for quantizing the orthogonal transformation coefficient by a quantization table, and dequantization for the quantized orthogonal transformation coefficient Inverse quantization means for obtaining the restored orthogonal transform coefficient, first error data generation means for obtaining, for each component, quantized error data that is the difference between the restored orthogonal transform coefficient and the orthogonal transform coefficient, the quantized error data and the quantum It is characterized in that it is provided with an encoding means for encoding compressed orthogonal transform coefficients to form compressed image data and a recording means for recording the compressed image data on a recording medium.

The image decompressing device according to the present invention reads the compressed image data recorded by the image compressing device from the recording medium, and decodes the quantized orthogonal transform coefficient and the fourth component from the compressed image data. Dequantizing means for dequantizing the quantized orthogonal transform coefficient to restore the orthogonal transform coefficient; error data separating means for separating the error data of each component forming the compressed image data from the fourth component; It is characterized in that it is provided with an adding means for adding the error data to the transform coefficient, and an inverse orthogonal transform means for performing an inverse orthogonal transform on the orthogonal transform coefficient output from the adding means to restore the image data.

[0007]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below with reference to illustrated embodiments. 1 is a block diagram of an image compression apparatus according to a first embodiment of the present invention.

The light coming from the subject S is collected by the condenser lens 11
The image of the subject is collected by the CCD (solid-state image sensor)
An image is formed on 12 light receiving surfaces. A large number of photoelectric conversion elements are arranged on the light receiving surface of the CCD 12, and color filters composed of, for example, R, G, and B color filter elements are provided on the upper surface of the photoelectric conversion elements. Each photoelectric conversion element corresponds to one pixel data. The subject image is converted into an electric signal corresponding to a predetermined color by each photoelectric conversion element and input to the A / D converter 13. In the configuration of FIG. 1, only one CCD 12 is provided, but two or more CCDs are used.
May be provided.

The signal A / D converted by the A / D converter 13 is converted into a luminance signal Y by a signal processing circuit (not shown).
Is converted into color difference signals Cb and Cr and input to the image memory 14. The image memory 14 is divided into mutually independent memory areas for storing the luminance signal Y and the color difference signals Cb, Cr, and each memory area has a storage capacity for one image.

The luminance signal Y and the color difference signals Cb and Cr read from the image memory 14 are input to the DCT processing circuit 21 for data compression processing. In the DCT processing circuit 21, the original image data such as the luminance signal Y is discrete cosine transformed (hereinafter referred to as DCT). That is, in this embodiment, the DCT transform is used as the orthogonal transform of the original image data. Although the DCT processing circuit 21 is shown as one processing circuit in FIG. 1, an independent DCT processing circuit is actually provided for each of the luminance signal Y and the color difference signals Cb and Cr.

DCT processing circuit 21, quantization processing circuit 2
2, Huffman coding processing circuit 23, inverse quantization circuit 24,
In the image compression apparatus including the error data synthesizing unit 25 and the like, image data such as the luminance signal Y is divided into a plurality of blocks for one screen and processed in block units. Each block is composed of 8 × 8 pixel data.

The DCT coefficients of the luminance signal Y and the color difference signals Cb and Cr obtained by the DCT processing circuit 21 are input to the quantization processing circuit 22, respectively. Similarly to the DCT processing circuit 21, the quantization processing circuit 22 is also provided for each signal. The DCT coefficients of the luminance signal Y and the color difference signals Cb and Cr input to the quantization processing circuit 22 are quantized by a quantization table composed of 8 × 8 quantization coefficients. This quantization is a linear quantization,
That is, each DCT coefficient is divided by the corresponding quantized coefficient.

In this embodiment, the quantization table for quantizing the DCT coefficient of the luminance signal Y and the quantization table for quantizing the DCT coefficient of the color difference signals Cb and Cr are different in accordance with the JPEG algorithm. However, the same quantization table may be used for each signal.

The quantized DCT coefficients of the luminance signal Y and the color difference signals Cb and Cr output from the quantization processing circuit 22 are input to the Huffman coding processing circuit 23 and Huffman coded by a predetermined algorithm. The image signal (Y ', Cb', Cr ') obtained by this Huffman coding is the first
~ It is recorded in the IC memory card (recording medium) M as the scan data of the third component.

The quantized DCT coefficient is input to the Huffman encoding processing circuit 23, and is also input to the inverse quantization processing circuit 24 and inversely quantized. That is, the quantized DCT coefficient is multiplied by the quantized coefficient of the quantization table in the inverse quantization processing circuit 24, and thereby the restored DCT coefficient is obtained. The restored DCT coefficient and the DCT coefficient output from the DCT processing circuit 21 are input to subtractors 26, 27 and 28, and the difference between these coefficients, that is, the quantization error data (ΔY ′, ΔCb ′, ΔCr ′). ) Is calculated. This quantized error data is stored in the error data synthesis unit 25 as
The data is quantized again as necessary to limit the maximum number of bits of data, Huffman-encoded in the Huffman-encoding processing circuit 23, and then recorded in the IC memory card M as scan data of the fourth component.

Information of the quantization table used in the quantization processing circuit 22 and the Huffman coding table of the Huffman coding processing circuit 23 when the image data is recorded in the IC memory card M is also included in the image data. And is recorded in the IC memory card M.

FIG. 2 is a flowchart of a processing program for compressing image data and recording it on the IC memory card M. In step 100, the analog color signals R, G, B read from the CCD 12 are converted into digital data by the A / D converter 13, and converted into a luminance signal Y and color difference signals Cb, Cr by a signal processing circuit. Are input to the image memory 14. When the data for one image is stored in the image memory 14, in step 101, the luminance signal Y and the color difference signals Cb and Cr are divided into blocks of 8 × 8 pixels and read from the image memory 14 to obtain the block image data. It is input to the DCT processing circuit 21 as P (Y) xy, P (Cb) xy, P (Cr) xy. In step 102, in the DCT processing circuit 21, image data P (Y) x is obtained for each block.
y, P (Cb) xy, P (Cr) xy are two-dimensional DC according to the equation (1)
T converted.

[Equation 1]

Note that P (Y) xy in the equation (1) corresponds to each pixel of the original image data and represents, for example, a luminance signal Y of 256 gradations (8-bit precision) and color difference signals Cb, Cr. In addition, the reason why Ls is subtracted from each pixel value P (Y) xy in the equation (1) is to make the expected value of the DC coefficient closer to zero. As a result of such subtraction, the DC coefficient can be encoded with a shorter code length in Huffman encoding described later,
A higher information compression rate can be expected.

In step 103, the DCT processing circuit 21
8 × 8 DCT coefficients S (Y) uv, S obtained in
(Cb) uv and S (Cr) uv are input to the quantization processing circuit 22,
Quantization is performed using the quantization tables Q (Y) uv, Q (Cb) uv, Q (Cr) uv.

As an example, FIG. 3 shows image data P (Y) xy of a block of 8 × 8 pixels, DCT coefficient S (Y) uv, quantized DCT coefficient R (Y) uv, and quantization table Q. (Y) uv is shown.

The image data P (Y) xy in FIG. 3A is
By the two-dimensional DCT transformation by the equation (1), FIG.
Are converted into 8 × 8 = 64 DCT coefficients S (Y) uv shown in FIG. Among these DCT coefficients, the DCT coefficient S (Y) _{00 at} the position (0,0) is a DC coefficient, and the remaining 63 DT coefficients.
The CT coefficient S (Y) uv is an AC coefficient. The AC coefficient indicates from the coefficient S (Y) ₀₁ or the coefficient S (Y) ₁₀ to the coefficient S (Y) ₇₇ how much higher spatial frequency component is in the image data of the 8 × 8 pixel block. There is. The DC coefficient represents the average value (DC component) of the pixel values of the entire 8 × 8 pixel block. That is, each DCT coefficient S (Y) uv
Respectively correspond to predetermined spatial frequencies.

FIG. 3D shows an example of the quantization table Q (Y) uv used in the quantization processing circuit 21. As such a quantization table Q (Y) uv, for example, the luminance signal Y and the color difference signals Cb and Cr may be different,
In this case, when the JPEG format image data is recorded in the IC memory card M, the contents of the quantization table Q (Y) uv used for quantizing the component data before the scan data for storing each component. Is recorded.

