JPH09116899A - Method for compressing image data - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the compression efficiency of image data. SOLUTION: A band split section 1 divides image data into plural spatial frequency band components LL, LH, HL, HH, and coding sections 3-5 applies modified Huffman coding to O-run length of the LH, HL, HH components in which at least one of horizontal/vertical directions is a high frequency component. Furthermore, the scanning direction of the coding is selected in horizontal, vertical and tilt direction and the coded data are more reduced.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image data compression method, and more particularly to an image data compression method in which image data is divided into spatial frequency bands and information in each band is encoded and compressed.

[0002]

2. Description of the Related Art In image data compression technology using sub-band coding or wavelet transform coding, image data is divided into high band and low band by its spatial frequency, and each band component is coded for compression. To do. In this case, the high-frequency component has the property that its value tends to concentrate in the vicinity of 0, and the property that the value 0 tends to be continuous except for the edge and the like, that is, the property of low entropy, and such property is used. The compressed data can be efficiently compressed.

[0003]

In the above-mentioned conventional technique, each component divided by the spatial frequency is divided into blocks of vertical 8 × horizontal 8 data as in the JPEG system,
Quantization and variable length coding are performed on each block. However, regarding the high-frequency component, even if the number of 0s that is much longer than the block size 8 continues, it is purposely subdivided into blocks and separately coded. There was a problem in that encoding could not be performed making full use of the property of low entropy.

It is an object of the present invention to provide a method of compressing image data which enables more efficient compression by performing encoding suitable for the characteristics of each frequency band component.

[0005]

According to the present invention, image data is divided into a plurality of spatial frequency bands, each of which is quantized to generate a component, and the component is scanned for at least one of the components. However, a method for compressing image data is disclosed in which a length of 0 runs in which the value 0 is continuous is encoded by the modified Huffman encoding method.

Further, according to the present invention, the division into the spatial frequency components is performed by first and second low frequency components in both the horizontal and vertical directions of the screen, one low frequency component and the other high frequency component. 3. The image data compression method according to claim 1, wherein the three components, both of which are high-frequency components, are divided into four components, and the fourth component is a Haar transform method. Disclose.

[0007]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below with reference to the drawings. FIG. 1 is a diagram showing an embodiment of a compression method of the present invention, in which a band division unit 1, an LL component encoding unit 2, an HL component encoding unit 3, an HH component encoding unit 4, and a combining unit 5 are provided. It consists of Here, each unit may be configured by a program that performs each process, or may be configured by using dedicated hardware. The input image data is monochrome multi-tone data (including binary) or image data for one plane of a color image.

FIG. 2 shows an example of band division of image data in the band division unit 1. A band division filter using a well-known Haar basis is used as a filter used for band division of spatial frequency, whereby LL, It is supposed to be divided into four bands of LH, HL, and HH components. Here, L represents a low frequency component, H represents a high frequency component, and the left side of the two characters represents a horizontal component of the screen and the right side thereof represents a vertical component. For example, the LH component is a low frequency component in the horizontal direction and a high frequency component in the vertical direction. FIG. 2 shows the four components in the spatial frequency plane.

FIGS. 3 to 5 show examples of divided data by the band division unit 1 described above. The image data is 4 × 4 pixels for simplicity. FIG. 3 shows a case of a screen having horizontal edges as shown in the image data D1 of FIG. 3A. When this is band-divided by spatial frequency, each component corresponding to FIG. ). Since the edge in the horizontal direction means a change component (high frequency component) in the vertical direction, L
In addition to the L component, only the LH component shows a non-zero value, and since there is no change in other directions, all other components are 0.

The image data D2 in FIG. 4A has an edge only in the vertical direction this time, and the band division result shows the change in the LL component and the horizontal direction as shown in FIG. A value other than 0 appears in the HL component shown, and all other components are 0.

Further, the image data D3 in FIG. 5A is an example of a screen having an edge in an oblique 45 ° direction, and the result of the band division is, as shown in FIG. 5B, horizontal and vertical directions. The HH component indicating the high frequency component has a value together with the LL component,
All other components are 0. As described above, in the band division using the Haar basis, the image data of 4 × 4 pixels is divided into four components, and each component is 2 × 2 data of half the size. Is also the same. That is, when the image data of n × n pixels is divided into bands as described above, each component becomes (n / 2) × (n / 2) data.

Next, the processing in the encoding units 2 to 5 for each component, which is a feature of the present invention, will be described. First, each component divided by the band division unit 1 is quantized by being divided by a quantization coefficient determined for each component. This post-quantization processing method uses the LL component of the low frequency component in both the horizontal and vertical directions and the LH component of which at least one component is the high frequency component, H
The L component and the HH component differ.

Since the LL component is generally less than a statistical piece such as a long run length of 0, it is encoded by a well-known general encoding method such as DPCM or JPEG. In the case of the quantized LH component, HL component, and HH component, variable length coding is performed with a run length of the number of consecutive 0s for a run of 0s having consecutive 0s. For non-zero runs in which values other than 0 are consecutive, variable-length coding is performed by entropy coding such as Huffman coding and arithmetic coding.

