JPS61147692A - Data compression/expansion method - Google Patents

Abstract

PURPOSE:To enable to select the picture of high quality according to the demands of users, by discarding the fraction and making quantization of the conversion co-efficient obtained by vertical cross conversion against the division block using the threshold value of discarding and coefficient quantization step width which are set in advance, forming the encoded data, transmitting or storing the encoded data, and decoding it at the expansion side. CONSTITUTION:8-bit picture data of one picture element is divided horizontally into 16 picture elements and vertically 16 picture elements, and multiplication of the 2-dimension discrete cosine conversion matrix with its transposed matrix is executed at converter 1-16, the conversion coefficient matrix is obtained. The output of coefficient discarding device 1-17 is uniformly quantized at step width 1-12 by coefficient quantizing device 1-18, and allotted with variable code specified in advance by encoder 1-19. The encoded data 1-21 generated by the encoder is transferred or stored as compressed data 1-13. The expanding device takes reverse procedure; the reproduced conversion coefficient reversely converted by 2-dimension discrete reverse conversion device 2-17, and the decode picture signal 2-18 is outputted.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a data compression/expansion method, and particularly to an image data compression/expansion method that allows selection of image quality.

(Prior art and its problems) There are many types of images used in the medical field, such as X-rays and 07 images, but these have conventionally been stored in the form of film. However, in hospitals, etc., the number of sheets generated per day is large (and it is necessary to store them for several years, resulting in a huge amount, making storage space, manpower, and time required for searching) a big problem. There is.

In recent years, with advances in digital image processing technology and device technology, it has become possible to digitize, store, and search medical images. Medical X-ray images require high accuracy in both resolution and number of gradations, at least 1024 x 1024 x 8 bits, and no matter how advanced storage elements are, the large amount of data is a problem. For this reason, data compression techniques are often used to reduce the amount of stored data. Data compression technology can be roughly divided into predictive coding and transform coding, but the compression rate is 5.
Transform coding is said to be advantageous when high compression is required.

(Object of the Invention) An object of the present invention is to provide a data compression/expansion method that allows image quality to be selected according to user requirements.

In medical images, there are strict requirements for image quality deterioration, and if the signal-to-noise ratio (hereinafter referred to as S/N ratio) becomes too low, findings and the like become blurred, causing practical problems. In conventional image data compression/expansion methods, the compression method was fixed, so compression parameters had to be set to obtain an S/N ratio that virtually caused no problems for all images. Sufficient compression ratio could not be obtained. However, some medical images require a high S/N ratio depending on their content, while others can be satisfied with a certain S/N ratio. For example, if we take an X-ray image, images that require detailed examination because the affected area has not yet been determined require a very high S/N ratio in order to fully reproduce the details. Images where the affected area is already clearly visible and you want to save for future reference, or X-ray images of a healthy person that are unlikely to be referenced in the future, but must be saved for several years. A very high S/N ratio is not required, but a high compression ratio is required in order to reduce the amount of data stored.

(Structure of the Invention) The image quality variable data compression/expansion method of the present invention allows the user to set the image quality on the compression side, and sets the compression parameters (coefficient truncation threshold, coefficient quantization step width, etc.) corresponding to the image quality. ), divide the original image into blocks and read them from the storage device, perform orthogonal transformation on these blocks to obtain transformation coefficients, and apply these transformation coefficients to the previously set coefficient truncation threshold and coefficient quantization step. Perform truncation and quantization using the width, assign a preset code to the truncation and quantization output to create encoded data, and apply the encoded data to the previously set coefficient truncation threshold and coefficient quantization step. On the decompression side, the transmitted or stored encoded data is decoded, and the transform coefficients are reproduced from the output of this decoding means using the transmitted or stored coefficient quantization step width and coefficient truncation threshold. The method is characterized in that the reproduced transform coefficients are subjected to inverse orthogonal transform to restore the image.