An equation for quantizing the DCT coefficient S (Y) uv using the quantization table Q (Y) uv is defined as follows. R (Y) uv = round (S (Y) uv / Q (Y) uv) {0≤u, v≤
7} Round in this formula means an approximation to the nearest integer. That is, the quantized DCT coefficient R (Y) uv as shown in FIG. 3C is obtained by dividing and rounding off each element of the DCT coefficient S (Y) uv and the quantization table Q (Y) uv. To be

The quantized DCT coefficients R (Y) uv, R (Cb) uv, R obtained in the quantization processing circuit 22 in this way.
(Cr) uv is input to the Huffman coding processing circuit 23 and the inverse quantization processing circuit 24.

Next, the processing of step 104, that is, the Huffman coding in the Huffman coding processing circuit 23 will be described with reference to FIGS. Quantized DC coefficient R
The encoding method is different between (Y) ₀₀ and the quantized AC coefficient (quantized DCT coefficient R (Y) uv other than the quantized DC coefficient R (Y) ₀₀ ). Encoding of the quantized DC coefficient R (Y) ₀₀ is performed as follows.

Firstly, the difference between the current quantized DC coefficient of a block to be coded R (Y) ₀₀ and one previously coded blocks quantized DC coefficient R (Y) ₀₀ is obtained. It is determined which of the groups shown in FIG. 4 the difference value belongs to, and the code word representing the number of the group is obtained from the code table (coding table of DC coefficient) shown in FIG.
For example, the quantized DC of the block currently to be encoded
When the coefficient R (Y) ₀₀ is 16 and the quantized DC coefficient R (Y) _{00 of the} previously encoded block is 25, the difference value is -9, so the group number of FIG. From the table, the number (SSSS) of the group to which the difference value = −9 belongs is determined to be “4”, and the code word of the group number (SSSS) is shown in FIG.
It is judged as "101" from the code table of.

Next, in the group number table of FIG. 4, the number of values in the group is represented by the additional bit. For example, since the difference value = -9 is the seventh smallest value in the group with the group number (SSSS) = 4, the additional bit is "0110". That is, the quantization D of the block currently being encoded
The Huffman code word of the C coefficient R (Y) ₀₀ is "1010110".

On the other hand, encoding of the quantized AC coefficient is performed by the processing routine shown in FIG. First step 12
At 0, 63 quantized AC coefficients are zigzag scanned in the order shown in FIG. 7 and rearranged into one-dimensional array data. Next, at step 122, it is judged whether or not the quantized AC coefficient values arranged one-dimensionally are "0". When the quantized AC coefficient is “0” (referred to as an invalid coefficient), in step 124, the number of consecutive “0” quantized AC coefficients is counted. Thereby, the length of continuous "0", that is, the run length (NNNN) is obtained.

On the other hand, when it is determined in step 122 that the quantized AC coefficient is not "0" (referred to as an effective coefficient), in step 126, the same grouping as that of the quantized DC coefficient is performed and the additional bits are added. Is required. The grouping of the quantized AC coefficients is different from the grouping of the quantized DC coefficients.
This is done for the C coefficient itself. That is, quantization A
When the C coefficient is, for example, "4", the group number (SSSS) "3" is obtained by referring to the table similar to FIG. Further, since the quantized AC coefficient “4” is the fifth from the smallest in the group having the group number (SSSS) = 3, the additional bit is “100”.

In step 130, an AC code table (not shown) is referred to. For example, if the run length of the data immediately before the quantized AC coefficient "4" is "0", this run length and group number (SSSS ) = 3 and the codeword "1
00 "is obtained. Then, the two-dimensional Huffman code word "100100" is obtained by combining the code word "100" and the additional bit "100" obtained in step 126.

The result of Huffman coding the quantized DCT coefficient shown in FIG. 3C is shown as coded data 99 in FIG.

Referring again to FIG. 2, in step 110, the encoded data obtained in step 104 is
Further, predetermined data is added to generate JPEG format data. In step 111, this data and the scan data of the fourth component composed of the quantization error data described later are sequentially recorded in the IC memory card M according to the JPEG format.

Here, the JPEG format will be described. 9 and 10 show the structure of the image data 80 configured in the JPEG format. As shown in these drawings, in the image data 80 configured in the JPEG format, the encoded data of one color still image is contained in at least one frame 84. The frame 84 is an SOI (Start of Start) indicating the beginning of encoded data.
Image) marker 82 and EOI indicating the end of encoded data
(End of Image) It is sandwiched by the marker.

A frame header 86 is inserted following the SOI marker 82, and the frame header 86 includes various parameters related to frame information. “DHT” included in the frame header 86 indicates the contents of the Huffman code table, and “DQT” indicates the contents of the quantization matrix. Following the frame header 86, a scan header 94 indicating the start of scanning and a scan 88 are provided. Encoded data is recorded in this scan 88.

The term "scan" means rearrangement of the order of the transform coefficients, which is performed when the transform coefficients obtained by DCT transforming the image data are Huffman-encoded. There are various methods for this scan, and there are non-interleaved scan and interleaved scan. For example, the conversion coefficients An = A1, A2 ... of each block of the component A
.., Bn = B1, B2 ..., Cn = C1, C2 ...
・ When, the non-interleaved scan is obtained by sequentially rearranging each transform coefficient independently for each component,
A1, A2 ..., B1, B2 ..., C1, C2 ...
-, And at this time, there are three scans. On the other hand, the interleaved scan is obtained by rearranging the conversion coefficients of corresponding pixel positions in each component so as to be continuous, and A1, B1, C1, A2, B2, C2 ...
・ There is one scan at this time.

Now, the first to the third of the JPEG format
The scan data (first to third components) is encoded data of the luminance signal Y and the color difference signals Cb and Cr to which the scan header 94 (FIG. 10) is added, and the fourth component is quantized error data. Is Huffman-encoded and stored. This process is described below.
It is executed in steps 106 to 111 of FIG.

In step 106, the inverse quantization processing circuit 2
4, the quantized DCT coefficients R (Y) uv, R (Cb) uv, R
(Cr) uv is the quantization table Q (Y) uv, Q (Cb) uv, Q (Cr) uv
Inversely quantized by using the restored DCT coefficient S '(Y) uv, S'
(Cb) uv and S '(Cr) uv are obtained. FIG. 11B shows a restored DCT coefficient S ′ obtained by dequantizing the quantized DCT coefficient R (Y) uv of FIG. 3C with the quantization table Q (Y) uv of FIG. 3D. An example of (Y) uv is shown.

In step 107, the restored DCT coefficient S '
(Y) uv, S '(Cb) uv, S' (Cr) uv are subtracters 26, 27,
DCT coefficients S (Y) uv, S (C
b) uv and S (Cr) uv (see FIG. 3B), that is, the quantization error data ΔYuv, ΔCbuv, and ΔCruv are the following (2),
It is calculated by the equations (3) and (4). ΔYuv = S (Y) uv-S '(Y) uv = S (Y) uv-R (Y) uv * Q (Y) uv (2) ΔCbuv = S (Cb) uv-S' (Cb) uv = S (Cb) uv-R (Cb) uv ・ Q (Cb) uv (3) ΔCruv = S (Cr) uv-S '(Cr) uv = S (Cr) uv-R (Cr) uv ・ Q (Cr ) uv (4)

At step 108, the quantization error data Δ
After Yuv, ΔCbuv, and ΔCruv are input to the error data synthesizing unit 25 and zigzag scanned (see FIG. 7) respectively,
As described below, bit allocation is performed for each of the luminance signal Y and the color difference signals Cr, Cb, and error data is obtained.

The maximum number of bits of one unit data before Huffman coding is 12 in the baseline process of the JPEG algorithm. That is, each value of the quantized DCT coefficient input to the Huffman coding processing circuit 23 is 1 at maximum.
It can be a 2-bit number. However, only the DC coefficient has a maximum of 12 bits, and the AC coefficient has a maximum of 11 bits.

On the other hand, the quantization error data ΔYuv, ΔCbuv,
The numerical value that ΔCruv can take is an integer that satisfies the following expressions (5) to (10).