Here, in order to perform variable length coding of the run length of 0, for example, the MH code (Modi
It is effective to use fied Huffman Code). This M
H code is run length i

## EQU1 ## The form of i = 64M + T, that is, expressed in 64-base number, the codes for the respective values of 64M and T (the former is Make up Codes, the latter is Term
Inating Codes) is defined as shown in FIGS. 6 and 7. The expression (Equation 1) means that the number of pixels in one line of the screen is 1728 pixels in the G3 type facsimile, so it is inefficient to assign different codes to each of the run lengths of this number. This is because the scale of the device becomes large. Also, the code of FIGS.
The entropy is determined to be small from the appearance probabilities for the respective values of M and T.

As shown in FIGS. 6 and 7, there are a white run length code word and a black run length code word in the case of MH coding. Therefore, encoding is performed using each of these, and the code word having the smaller total code amount is used. Also, as shown in FIGS. 6 and 7, the maximum value of 64M and T in (Equation 1) is 2560.
And 63, the maximum run length that can be represented by this MH code is 2623 (= 2560 + 63). Therefore 0
When the run length of the above exceeds this, it is divided so that it becomes 2623 or less, and a code EOL indicating that it is a break is inserted into the break. As described above, when the run length of 0 of the LH component, the HL component, and the HH component is variable-length coded, the scanning direction of pixels is taken into consideration in each component, so that a more efficient code is obtained as shown below. Becomes possible.

First, the LH component represents a component obtained by extracting edges in the horizontal direction as described with reference to FIG. 3. Therefore, in variable length coding of this component, when pixels are scanned in the horizontal direction as shown in FIG. A longer run length of 0 can be detected. However, FIG. 8 shows the scanning direction as 10 × 10 pixels, and this is the same in the following. Further, when there is a linear edge in the horizontal direction as in the image data D1 of FIG. 3A, it is highly possible that the same value other than 0 still exists continuously in the horizontal direction. The other entropy coding is also efficient.

As described with reference to FIG. 4, the HL component has only the property that the vertical and horizontal directions are interchanged with the LH component of FIG. Therefore, by scanning pixels in the vertical direction as shown in FIG. 9, a longer run length of 0 can be detected, and entropy coding of a portion other than 0 can be efficiently performed.

As described with reference to FIG. 5, the HH component represents a component obtained by extracting an edge in the diagonal direction. Therefore, when variable length coding this component, pixels are diagonally arranged as shown in FIG. 10 or 11. A longer run length of 0 can be detected by scanning. However, in the case of FIG. 10, the inclination is 45 ° to the right, but it may be 45 ° to the right as shown in FIG. Considering the case of an angle other than 45 °, it is efficient to change the scan direction to each of the directions shown in FIGS. 8 to 11 and use the scan direction with the smallest code length.

In the above, the method of efficiently coding the part in which 0s are continuous by the MH code has been described, but the following measures are effective for the coding of the part of the value other than 0. One of them is a method in which, as shown in FIG. 12, when a value other than 0 is continuous along the scanning direction, the difference between the adjacent pixel values is calculated and then encoded. That is, FIG.
In (a), when the scanning direction is the same horizontal direction as in FIG. 8, a portion surrounded by a thick line is a portion where values other than 0 continuously appear. In general, adjacent values of such a part do not change abruptly and the correlation is strong. Therefore, the difference value between the adjacent values along the scanning direction in this bold line frame is shown in FIG.
And entropy-encoding this, the compression rate is improved. However, when the difference is obtained, the difference with 0 at the head position is obtained.

Further, like the pixels surrounded by the thick line in FIG. 13A, pixels in which the pixels on the upper, lower, left and right sides are all 0 are often like noise, which is shown in FIG. It can be replaced with 0 as a singular point as in b). In this case, the portion of the singular point is included in the run of 0, so that the coding efficiency is improved.

The processing of the four coding sections 2 to 5 in FIG. 1 has been described in detail above, but the coded data generated by these processings are combined in the combining section 6. This synthesis is, for example, LL
The coded data of each of the component, the LH component, the HL component, and the HH component is collected as one time series data. Here, for the LL component, as described above, DP
It is encoded data generated using a method such as CM or JPEG. On the other hand, in the case of LH, HL and HH components,
As described above, the run of 0 and the run of non-zero are alternately encoded. Therefore, for example, looking only at the LH component, the first 0
, Encoded data of the first non-zero run, encoded data of the second zero run, encoded data of the second non-zero run, and so on. FIG. 14 and FIG.
FIG. 5 shows an example thereof. In FIG. 14, data CO (1) obtained by MH encoding the run length of the first 0 run, data CN (1) obtained by encoding the first non-zero run and its Code EOD for indicating the end, data CO (2) obtained by MH encoding the run length of the second 0 run, data CN (2) obtained by encoding the second non-zero run, and a code indicating the end thereof. EOD, ...
It is configured as. Also, in the case of FIG. 15, instead of the code EOD indicating the end of the non-zero run, codes LN (1) and LN (2) indicating the length of the data obtained by encoding the non-zero run are shown.
Is used. However, this code LN (1), LN
It is assumed that (2) ... Are coded by the MH code similarly to the run length of 0 run.