(Example) The present invention will be described in detail below with reference to the drawings. Here, we will focus on X-ray images as the main target and two-dimensional discrete cosine transform as the orthogonal transformation, but it is also possible to target both CT images and use orthogonal transformations such as Hadamard transform. It's obvious.

FIG. 1(a) is a block diagram showing an example of an image data compression device that implements the data compression/decompression method of the present invention, and FIG. 1(b) is a block diagram showing an example of the decompression device.

In FIG. 1(a), the image data stored in the image memory 1-14 is block data readout 1s1-1.
5, the data is read out in block units for two-dimensional discrete cosine transformation. In this embodiment, image data of 8 bits per pixel is divided into blocks of 16 pixels vertically and 16 pixels horizontally and read out. Next, the read block data is processed by two-dimensional discrete cosine transformer 1-
At 16, a two-dimensional discrete cosine transform matrix is multiplied by its transposed matrix to obtain a transform coefficient matrix. Since the above calculations are performed with a finite word length, the calculation results have already been quantized by rounding. In the image quality selector 1-22, a coefficient truncation threshold is set to a value between 1/8160 and 8/8160 of the dynamic range of the conversion coefficient, according to the image quality switching signal selected by the user corresponding to the required image quality. (hereinafter referred to as threshold value) and 1/8160 to 4/816 of the dynamic range of the conversion coefficient.
The combination with the coefficient quantization step width (hereinafter referred to as step width) set to 0 is the threshold 1-11 and step width 1.
Output as -12. The transform coefficient matrix obtained by the two-dimensional discrete cosine transformer 1.16 is compared with a threshold value 1.11 in a coefficient truncation unit 1-17. According to the comparison result, conversion coefficients whose absolute value is smaller than the threshold value are rounded to zero, and the threshold value is subtracted from the absolute value of other conversion coefficients. However, the above processing is not performed on the DC component, and the two-dimensional discrete cosine transformer 1-16
The output of is output as is. Next, the output of the coefficient truncation unit 1-17 is uniformly quantized in a coefficient quantizer 1-18 with a step width of 1-12. The uniformly quantized transform coefficients are transformed in an encoder 1-19. A preset variable length code is assigned according to the coefficient occurrence frequency distribution.The code data 1°21 created by the encoder is
A threshold value 1-11 and a step width 1-12 are added by a parameter adder 1-13, and the data is transmitted or stored as compressed data 1.13.

In the decompression device shown in FIG. 1(b), the transmitted or stored compressed data 2-13 is passed through a parameter separator 2-19.
It is separated into encoded data 2-20 and parameters (threshold value 2-12, step width 2-11). Encoded data 2-
13 is decoded by a decoder 2-14 and a transform coefficient is output. The coefficient inverse quantizer 2-15 has a step width of 2-1.
Multiply the transform coefficient obtained by the decoder by 1. However, this processing is not performed on the DC component and the value is output as is.

Next, cut off the coefficients! In 2-16, threshold 2-1
2 is added to the absolute value of the output of the coefficient inverse quantizer to reproduce the transform coefficients. Here again, the DC component and the conversion coefficient whose value is zero are output as they are without performing the above processing.

The transform coefficients reproduced through the above procedure are inversely transformed by a two-dimensional discrete inverse transformer 2-17, and a decoded image signal 2.18 is output.

FIG. 2 shows a block diagram of the block data reader. A clock is applied to the terminal 21 from a clock generator (not shown) to drive the row address counter 26 and column address counter 27. The output of each address counter is output to terminals 22 and 23, and the image memory 1
The image data of the corresponding address is read from 4, and the image data is written to the block data memory 28 from the terminal 24. The written image data is outputted to the two-dimensional discrete cosine transformer via the terminal 25.