[0042]   −Q (Y) uv / 2 ≦ ΔYuv <Q (Y) uv / 2 (S (Y) uv ≧ 0) (5)   -Q (Y) uv / 2 <ΔYuv ≦ Q (Y) uv / 2 (S (Y) uv <0) (6)   −Q (Cb) uv / 2 ≦ ΔCbuv <Q (Cb) uv / 2 (S (Cb) uv ≧ 0) (7)   -Q (Cb) uv / 2 <ΔCbuv ≦ Q (Cb) uv / 2 (S (Cb) uv <0) (8)   -Q (Cr) uv / 2 ≦ ΔCruv <Q (Cr) uv / 2 (S (Cr) uv ≧ 0) (9)   -Q (Cr) uv / 2 <ΔCruv ≦ Q (Cr) uv / 2 (S (Cr) uv <0) (10)

As can be understood from these equations (5) to (10), the numerical ranges of the quantization error data ΔYuv, ΔCbuv, ΔCruv are the quantization tables Q (Y) uv, Q (Cb) uv, Q ( Cr)
It depends on the value of uv. Furthermore, the quantization error data ΔYuv, ΔCbuv, maximum number of bits B _Y which can take ΔCruv, B _Cb, B _Cr is defined by the following (11) to (13) below.

B _Y = round (log ₂ (Q _Y −1) +0.5) (11) B _Cb = round (log ₂ (Q _Cb −1) +0.5) (12) B _Cr = round (log ₂ (Q _Cr -1) + 0.5) (13)

Q _Y , Q _Cb , and Q _Cr in these equations are
It is the maximum value in each quantization table Q (Y) uv, Q (Cb) uv, Q (Cr) uv. For example, in the case of the quantization table Q (Y) uv in FIG. 3D, Q _Y = 121. Therefore, B _Y
= Round (6.90689 + 0.5) = 7, the quantization error data Δ
When Yuv is expressed in binary, the maximum number of bits is 7 bits. In this way, each quantization error data ΔYuv, ΔCbu
The number of bits of v and ΔCruv is the quantization table Q (Y) uv, Q (C
Since b) uv and Q (Cr) uv are uniquely determined, the sum of the bit numbers of the quantization error data ΔYuv, ΔCbuv, and ΔCruv, that is, B _Y obtained by the equations (11) to (13),
The sum of B _Cb and B _Cr may exceed 11 bits.

Therefore, in this embodiment, (11)-
The sum of B _Y , B _Cb , and B _Cr obtained by the equation (13) is 1
If it exceeds 1 bit, each quantization error data ΔYuv,
ΔCbuv and ΔCruv are requantized by the following equations (14) to (16), and the number of bits is limited so that the total sum of these bits is 11 bits or less.

ΔY'uv = rol (ΔYuv / 2 ^(BY-k) ) (14) ΔCb'uv = rol (ΔCbuv / 2 ^(BCb-m) ) (15) ΔCr'uv = rol (ΔCruv / 2 ^{(BCr -n)} ) (16) However, k + m + n = 11, for example, k = 7, m =
n = 2. Also, rol means rounding down after the decimal point.

Expressions (14) to (16) show the simplest method of requantization of the quantization error data ΔYuv, ΔCbuv, and ΔCruv.

In step 108, first, the number of bits for storing the quantization error data ΔYuv, ΔCbuv, ΔCruv is assigned as described below.

Of the luminance signal Y and the color difference signals Cb and Cr, it is the luminance signal Y that most affects the visual characteristics of the reproduced image. Therefore, the error data ΔY of the luminance signal Y
The number of bits assigned to uv is made larger than the number of bits assigned to the error data ΔCbuv and ΔCruv of the color difference signals Cb and Cr. For example, the number of bits assigned to the error data ΔYuv of the luminance signal Y is 7 bits, and the number of bits assigned to the error data ΔCbuv and ΔCruv of the color difference signals Cb and Cr is 2 bits, respectively.

FIG. 12 is a flowchart of the subroutine executed in step 108 for allocating bits and synthesizing error data.

In step 140, the number of bits of each quantization error data ΔYuv, ΔCbuv, ΔCruv is (11) to (1).
It is calculated by the equation 3). The sum of these bit numbers is 1
When it is 1 or less, and when the number of bits of the quantization error data ΔYuv of the luminance signal Y is 7 or less even if the total sum is greater than 11, the quantization error data ΔYuv is not requantized. On the other hand, the total number of bits is larger than 11, and the number of bits of the quantization error data ΔYuv is 7
If it is larger than, the quantization error data ΔYuv becomes (1
It is requantized by the equation 4). That is, the luminance signal Y
The number of bits for the quantized error data ΔYuv is determined. Further, when the total number of bits is larger than 11, but the number of bits of the quantization error data ΔYuv is 7 or less, only the quantization error data ΔCbuv and ΔCruv are (15), (1
It is requantized by the equation (6).

In step 142, the quantization error data ΔCbu of the color difference signals Cb and Cr is calculated based on the number of bits of the quantization error data ΔYuv determined in step 140.
The number of bits for v and ΔCruv is assigned. For example, if the number of bits of the quantization error data ΔYuv is 14
If 0 is set to 7, then in step 142,
Quantization error data ΔCbu according to equations (15) and (16)
Requantization is performed so that the number of bits of v and ΔCruv becomes (11−7) / 2 = 2.

In step 144, steps 140, 1
The quantized error data obtained at 42 is zigzag scanned (see FIG. 7) to generate the fourth component ΔKuv.

This fourth component ΔKuv is processed by the Huffman coding processing circuit 23 in step 109 of FIG.
And is Huffman encoded in the same manner as the Huffman encoding of the first to third components. Step 110
Then, the Huffman-encoded fourth component ΔKuv
Is J together with the encoded data of the first to third components
The data is recorded on the IC memory card M in the form of data according to the PEG format.

FIG. 13 shows another example of a flowchart of a subroutine of bit allocation and error data combination.

In step 146, step 10 in FIG.
Quantization error data ΔYuv, ΔCb obtained by 7.
The number of bits of uv and ΔCruv is calculated by the equations (11) to (13). Then, based on these bit numbers, FIG.
In the same manner as in steps 140 and 142 of step 2, the total number of bits is 11 or less, and the number of bits of the quantization error data ΔYuv of the luminance signal Y is the largest.
The number of bits of each quantization error data ΔYuv, ΔCbuv, ΔCruv is determined.

In step 148, step 10 in FIG.
In the Huffman encoding of the quantized error data in 9, in order to perform the information compression more efficiently,
The positive and negative signs of the quantization error data are adjusted. This adjustment will be described with reference to FIG.

For example, the original DCT coefficient S (Y) uv "95"
When "104" is quantized by the quantization coefficient "10", the quantized DCT coefficient is "10", and the quantization error data Q (Y) uv / 2 is "-5" to "4". On the other hand, Hara D
The occurrence frequency of each numerical value of the CT coefficient is higher as the absolute value is smaller, regardless of whether the numerical value is positive or negative. Therefore, the occurrence frequency distribution (histogram) of quantization error data is
As shown in FIG. 14A, -Q (Y) uv / 2 to Q (Y) uv /
It gradually decreases to 2. On the other hand, for example, the original DCT coefficient S
(Y) uv “−104” to “−95” is quantized coefficient “10”
, The quantized DCT coefficient is "-10".
And the quantized error data Q (Y) uv / 2 is from "-4" to
It becomes "4". Therefore, the histogram of the quantization error data is, as shown in FIG. 14B, from the larger absolute value of the original DCT coefficient to the smaller absolute value, that is, -Q.
It gradually increases from (Y) uv / 2 to Q (Y) uv / 2.

As described above, the histogram of the quantized error data shows the opposite tendency depending on the positive / negative sign of each numerical value of the original DCT coefficient. Therefore, when the data obtained by zigzag scanning the quantized error data is Huffman coded. Since the entropy of the quantization error data is large, the efficiency of information compression is not always good.

Therefore, in step 148, in order to prevent the histogram of the quantized error data from being reversed by the positive / negative sign of the original DCT coefficient, when the original DCT coefficient is negative, the quantization error data of the DCT coefficient has "-". It is multiplied by 1. As a result, the frequency of occurrence of each numerical value of the quantized error data increases as the absolute value decreases, regardless of whether the numerical value of the original DCT coefficient is positive or negative. As a result, in the Huffman coding of the quantized error data in step 109 of FIG. 2, the entropy of the quantized error data becomes small, and the information compression is performed more efficiently.

In step 150, the quantization errors ΔYuv, ΔCbuv, ΔCruv whose codes have been adjusted in step 148 have been adjusted.
Is scaled. In this scaling, a smaller code value is assigned to a numerical value of the quantization error data whose code has been adjusted, with a higher occurrence frequency. For example, the quantization table Q (Y) uv in FIG.
When the CT coefficient S (Y) uv is quantized, if the numerical range of the quantization error data ΔYuv is in the range of −8 to 7, −
8 → “0000”, −7 → “0001”, ... 6 → “1110”,
7 → Replace with "1111".

In step 152, each bit of the requantized quantization error data ΔY'uv, ΔCb'uv, ΔCr'uv according to the bit allocation in step 146 is rearranged as follows.

For example, the quantization error data ΔY'uv, ΔC
The number of bits of b'uv and ΔCr'uv is respectively 6, 3, 2 by the requantization based on the bit allocation in step 146.
Quantized error data ΔY'uv, ΔC
It is assumed that one piece of data b'uv and ΔCr'uv is represented in binary as follows. Quantization error data ΔY'uv = Y (5) Y (4) Y (3) Y (2) Y (1) Y (0) Quantization error data ΔCb'uv = Cb (2) Cb (1) Cb ( 0) Quantization error data ΔCr'uv = Cr (1) Cr (0)

Such quantization error data ΔY'uv, Δ
The fourth component ΔKuv of the quantization error data for Cb'uv and ΔCr'uv is ΔKuv = Y (5) Y (4) Y (3) Y (2) Cb (2) Y (1) Cb (1) Cr (1) Y (0) Cb
Each bit of each quantization error data is rearranged so that it becomes (0) Cr (0).

That is, the quantization error data ΔY'uv, Δ
The number of bits of Cb'uv and ΔCr'uv is k, m, and n (k
When ≧ m ≧ n, k + m + n = 11), as shown in FIG. 15, the quantization error data having the largest number of bits in the bit string 160 from the most significant bit to the (2m + n + 1) th bit of the fourth component ΔKuv. Bit string 166 at the beginning of ΔY'uv (from the most significant bit to the (m +
1) up to bits) are placed.

In the bit string 162 (from the (2m + n) th bit to the (3n + 1) th bit) following the bit string 160, the mth bit to the (nth) th bit of the quantization error data ΔY′uv.
The bit string 168 up to +1) bits and the bit string 168 from the most significant bit to the (n + 1) th bit of the quantized error data ΔCb′uv are alternately arranged in the order of Y and Cb. Then, the last bit string 164 of ΔKuv (the third (3
(n) to the least significant bit), the bit string 170 from the nth bit to the least significant bit of the quantization error data ΔY'uv and the nth bit to the least significant bit of the quantization error data ΔCb'uv. The bit string 170 and all the bit strings 170 of the quantization error data ΔCr'uv are Y, C
They are cyclically arranged in the order of b and Cr.

As described above, in the scaling in step 150, the quantization error data ΔY'uv, ΔCb'uv, ΔC
Since r'uv is replaced with a smaller code value as the frequency of occurrence is higher, the probability that the leading bit of each quantized error data will be 1 becomes lower. Further, in step 152, since the bits in the same place of the quantization error data ΔY'uv, ΔCb'uv, and ΔCr'uv are arranged in the order of Y, Cb, Cr,
The number of components will be smaller. Therefore, the entropy of the quantization error data ΔY′uv, ΔCb′uv, and ΔCr′uv, which are the fourth component ΔKuv, becomes small, and the information compression is efficiently performed.

The bit rearrangement order in the generation of the fourth component ΔKuv is always such that error data having a large number of bits comes first. For example, in the case of the example of FIG. 15, in the bit strings 162 and 164 of ΔKuv, the number of bits of the quantization error data ΔY′uv, ΔCb′uv, and ΔCr′uv is k ≧ m ≧ n, so that ΔY′uv and ΔCb must be set. 'uv, ΔC
Make sure they are in r'uv order. If the number of bits is the same, for example, the quantization error data ΔY'uv, ΔCb'uv,
When ΔCr′uv has the same number of bits, ΔY′uv is always arranged first, and the order is the same for all elements of ΔKuv.

The fourth component ΔKuv thus obtained is Huffman coded by the Huffman coding processing circuit 23 in step 109 of FIG. 2 and is output from the quantization processing circuit 22 to be Huffman coded. 1
~ With the third scan data, the JPEG format image data 80 (see FIG. 10) is used to build the IC
It is recorded in the memory card M (step 110 in FIG. 2,
111). When the fourth component ΔKuv is incorporated in the image data 80, each quantization error data Δ
Information on the number of bits (k, m, n) assigned to Y'uv, ΔCb'uv, and ΔCr'uv is also incorporated.

FIG. 16 is a block diagram of the image compression apparatus of FIG.
3 is a block diagram of an image decompression device that reproduces an image from image data 80 recorded in a C memory card M. FIG. The configuration and operation of the image expansion apparatus will be described with reference to this figure and FIG.

The image data 80 read from the IC memory card M is input to the Huffman decoding processing circuit 31. The Huffman decoding processing circuit 31 extracts the first to fourth scan data 88a to 88d from the image data 80, and also extracts the information on the Huffman coding table used for coding these scan data. Using this Huffman coding table, the first scan data 88a, the second scan data 88b, and the third scan data 88b
The encoded data of the scan data 88c is decoded to obtain the quantized DCT coefficient, and the encoded data of the fourth scan data 88d is decoded to obtain the quantized error data.

The quantized DCT coefficient is the inverse quantization processing circuit 32.
Entered in. The inverse quantization processing circuit 32 restores the DCT coefficient for each of the luminance signal Y and the color difference signals Cb and Cr using the quantization table incorporated in the image data 80. The restored DCT coefficient is input to the adders 34 to 36. The quantization table is the image data 80
It is recorded in.

The quantized error data is stored in the error data decomposition unit 33.
Quantization error data ΔY ′, ΔCb ′, and ΔCr ′ of the luminance signal Y and the color difference signals Cb and Cr are decomposed into each other and input to the adders 34 to 36. In the adders 34 to 36, the luminance signal Y input from the inverse quantization processing circuit 32,
The quantized error data ΔY ′, ΔCb ′, and ΔCr ′ are added to the restored DCT coefficients of the color difference signals Cb and Cr. The data obtained by this addition is stored in the IDCT processing circuit 37 as 8
Two-dimensional inverse DCT conversion is performed for each block of × 8 pixels, and the luminance signal Y and the color difference signals Cb and Cr of the original image data are restored.

The luminance signal Y and the color difference signals Cb, Cr of the original image data are divided into blocks of 8 × 8 pixels and input to the image memory 38. When the data storage for one image is completed in the image memory 38, the data of the luminance signal Y and the data of the color difference signals Cb and Cr are simultaneously read by the same scanning method as the normal video signal and input to the D / A converter 39. It is converted into an analog video signal. The analog luminance signal Y and the color difference signals Cb and Cr are converted into a composite video signal or the like by a circuit (not shown) as necessary and output to the display 40.

Next, the operation of the image decompression device shown in FIG.
This will be described with reference to the flowcharts of FIGS. 17 and 18 and FIG.

In step 170, the image data 80
Is read from the IC memory card M. The first to third components of this image data 80 are step 1
At 72, the quantized DCT coefficients R ′ (Y) uv, R ′ (Cb) uv, R ′ (Cr) uv, which are input to the Huffman decoding processing circuit 31 and are composed of blocks of 8 × 8 pixels, are input. It is decrypted (FIG. 11 (a)). This quantized DCT coefficient R '
(Y) uv, R '(Cb) uv, R' (Cr) uv are input to the inverse quantization processing circuit 32 in step 174 and inversely quantized using the quantization table incorporated in the image data 80. To be done. As a result, the DCT coefficients S '(Y) uv, S' (Cb) uv
, S ′ (Cr) uv are restored (FIG. 11 (b)).

On the other hand, the fourth component ΔKuv decoded by the Huffman decoding processing circuit 31 is
At 82, it is input to the error data decomposition unit 33. In the error data decomposition unit 33, the quantized error data ΔY'uv
, ΔCb'uv, ΔCr'uv are obtained. Also step 1
At 82, if the quantized error data is requantized, dequantization is performed and the quantized error data ΔY ″ uv,
ΔCb "uv and ΔCr" uv are restored.