FIG. 3 shows a block diagram of a two-dimensional discrete cosine transformer. When an image signal divided into small regions is represented by a matrix P, a two-dimensional discrete cosine transformation coefficient F is given by F:APAT, where A is a transformation matrix and AT is its transposed matrix. θ of P from the memory 28 for one block.

k) Input the components from the read terminal 32 and convert (i, j) of the transformation matrix A stored in the read-only memory 34
An address from an address generator whose components are not shown is input from a terminal 31 and read out, and the product of both is sent to a multiplier 35.
Find it with A. This result is sent to register 37A by adder 36A.
The outputs of are added and stored in the register 37A again. By repeating the above operation from 1 to n with i and k fixed, the (i, k) component of the matrix, which is the product of A and P, is obtained, and its value is stored in the block data memory 38 (i,
k). However, n is the size of the block, and when j starts from 1, the contents of the register 37A are cleared. By repeating the above operations for i and from 1 to n, a matrix of the product of A and P is stored in the block data memory 38. Next, the product of AP and A-1 is obtained in exactly the same manner as above. That is, the (i, j) components of A-P stored in the block data memory 38 and the read-only memory 34
The (k, j) component of the conversion matrix A is input from the terminal 31 and read out from the address generator (not shown) to calculate the product for j from 1 to n, and the (i, j) component of APAT is k) Calculate the components, repeat the process, and output the transformation coefficient matrix to the terminal 33. Here A and A
T is created by changing the reading of exactly the same read-only memo +7-34. Furthermore, since the calculation is performed with a finite word length, the output has already been quantized.

FIG. 4 shows a block diagram of the coefficient truncator. terminal 41
The conversion coefficients input from the absolute value circuit 44 together with the threshold values supplied from the image quality selector 4-22 via the terminal 42
Absolute values are taken at A and 44B, and their magnitudes are compared by a comparator 45. At the same time, a subtracter 46 subtracts the threshold value from the absolute value of the conversion coefficient, and the sign of the conversion coefficient is determined by a polarity determiner 48.
The multiplier 47 multiplies the two to reproduce the sign of the transform coefficient. Then, if the absolute value of the conversion coefficient is smaller than the threshold according to the result of the comparator 49, the AND gate 49 is closed;
Outputs zero to the terminal 43, otherwise the previous multiplier 47
The output is outputted to the terminal 43. Also, when the conversion coefficient corresponds to a DC component, the absolute value output of the threshold value is set to zero, and the output of the two-dimensional discrete cosine converter is sent directly to the terminal 4.
3 is output. The above operation can be expressed as follows.

However, X is a transformation coefficient, t is a threshold value, sgn(·) is a function for extracting the sign, and i and j are row and column numbers of the transformation coefficient matrix.

FIG. 6 shows a block diagram of the quantizer. The truncated transform coefficient input from the terminal 61 is truncated and quantized by the divider 64 according to the quantization step width input from the image quality selector 1-22 via the terminal 62. 63. However, here too, in the case of a DC component, the quantization step width is set to 1, and the result is output without substantially any processing. Quantization step width is 1.2,
4. ...9. If it is in the form of a power of 2, the divider 64 only needs to shift.

FIG. 5 shows a diagram in which the relationship between the two is plotted for various threshold values using a typical X-ray image as an example, with entropy on the horizontal axis and S/N on the vertical axis. (Fixed step width) As a result of considering various values for the step width, the threshold value was 1/8160 to 8/816 of the dynamic range of the conversion coefficient.
In the range of 0, the S/N and entropy maintain a linear relationship, but when the value exceeds this value, the entropy does not decrease even though the S/N decreases, and medical X-ray images with a value of about 40 dB or more It has been confirmed that if the S/N is not maintained, delicate affected areas will become blurred, which will pose a practical problem. Therefore, it is appropriate to set the threshold value for coefficient truncation to 1/8160 to 8/8160 of the dynamic range of the conversion coefficient.

FIG. 7 shows the relationship between entropy and S/H as a step width parameter. Here, the threshold is fixed, but
As mentioned earlier, while ensuring the S/N at 40 dB or more, we also considered various threshold values to maintain a linear relationship between S/N and entropy, and the step width was converted. 1/8160 to 4 of the dynamic range of the coefficient
It has been confirmed that selecting /8160 is appropriate.