The quantization error data ΔY "uv, ΔCb" uv
, ΔCr ″ uv will be described with reference to the flowchart of the processing routine shown in FIG.

In step 184, the image data 80
Based on the allocated bit number (k, m, n) of each quantized error data incorporated in, the bit string of the fourth component ΔKuv having the form as shown in FIG. 15 is obtained for each of the luminance signal Y, the color difference signals Cb, Cr. And the quantization error data ΔY "uv, ΔCb" uv, and ΔCr "uv are extracted.At step 186, at step 150 of FIG. The scaling corrections performed are performed, for example, suppose that the quantization error data ΔY′uv is changed from 8 bits to 6 bits by requantization according to equation (14), and the decoded data is “000000”. To do. Quantization coefficient Q _Y =
245, when the quantization error ΔYuv> 0, “000000” means that the requantization error is the minimum (negative and the absolute value is the largest), and this requantization error data is ΔY ″ uv = −Q _Y / 2/2
² = −30. Therefore, the quantization error data ΔY'uv before being requantized is multiplied by "-4" by "-30" in order to correct it from 6 bits to 8 bits.
It becomes 0.

After the scaling correction, in step 188, B _Y obtained by the equations (11) to (13),
Based on B _Cb and B _Cr and the number of bits (k, m, n), the quantized error data ΔY ″ uv, ΔCb ”uv and ΔCr” uv are obtained, where the quantized error data ΔYuv and ΔCbuv, ΔCr
If the uv has been quantized, the quantization error data ΔY "uv, ΔCb" uv, ΔCr "uv after the inverse quantization is performed.
Is required. Further, the DCT transform coefficients S ′ (Y) uv, S ′ (Cb) uv, S ′ (C
r) From the positive and negative signs of uv, the quantization error data ΔY "uv,
The signs of ΔCb "uv and ΔCr" uv are corrected.

Referring again to FIG. In step 176, the quantization error data ΔY ″ uv, ΔCb ”uv, and ΔCr” uv obtained in step 188 are added to the adders 34, 35,
The restored DCT coefficients S ′ (Y) uv, S ′ (Cb) uv, and S ′ (Cr) uv input to the inverse quantization processing circuit 32 are added to the input 36. The DCT coefficient S '(Y) uv, S' that is the result of this addition
(Cb) uv, S ′ (Cr) uv is 8 × in step 178.
The inverse discrete cosine transform represented by the following equation (17) is applied to each block of 8 pixels, and the image data P ′ (Y) is thereby obtained.
xy, P '(Cb) xy, P' (Cr) xy are restored (Fig. 11
(C)).

[0083]

[Equation 2]

Restored image data P '(Y) xy, P' (Cb) xy,
P ′ (Cr) xy is sequentially written and accumulated in the image memory 38 for each block. When the image data for one image is stored in the image memory 38, in step 180, the image data is read from the image memory 38 in the same order as the original image data writing to the image memory 14 in FIG. After being converted into an analog video signal in the A converter 39,
It is displayed on the display 40.

As described above, in this embodiment, the original image information lost by the quantization or DCT conversion is stored as error data in the fourth component of the JPEG format and recorded in the IC memory card M. Therefore, the image can be restored by a device that reproduces the image using only the first to third components, and FIG.
With the reproducing device shown in FIG. 1, it is possible to reproduce higher quality images.

In the present embodiment, the error data stored in the fourth component is assigned a large number of bits for the luminance signal Y that most affects the visual characteristics of the reproduced image. It is possible to further reduce the deterioration of image quality due to. Further, since the processing for reducing the entropy of the fourth component is performed, the memory occupation amount of the IC memory card M due to the increase of the fourth component can be minimized.

FIG. 19 is a block diagram of an image compression apparatus according to the second embodiment of the present invention. In the second embodiment,
Unlike the first embodiment of FIG. 1, the color signals R, G and B output from the A / D converter 13 are directly stored in the image memory 14, and the DCT processing circuit 21, the quantization processing circuit 22 and the dequantization are performed. In the processing circuit 24, the bit allocation processing / error data synthesizing unit 25, and the Huffman coding processing circuit 23, each color signal R, G, B is coded and recorded in the IC memory card M. The DCT processing circuit 21 and the quantization processing circuit 2
2. The processing in the inverse quantization processing circuit 24, the bit allocation processing / error data synthesis unit 25, and the Huffman coding processing circuit 23 is the same as that in the first embodiment.

FIG. 20 is a block diagram of the image expansion apparatus of the second embodiment. In this image decompression device, the JPE recorded in the IC memory card M by the image compression device of FIG.
From the G format image data 80, the color signal R,
The image composed of G and B is reproduced. In this figure, the circuits corresponding to those of the image expanding apparatus of the first embodiment of FIG. 16 are designated by the same reference numerals.

The image expanding apparatus of the second embodiment is such that the color signals R,
This is the same as the first embodiment except that G and B are processed.

FIG. 21 is a block diagram of an image compression apparatus according to the third embodiment of the present invention. In this figure, the same circuits as those in the first and second embodiments are designated by the same reference numerals.

In the third embodiment, the quantization error data stored in the fourth component of the JPEG format is
It is obtained from the difference between the original image data before the DCT transformation and the image data restored from the quantized DCT coefficient after the DCT transformation and the quantization. In other words, the information of the original image data lost by the DCT transform and the quantization is stored in the fourth component as error data.

That is, the data of the luminance signal Y and the color difference signals Cb and Cr of the original image read out from the image memory 14 are processed by the DCT processing circuit 21 and the quantization processing circuit 22 and inverse quantized as quantized DCT coefficients. Conversion processing circuit 24
Entered in. The quantized DCT coefficient is processed by the inverse quantization processing circuit 24 and the IDCT processing circuit 37, and is input to the subtracters 26, 27 and 28 as restored image data of the luminance signal Y and the color difference signals Cb and Cr. On the other hand, the luminance signal Y, which is the original image data read from the image memory 14,
The color difference signals Cb, Cr are directly input to the subtractors 26, 27, 28, and therefore the error data which is the difference between the original image data and the restored image data is obtained for each pixel.

In the data generator 41, the error data of the luminance signal Y and the color difference signals Cb and Cr are quantized as necessary and rearranged to 11 bits as in the first and second embodiments. It is converted into the fourth component of fixed length data and input to the Huffman coding processing circuit 23. The error data Huffman-encoded by the Huffman-encoding processing circuit 23 is recorded in the IC memory card M as the fourth scan data in the JPEG format.

FIG. 22 is a block diagram of an image expansion device in the third embodiment. Each scan data of the JPEG format read from the IC memory card M is processed by the Huffman decoding processing circuit 31, and the quantized DCT coefficient of the luminance signal Y and the color difference signals Cb, Cr and the error data are decoded. The quantized DCT coefficient is restored by the inverse quantization processing circuit 32 based on the quantization table information incorporated in the image data 80, and the restored DCT coefficient is restored.
The coefficient is input to the IDCT processing circuit 37. Restore D
The CT coefficient is subjected to inverse DCT conversion in the IDCT processing circuit 37, whereby the luminance signal Y and the color difference signals Cb and C are obtained.
r is required. The luminance signal Y and the color difference signals Cb and Cr
Is input to the adders 34, 35 and 36. On the other hand, the error data output from the Huffman decoding processing circuit 31 is separated into error data of the luminance signal Y and the color difference signals Cb and Cr for each pixel position in the data separation unit 42, and the adder 3
4, 35, 36 are input.

In the adders 34, 35 and 36, the luminance signal Y, the color difference signals Cb and Cr and the error data corresponding to these are added to each other and written sequentially in the image memory 38. When the data of the luminance signal Y and the color difference signals Cb and Cr for one image is accumulated in the image memory 38, it is converted into an analog signal in the D / A converter 39 and displayed as in the first and second embodiments. 40 is output.

FIG. 23 shows the quantization error data and the like from the first, second and third components of the image data 80 recorded in the IC memory card M by the image compression apparatus of the first, second and third embodiments. The block block diagram of the image expansion apparatus which reproduces an image without using is shown.

23, only the first to third scan data 88a, 88b, 88c of the image data 80 read from the IC memory card M are separated and input to the Huffman decoding processing circuit 31. It In the Huffman decoding processing circuit 31, the scan data 88a-8
The quantized DCT coefficient is restored from 8c in the same manner as in the first, second and third embodiments, and is input to the inverse quantization processing circuit 32. Quantized DCT input to the inverse quantization processing circuit 32
The coefficient is restored to a DCT coefficient using the quantization table and input to the IDCT processing circuit 37. In the IDCT processing circuit 37, the DCT coefficient is inversely DCT-converted for each block of 8 × 8 pixels, and the luminance signal Y and the color difference signals Cb and Cr of the image data are restored. These signals are sequentially input to the image memory 38, and signals for one image are stored in the image memory 38.
When stored in, the D / A converter 39 converts the analog signal into an analog signal and outputs the analog signal to the display 40.

[0098] Now, in the first to third embodiment, the quantization error data ΔYuv, ΔCbuv, maximum number of bits B _Y which can take ΔCruv, B _Cb, the sum of B _Cr, with respect to the quantization error data of the DC coefficients If it exceeds 12 bits, or if it exceeds 11 bits for the quantization error data of the AC coefficient,
The number of bits of each quantization error data was adjusted based on the equations (14) to (16). For example, as shown in FIG. 24A, for the luminance signal Y, the quantization table Q (Y) u
The maximum value of the quantized coefficient of v is "121", and all the constituent elements of the quantized error data ΔYuv have been re-quantized so that the number of bits becomes 7 based on the equation (11). .

On the other hand, for the quantization coefficient "16", for example, the number of bits of the quantization error data ΔYuv is 4 bits, and 7 bits are not necessary to store this. That is, the numerical values of the quantized coefficients are greatly different, and the quantized coefficient having a small numerical value can be re-quantized so as to have a shorter bit number. -It is possible to improve further than the third embodiment. The fourth and fifth embodiments described below have a configuration for more efficiently using the fourth component.

25 and 26 are flowcharts showing the processing routine for generating the quantized error data in the fourth embodiment. In step 201, the parameter v is set to "0", and in step 202, the parameter u is set to "0". The parameters u and v are as shown in FIG.
Pixel positions in the horizontal and vertical directions of the quantization tables shown in (a) and (b) are respectively shown, and the values are "0" to "7". In the quantization table, DC coefficients are stored at pixel positions where both parameters u and v are “0”, and AC coefficients are stored at other pixel positions.

At step 203, the quantization error data Δ
The number of bits B _Y , B _Cb , and B _Cr required to store Yuv, ΔCbuv, and ΔCruv are (11) to
It is calculated using the equation (13). In the case of the DC coefficient of the quantization table shown in FIG. 24, the quantization coefficient Q (Y) ₀₀ = 16,
Since Q (Cb) ₀₀ = Q (Cr) ₀₀ = 17, the required bit numbers are B _Y = 4 and B _Cb = B _Cr =, respectively, as shown in FIG.
It becomes 5. In step 204, the total sum B _{all of} these required number of bits B _Y , B _Cb , and B _Cr is calculated, and D of FIG.
In the case of the C coefficient, the total number of required bits B _all is “14”.

In step 205, the total sum B _all is the set bit number S _Y assigned to the quantization error data ΔYuv.
Then, it is determined whether or not it is larger than the sum S _{all of the} setting bit number S _Cb assigned to the quantization error data ΔCbuv and the setting bit number S _Cr assigned to the quantization error data ΔCruv. The sum S _all of the set number of bits is “11” for the AC coefficient and “12” for the DC coefficient.
Therefore, in the case of the DC coefficient in FIG. 24, the total sum B _all of the required bits is “14”, and the total sum S _{all of the} set bits is S _all.
Therefore, step 206 and the following steps are executed, and it is determined whether or not the required bit numbers B _Y , B _Cb , and B _Cr are larger than the set bit numbers S _Y , S _Cb , and S _Cr , respectively.

As the bit distribution of each quantization error data, regarding the DC coefficient, for example, the quantization error data Δ
Yuv setting bit number S _Y is “6”, quantization error data Δ
The set bit numbers S _Cb and S _Cr of Cbuv and ΔCruv are set to “3”, respectively. Regarding the AC coefficient, for example, the set bit number S _Y of the quantized error data ΔYuv is set to “5”, and the set bit numbers S _Cb and S _Cr of the quantized error data ΔCbuv and ΔCruv are set to “3”, respectively.

At step 206, the quantization error data Δ
It is determined whether the required bit number _{BY of} Yuv is larger than the set bit number S _Y. In the case of the DC coefficient in FIG. 24A, B _Y = 4 and S _Y = 6, the required number of bits is smaller, and it is not necessary to perform requantization.

Therefore, in this case, the routine proceeds to step 208, where it is judged whether or not the required bit number B _Cb of the quantized error data ΔCbuv is larger than the set bit number S _Cb . Figure 24
In the case of the DC coefficient of (b), B _Cb = 5 and S _Cb = 3,
The required number of bits is larger. Therefore, in this case, requantization is performed in step 209. That is, the set bit number S _Cb is defined as the maximum bit number B ′ _Cb of the quantization error data ΔCbuv, and the quantization error data ΔCbuv is divided by the requantization coefficient F (B _Cb , B ′ _Cb ) to be requantized. Be converted. The requantization coefficient F (B _Cb , B ′ _Cb ) has a quantization value for changing the number of bits of the quantization error data ΔCbuv from B _Cb to B ′ _Cb, and the DC coefficient of FIG. In the case of, in order to change from 5 bits to 3 bits,
7, the requantization factor F
(B _Cb , B ′ _Cb ) is set to “5”.

At step 210, the quantization error data Δ
It is determined whether the required number of bits B _{Cr of} Cruv is larger than the set number of bits S _Cr , and the same processing as the quantization error data ΔCbuv is performed. That is, in the case of the DC coefficient of FIG. 24 (b), since the required number of bits is larger, requantization is performed in step 211, and the requantization coefficient F (B _Cr , B '
_Cr ) is "5".

Then, in step 212, a fixed length bit string in which the quantized error data ΔYuv, ΔCbuv, and ΔCruv are arranged in this order is generated. This fixed length is "12" for the DC coefficient and "11" for the AC coefficient. Note that the arrangement in step 212 may be replaced by a cyclic arrangement as shown in FIG. 15, which can reduce entropy. In step 213, the parameter u is incremented by "1". If the parameter u is 7 or less, the process returns from step 214 to step 203, and steps 203 and subsequent steps are executed.

At step 203, the same processing as described above is performed. In the quantization table shown in FIG. 24, for example, when the parameter u is “4” and the parameter v is “0”, the quantization coefficient Q (Y) ₄₀ = 24, Q (Cb) ₄₀ = Q (C
Since r) ₃₀ = 99, the required bit numbers are B _Y = 5 and B _Cb = B _Cr = 7, respectively, as shown in FIG. In step 204, the required number of bits B _Y ,
B _Cb, sum B _all of B _Cr is calculated, the sum B _all of the required number of bits is "24".

At step 205, it is judged whether the total sum B _all is larger than the total sum S _all of the set number of bits.
The sum S _all of the number of set bits is “11” in the case of AC coefficient.
Therefore, the parameters u and v are “4”,
If it is “0”, the total sum B _all of the required number of bits is larger than the total sum S _all of the set number of bits, and step 206 and subsequent steps are executed.

At step 206, the quantization error data Δ
It is determined whether or not the required number of bits _{BY of} Yuv is larger than the set number of bits S _Y. In FIG. 24, the parameter u,
When v is "4" and "0", respectively, the required number of bits B _Y = 5 and the set number of bits S _Y = 5, the required number of bits is equal to the set number of bits, and in this case, requantization is performed. Since it is not necessary, steps 208 et seq. Are executed next.
In steps 208 to 211, the quantization error data ΔCbuv and ΔCruv are requantized similarly to the DC coefficient, and the number of bits of the quantization error data ΔCbuv and ΔCruv is changed from 7 bits to 3 bits.

After the bit string is generated in step 212, the parameter u is set to "7" in step 214.
If it is determined that the parameter v is larger than the value, the parameter v is incremented by “1” in step 215. In step 216, it is determined whether or not the parameter v is larger than “7”. If the parameter v is equal to or smaller than “7”, the process returns to step 202 and the above-mentioned processing is performed.

When it is determined in step 216 that the parameter v is larger than "7", this processing routine ends.