FIG. 12 shows an example of average values of the S/N ratio and compression rate for a plurality of X-ray images for three combinations of threshold value and step width. From this table, as an example of an image quality selector, the combination of (threshold value, step width) is set to (2/8160.1/8160) (4/8
160°2/8160) (8/8160.4/816
It is also possible to switch to three stages (0). In this case 4
If image quality of about 8 dB is required, threshold value 1-11
As a combination of step width 1-12 (2/81
60. 1/8160) and an image quality of about 44 dB is required, (4/8160. 12/81
60), and if an image quality of about 41 dB is acceptable, output (8/8160. 4/8160). By switching the threshold value and step width in the image quality selector in accordance with requests in this way, it is possible to increase the compression rate for images that do not require particularly high image quality.

Next, the encoder 1-19 performs entropy encoding on the transform coefficients that have been truncated and quantized as described above in accordance with their frequency of occurrence. At that time, we not only simply allocate a variable-length Huffman code to each coefficient of each component, but also take advantage of the fact that high frequency components within a block are often zero. The encoding order is zigzag so that zeros continue in the latter half, and the zero part is run-length encoded, or if there are all zeros from the middle to the end within one block, the encoding is stopped there. , it is also possible to improve the encoding efficiency by sending an end mark of the block. A threshold value and a step width are added to the encoded data created by the encoder 1-19 by a parameter adder to create compressed data 1-13.

An example of the compressed data structure is shown in FIGS. 9(a), (b), and (c). One entire image consists of a header and codes for each block, as shown in FIG. 9(a). The header has the structure shown in FIG. 9(b), and stores the threshold value and step width selected by the image quality selector. Each block has the configuration shown in FIG. 9(c), in which non-zero coefficients are assigned Huffman codes, and when zeros continue, their run lengths are stored. If zeros continue from the middle to the end of the block, a block end mark is added immediately after the coefficient sequence without sending a run length of zero.

The compressed data obtained in this manner is sent to a transmission path and stored in a file device.

Next, the transmitted or accumulated compressed data is processed by a parameter separator 2-19 into encoded data 2-20 and threshold value 2-1.
2. After being separated into step widths of 2-11, code data 2
-20 is the encoder 1-19 in the decoder 2-14.
Decoding is performed for the entropy encoding used in . This output is a transform coefficient that has been quantized and truncated. This transform coefficient is subjected to coefficient dequantization in a coefficient dequantizer 2-15 using a step width 2-11 separated from the compressed data by a parameter separator 2-19. FIG. 10 shows a block diagram of the coefficient dequantizer. A multiplier 104 multiplies the transform coefficient input from a terminal 101 and a quantization step width input via a terminal 102 and outputs the result to a terminal 103 . Here, if the conversion coefficient is a DC component, the step width is forced to 1.
The result is output without any processing. The output of the coefficient inverse quantizer is 52-16
In this step, coefficients are rounded down using the threshold values separated by the parameter separator.

FIG. 11 shows a block diagram of the coefficient rounding and discarding unit. The absolute value of the conversion coefficient inputted from the terminal 111 is taken by the absolute value circuit 114A, and the absolute value of the conversion coefficient inputted from the terminal 112 is taken by the absolute value circuit 114B and the adder 116.
The polarity determination circuit 118 extracts the sign of the conversion coefficient, and the multiplier 117 reproduces the sign. Also, the zero determination circuit determines whether the conversion coefficient takes zero, and if it is zero, the AN
Close the D gate 119 and output zero to the terminal 113, and if it is not zero, the output of the multiplier 117 is directly sent to the terminal 113.
Output to. In addition, when the conversion coefficient input from the terminal 111 corresponds to a DC component, the threshold value is set to zero and the terminal 111
The conversion coefficients input from the terminal 113 are output as they are to the terminal 113. 2-dimensional discrete cosine inverse transformer 2-17
can be realized with the same configuration as the two-dimensional discrete cosine transformer shown in FIG. The transformation coefficient is F, the transformation matrix is A
, If the inverse matrix of A is A-1, then A is regular, so (
AT)-1 = (A-1)T, so the decoded image P obtained by inverse transformation is P: A-1・F ・(A'r)-1= A-1・F
, (A-1) given by T. Therefore, if the contents of the read-only memory 34 are changed to A-1 instead of A,
Two-dimensional discrete cosine inverse transformation can be performed using the configuration shown in FIG. 3 as is.