As described above, in the fourth embodiment, when the sum B _{all of the} number of bits required to store the quantized error data ΔYuv, ΔCbuv, ΔCruv is “11” for the AC coefficient, and also for the DC coefficient. When it is "12" or less, the quantization error data is not requantized. When requantizing the quantized error data, the requantized coefficients F (B _Y , B ′ _Y ), F (B _Cb at each pixel position in the quantization table having a value corresponding to the quantized coefficient. , _B'Cb ), F (B _Cr ,
_B'Cr ) is used. Therefore, the quantized error data can be effectively allocated to bits, and requantization of the quantized error data more than necessary is avoided. That is, according to this embodiment, the utilization efficiency of the fourth component of the JPEG format is further improved as compared with the first to third embodiments.

Further, according to the fourth embodiment, the bit allocation of the quantization error data can be easily performed based on only the value of each quantization coefficient in the quantization table (FIG. 24) and the requantization coefficient (FIG. 27). It is determined and there is no need to use complicated formulas.

29 to 31 are flowcharts showing the processing routine for generating the quantized error data in the fifth embodiment. The contents of steps 301 to 304 are shown in FIG.
This is the same as steps 201 to 204 shown in FIG. As in the case of the fourth embodiment, as the bit distribution of each quantization error data, regarding the DC coefficient, the set bit number S _Y of the quantization error data ΔYuv is “6”, the quantization error data ΔCbuv,
The set bit numbers S _Cb and S _Cr of ΔCruv are each set to “3”, and regarding the AC coefficient, the quantization error data Δ
Yuv setting bit number S _Y is “5”, quantization error data ΔC
The set bit numbers S _Cb and S _Cr of buv and ΔCruv are each set to “3”.

At step 305, it is judged whether the total sum B _all of the required number of bits is larger than the total sum S _all of the set number of bits. When the total sum B _all of the required number of bits is less than or equal to the total sum S _all of the set number of bits, steps 312 and below are executed without requantizing the quantization error data. The contents of steps 312 to 316 are the same as steps 212 to 216 shown in FIG.

The total sum B _all of the required bits is larger than the total sum S _all of the set bits, and each required bit number B _Y ,
B _Cb, B _Cr number setting bits each S _Y, if S _Cb, greater than S _Cb, step 305,306,308,3
It is executed in the order of 10 and 320. That is, all the quantization error data ΔYuv, ΔCbuv, ΔCruv are requantized,
Then, in step 312, the quantization error data Δ
A fixed-length bit string arranged in the order of Yuv, ΔCbuv, and ΔCruv is generated.

The operation described so far is the same as that of the fourth embodiment.

In the case of the DC coefficient of the quantization table shown in FIG. 24, the required number of bits is B _Y =
4, B _Cb = B _Cr = 5 (see FIG. 28). Step 3
In 04, the total number B _{all of} the required bits B _Y , B _Cb , and B _Cr is B _all.
Is calculated and “14” is obtained. In step 305, the total sum B _all of the required bits is the total sum S of the set number of bits.
Since it is determined that it is larger than _all , step 306 is executed. In step 306, it is determined whether the required bit number B _Y is larger than the set bit number S _Y. In the case of the DC coefficient in FIG. 24, the required bit number B _Y “4” is smaller than the set bit number S _Y “6”. In other words, the number of bits required to store the quantized error data ΔYuv is smaller than the set number of bits.
For storing the other quantization error data ΔCbu
It can be used to store v, ΔCruv.

Then, next, the routine proceeds to step 321, where it is judged if the required bit number B _Cb is larger than the set bit number S _Cb . In the case of the DC coefficient in FIG. 24, the required number of bits B
_{Since Cb} = B _Cr = 5 and the number of set bits S _Cb = S _Cr = 3, the steps 321, 322, 323 are executed in order, and at step 323, the quantization error data ΔCbu
The bits for storing v and ΔCruv are expanded.

That is, in step 323, (18),
The maximum bit numbers B ″ _Cb and B ″ _Cr are expanded according to the equation (19). _{B "Cb = S Cb + (} S Y -B Y) / 2 (18) B" Cr = S all -B Y -B "Cb (19) for DC coefficients of Figure 24, the maximum number of bits B" _Cb = B " _Cr =
4 (see FIG. 28).

If the maximum number of bits B " _Cb expanded in step 323 is equal to or larger than the required number of bits B _Cb , that is, the number of bits B _Cb for storing the quantization error data ΔCbuv is expanded B" _Cb. If not so, go from step 324 to step 327. That is, the quantization error data ΔCruv is requantized, and the quantization error data ΔCbuv is not requantized. On the other hand, if the maximum number of bits B ″ _Cb is smaller than the required number of bits B _Cb , the process proceeds from step 324 to step 325, where the quantization error data ΔCbuv is requantized, and then the maximum number of bits is determined in step 326. It is determined whether B " _Cr is smaller than the required number of bits B _Cr . When the maximum number of bits B ″ _Cr is smaller than the required number of bits B _Cr , in step 327 the quantization error data ΔCr
If requantization is performed on uv and the maximum number of bits B ″ _Cr is greater than or equal to the required number of bits B _Cr , it is not necessary to requantize the quantization error data ΔCruv, and thus step 327.
Is skipped.

Then, when the step 312 and subsequent steps are executed and the bit string is generated for the quantized error data of all the pixel positions, this routine ends.

In the quantization table shown in FIG. 24, for example, when the parameter u is "4" and the parameter v is "0", the required bit numbers are B _Y = 5 and B _Cb = B, respectively, as described above. _Cr = 7 (see FIG. 28).
That is, the total number B _{all of} the required number of bits B _Y , B _Cb , and B _Cr
“19” is larger than the sum S of _all set bits S _all “11”, the required number of bits B _Y is equal to the set number of bits S _Y “5”, and the required number of bits B _Cb and B _Cr are the set number of bits S _Cb , S
_Cr Greater than "3". Therefore, in step 305,
Steps 306, 321, and 322 are performed in order, and step 32 is performed.
3 to 327 are executed. In step 323, the maximum number of bits B " _Cb = B" _Cr = 3 (see FIG. 28). In this case, unlike the case of the DC coefficient, the maximum number of bits B " _Cb = B" _Cr is the set number of bits. Same value as.

As shown in FIG. 28, the quantization coefficient Q (Y) uv
= 7 and Q (Cb) uv = Q (Cr) uv = 65, the required number of bits is B _Y = 3 and B _Cb = B _Cr = 7, respectively, and the total number B _all is the sum of the set number of bits. It is larger than S _all “11”. The required bit number B _Y is smaller than the set bit number S _Y “5”, and the required bit numbers B _Cb and B _Cr are the set bit number S
_Cb and S _Cr are larger than “3”. Therefore, step 3
05, 306, 321, and 322 are executed in this order, and steps 323 to 327 are executed. Note that in step 323, the maximum number of bits B ″ _Cb = B ″ _Cr = 4 (see FIG. 28).
In this case, as in the case of the DC coefficient, the maximum bit number B " _Cb = B" _Cr increases more than the set bit number.

As shown in FIG. 28, the quantization coefficient Q (Y) uv
= 129, Q (Cb) uv = Q (Cr) uv = 3, the required bit numbers are B _Y = 8 and B _Cb = B _Cr = 2, respectively, and the total number B _all is the sum of the set bit numbers. It is larger than S _all “11”. Required bit number B _Y is set bit number S _Y “5”
And the required bit numbers B _Cb and B _Cr are smaller than the set bit numbers S _Cb and S _Cr “3”. Therefore, in order to expand the bits for storing the quantized error data ΔYuv, after executing steps 305, 306, 308 and 328, the maximum number of bits B ″ _Y is obtained in step 330 according to the equation (20).

B " _Y = S _all -B _Cb -B _Cr (20) In this example, the maximum number of bits B" _Y = 7 (FIG. 2).
8).

At step 331, the quantization error data Δ
Requantization is performed on Yuv, and step 312
The following is done:

Steps 343 to 347 and step 353
~ 357, steps 323 to 3 respectively.
Similar to 27. Also steps 360 and 361,
The processes in steps 370 and 371 are similar to steps 330 and 331, respectively.

As described above, according to the fifth embodiment, when the required bit number of one or two of the quantized error data is smaller than the set bit number, it is used for storing this quantized error data. The missing bits are used to store other quantized error data. Therefore, in comparison with the fourth embodiment, the fourth JPEG format is further added.
Use efficiency of components is improved.

In each of the above embodiments, the set bit number S
_Y , S _Cb and S _Cr are recorded in the IC memory card M.

In each of the above embodiments, the DCT transform is used as the orthogonal transform of the original image data, but the present invention is not limited to this. Also, the recording medium for the image data 80 is not limited to the IC memory card M.

[0133]

As described above, according to the image compression apparatus of the present invention, it is possible to perform image compression capable of reproducing an image as close to the original image as possible. Further, according to the image expanding apparatus of the present invention, it is possible to reproduce a still image from the data obtained by the image compressing apparatus of the present invention and reproduce an image close to the original image.

[Brief description of drawings]

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

FIG. 2 is a flowchart showing an image compression process in the first embodiment.

FIG. 3 is image data P (Y) xy, DC in the first embodiment.
It is a figure which shows the example of T conversion coefficient S (Y) uv and quantized DCT coefficient R (Y) uv.

FIG. 4 is a diagram showing grouping of difference values of DC coefficients.

FIG. 5 is a diagram showing a code table of DC coefficients.

FIG. 6 is a flowchart showing a processing routine for encoding quantized AC coefficients.

FIG. 7 is a diagram showing zigzag scanning in Huffman coding of AC coefficients.

FIG. 8 is a diagram illustrating an example of encoded data by Huffman encoding.

FIG. 9 is a diagram showing the structure of image data in JPEG format.

FIG. 10 is a diagram showing the structure of image data in JPEG format.

FIG. 11 is a diagram showing an example of quantized DCT coefficients, restored DCT coefficients, and decoded image data.

FIG. 12 is a flowchart showing a first example of performing bit allocation and error data combination.

FIG. 13 is a flowchart showing a second example of performing bit allocation and error data combination.

FIG. 14 is a diagram showing a histogram of quantized error data ΔYuv.

FIG. 15 is a diagram showing a configuration of a fourth component ΔKuv.

FIG. 16 is a block diagram showing an image expansion device of the first embodiment.

FIG. 17 is a flowchart showing an image expansion process in the first embodiment.

FIG. 18 is a flowchart showing a process of obtaining error data in the image expansion process of the first embodiment.

FIG. 19 is a block diagram showing an image compression apparatus according to a second embodiment.

FIG. 20 is a block diagram showing an image expansion device of a second embodiment.

FIG. 21 is a block diagram showing an image compression apparatus according to a third embodiment.

FIG. 22 is a block diagram showing an image expansion device of a third embodiment.

FIG. 23 is a block diagram of an apparatus for reproducing an image without using error data of a fourth component ΔKuv.

FIG. 24 is a diagram showing an example of a quantization table of luminance signals and color difference signals.

FIG. 25 is a flowchart showing a processing routine for generating quantized error data in the fourth embodiment.

FIG. 26 is a flowchart showing a processing routine for generating quantized error data in the fourth embodiment.

FIG. 27 is a diagram showing requantization coefficients.

FIG. 28 is a diagram showing an example of bit lengths and bit allocations required to store quantized error data.

FIG. 29 is a flowchart showing a processing routine for generating quantized error data in the fifth embodiment.

FIG. 30 is a flowchart showing some steps of the processing routine shown in FIG. 29.

FIG. 31 is a flowchart showing some steps of the processing routine shown in FIG. 29.

[Explanation of symbols]

Mic memory card

Claims (11)

(57) [Claims] 1. An orthogonal transform unit for subjecting original image data composed of first to third components to orthogonal transform to obtain an orthogonal transform coefficient corresponding to a predetermined spatial frequency for each component, and the orthogonal transform coefficient being a quantum. Quantization by the conversion table,
Quantization means for obtaining a quantized orthogonal transformation coefficient, dequantization means for dequantizing the quantized orthogonal transformation coefficient to obtain a restored orthogonal transformation coefficient, and a difference between the restored orthogonal transformation coefficient and the orthogonal transformation coefficient. The first error data generating means for obtaining the quantized error data for each of the components and the quantized error data for the respective components are the same.
Collected in one unit data of fixed length for each spatial frequency.
Error data synthesizing means for synthesizing components, encoding means for encoding the fourth component and the quantized orthogonal transform coefficient to form compressed image data, and recording means for recording the compressed image data on a recording medium. An image compression device comprising. 2. The number of bits of the quantized error data stored in the fourth component corresponds to the component that most affects the visual characteristics of the reproduced image in the first to third components, which is longer. as, according to comprising the bit number allocation means for allocating
The image compression device according to Item 1 . 3. An error data quantization means for requantizing quantization error data that does not fit in the number of bits assigned by the bit assigning means.
The image compression device according to claim 2 . Wherein said error data quantizing means, 請, characterized in that using the maximum value of the quantization coefficients of the quantization table, and re-quantizing all of the components of the quantization error data
The image compression device according to claim 3 . Wherein said error data quantization means, wherein by using the respective quantization coefficients of the quantization table, characterized by re-quantizing the components of the quantized error data corresponding to the quantized coefficients Item 4. The image compression device according to Item 3 . 6. The error data quantization means, when any number of bits of the requantized quantized error data of the first to third components is smaller than the assigned number of bits, the error data quantization means of the other component The image compression apparatus according to claim 5 , wherein the number of bits of the quantization error data is increased by an amount corresponding to the fixed length. 7. When the orthogonal transform coefficient output from the orthogonal transform means is negative, a code adjusting means for multiplying the quantization error data corresponding to the orthogonal transform coefficient by “−1” is provided. The image compression apparatus according to claim 2 . 8. Scaling means for deriving the coding error data by assigning shorter code words to the quantized error data code-adjusted by the code adjusting means as the frequency of occurrence is higher. Claim
7. The image compression device according to 7 . 9. The error data synthesizing means for generating the fourth component from the encoded error data, wherein the number of bits of the encoded error data is k, m, n (k ≧ m ≧).
n), the most significant bit to the (m + 1) th bit of the quantized error data to which the number of bits k has been allocated, and the most significant bit to the (2nd) bit of the fourth component
Stored in the first area up to (m + n + 1) bits, the m-th bit to (n + 1) th bit of the quantization error data to which the bit number k is assigned, and the most significant bit of the quantization error data to which the bit number m is assigned. To (n + 1) th bit to the (2m + n) th bit of the fourth component.
Bits to the (3n + 1) th bit are alternately stored in the second area, and the quantization error data to which the bit number k is assigned, from the nth bit to the least significant bit, and the quantization error to which the bit number m is assigned. The nth bit to the least significant bit of the data and all the bits of the quantized error data assigned the number of bits n are cyclically placed in the third region from the 3nth bit to the least significant bit of the fourth component. The image compression apparatus according to claim 5 , wherein the image compression apparatus stores the image compression apparatus. 10. An orthogonal transform means for subjecting original image data composed of first to third components to orthogonal transform to obtain an orthogonal transform coefficient for each component, and the orthogonal transform coefficient is quantized by a quantization table and quantized. Quantization means for obtaining an orthogonal transformation coefficient, dequantization / inverse orthogonal transformation means for dequantizing the quantized orthogonal transformation coefficient and inverse orthogonal transformation to obtain a restoration orthogonal transformation coefficient, and the restoration inverse orthogonal transformation coefficient Second error data generating means for obtaining error data which is a difference between the original image data and each of the components, and the error data in one unit data having a fixed length for each same image element position of each component. Error data synthesizing means for synthesizing the fourth component together, and compressed image data obtained by encoding the fourth component and the quantized orthogonal transform coefficient. An image compression apparatus, comprising: an encoding unit that forms a recording medium; and a recording unit that records compressed image data on a recording medium. 11. The compressed image data recorded by the compressed image device according to claim 1 is read from the recording medium, and the quantized orthogonal transform coefficient and the fourth component are read from the compressed image data. And a dequantization unit that dequantizes the quantized orthogonal transform coefficient to restore the orthogonal transform coefficient, and error data of each component that constitutes the compressed image data from the fourth component. Error data separating means for separating the quantized orthogonal transform coefficient, error data adding means for adding the error data to the quantized orthogonal transform coefficient, and inverse orthogonal transform for the orthogonal transform coefficient output from the adding means to restore the image data. An image decompression device comprising an inverse orthogonal transformation means.