In the above example, the threshold and step width values selected by the image quality selector are directly entered as the header of the compressed data, but some combinations are numbered and the numbers corresponding to the selected combinations are used as the header. You can also attach it. In this case, of course, a conversion circuit is required on the expansion device side to convert the selection number into the actually used threshold value and step width. Further, in the above embodiment, a variable length code preset according to the frequency of occurrence is used as the code, but it is of course possible to use an equal length code.

(Effects of the Invention) As described above, the data compression/expansion method of the present invention selects and uses threshold values and step widths corresponding to the required image quality, thereby achieving compression corresponding to the required image quality for each image. Data can be compressed and expanded at a high rate.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a block diagram of a block data reader, FIG. 3 is a two-dimensional discrete cosine transformer, and FIG. 4 is a block diagram of a coefficient truncation device. Figure 5 is a diagram showing the relationship between entropy and S/H using the coefficient truncation threshold as a parameter.
Fig. 6 is a block diagram of the coefficient quantizer, Fig. 7 is a diagram showing the relationship between entropy and S/N using the coefficient quantization step width as a parameter, and Fig. 8 is a diagram showing the coding order within a block. Figure 9 shows the configuration of the code, Figure 10 is a block diagram of the coefficient inverse quantizer, Figure 11 is a block diagram of the coefficient rounding and discarding unit, and Figure 12 shows the combination of threshold value, step width, and S/ N
It is a figure showing the relationship between a ratio and a compression rate. In the figure, 1-14... Image memory 1-15... Block reader 1-16...
・2-dimensional discrete cosine converter 1-17・
... Coefficient truncation device 1-18 ... Coefficient quantizer 1-19 ... Encoder 1-22 ... Image quality selector 1-23 ... Parameter adder 2-14...Decoder 2-15...Coefficient inverse quantizer 2-16...Coefficient rounding and discarding unit 2-17...Two-dimensional Discrete cosine inverse transform 22-19...Parameter separator 26...Row address counter 27...Column address counter 28...Block data memory 34...
Read-only memory 35A, 35B...Multiplication 5 36A, 36B...
... Adder 37A, 37B ... Register 38 ... Block data memory 44A, 44B
...Absolute value circuit 45, song comparator 46...
・Adder 47... Multiplier 48°°
°°° Polarity determiner 49...AND gate 64...Divider 104/Music multiplier 114A, 114B...Absolute value circuit 115...
...Zero judgment circuit 116...Adder 117
. . . Multiplier 118 . . . Polarity determination circuit 119 . . . AND gate. Figure 3 Figure 4 Figure 5 Entropy (bit/pet) Figure 6 Bait Figure 7 Figure 8 Figure 9 (b) Figure 10 Figure 11

Claims (1)

[Claims] At the time of compression, the user sets the image quality as necessary, sets the coefficient truncation threshold and coefficient quantization step width corresponding to the image quality, reads the original image from the storage device in blocks, and processes the blocks. The transform coefficients obtained by performing orthogonal transformation on the unit are truncated and quantized using the coefficient truncation threshold and the coefficient quantization step width, and a preset sign is assigned to the truncated and quantized transform coefficients. create encoded data, transmit or store the encoded data, the coefficient truncation threshold, and the coefficient quantization step width; decode the transmitted or accumulated encoded data during decompression; A data compression/expansion method characterized in that a transform coefficient is reproduced using the coefficient truncation threshold and the coefficient quantization step width, and an inverse transform of the orthogonal transform is performed on the reproduced transform coefficient to restore an image.