As described above, the image data in which the high frequency components are efficiently compressed by utilizing the statistical property thereof is generated. The only LL component is the conventional DPCM or JP.
Although the EG method is used, the data of this LL component is 1
The image data can be regarded as one image data, and the band division similar to that in FIG. By recursively repeating such subdivision of band division, more efficient and accurate compression becomes possible.

[0023]

As described above, according to the present invention, there is an effect that a run of 0 can be efficiently coded as compared with the conventional coding method of a small coding unit. This has the effect of enabling accurate encoding with suppressed distortion.

[Brief description of the drawings]

FIG. 1 is a diagram showing a flow of a compression method of the present invention.

FIG. 2 is an explanatory diagram of spatial frequency band division of image data.

FIG. 3 is an explanatory diagram of Haar transformation.

FIG. 4 is an explanatory diagram of Haar transform.

FIG. 5 is an explanatory diagram of Haar transformation.

FIG. 6 is a diagram showing Make up Codes of an MH code.

FIG. 7 is a diagram showing Terminating Codes of MH code.

FIG. 8 is a diagram showing an example of an LH band scan order.

FIG. 9 is a diagram showing an example of an HL band scan order.

FIG. 10 is a diagram showing an example of an HH band scan order.

FIG. 11 is a diagram showing another example of the scanning order of the HH band.

FIG. 12 shows an example of differential encoding of values other than 0.

FIG. 13 shows an example of deleting a singular point having a value other than 0.

FIG. 14 is an explanatory diagram of encoded data.

FIG. 15 is an explanatory diagram of encoded data.

[Explanation of symbols]

 1 band division unit 2 LL component coding unit 3 LH component coding unit 4 HL component coding unit 5 HH component coding unit 6 combining unit

Claims (15)

[Claims] 1. The image data is divided into a plurality of spatial frequency bands, each of which is quantized to generate a component, the constituent element is scanned for at least one of the components, and the value 0 is consecutive 0. A method for compressing image data, characterized in that the run length is encoded by a modified Huffman encoding method. 2. The division into the spatial frequency components includes the first component of low frequency components in both the horizontal and vertical directions of the screen,
One is a low frequency component, the other is a high frequency component of the second and third components, both are high frequency components of a fourth component into four components, and the method of division is the Haar transform method. The image data compression method according to claim 1. 3. The first component is JPEG method or DPC
3. The method of compressing image data according to claim 2, wherein the length of the 0 run is encoded by the M method, and the other three components are encoded by the modified Huffman encoding. 4. The method of compressing image data according to claim 1, wherein the component is replaced with a difference between its adjacent components before scanning the component. . 5. The method according to claim 1, wherein before the scanning of the component, the component value of the component adjacent to each other in both the horizontal and vertical directions is set to 0 forcibly. Of 1
The method of compressing image data described in 1. 6. The code word of which the length of the 0 run is encoded by each of the white run length code word and the black run length code word of the modified Huffman code, and the generated encoded data becomes shorter. 4. The image data compression method according to claim 1, wherein the coded data according to claim 1 is coded data of the 0 run. 7. The method of compressing image data according to claim 1, wherein the scanning direction of the constituent elements is the horizontal direction of the screen. 8. The method of compressing image data according to claim 1, wherein the scanning direction of the constituent elements is a vertical direction of the screen. 9. The method of compressing image data according to claim 1, wherein the scanning direction of the constituent elements is set to the direction of upward 45 ° or downward 45 ° of the screen. 10. Coded data is generated for each of the scanning directions of the constituent elements in the horizontal, vertical, upward 45 ° rightward, and downward 45 ° rightward directions of the screen, and the shortest one of the generated encoded data is generated. The encoded data of the said component is used, The compression method of the image data of Claim 1 characterized by the above-mentioned. 11. A non-zero run in which non-zero values are consecutive in a component for coding the length of the zero run by the modified Huffman coding method is coded by a Huffman coding method or an arithmetic coding method. The method for compressing image data according to claim 1, wherein the image data is compressed. 12. The method for compressing image data according to claim 11, wherein the end of the coded data of the non-zero run is indicated by a delimiter code for indicating a data delimiter. 13. The non-zero run coded data is represented by a code indicating the length and added to the head of the non-zero run coded data. Image data compression method. 14. The method of compressing image data according to claim 13, wherein the code indicating the length of the non-zero run is a modified Huffman code. 15. The division into the spatial frequency components is first performed by first and second components of a low frequency component in both horizontal and vertical directions of the screen, one of which is a low frequency component and the other of which is a second component of a high frequency component. The third component is divided into a fourth component, which is a high-frequency component, and the first component is regarded as image data. A fifth component, which is a low-frequency component in both the horizontal and vertical directions of the screen, and one of which is a low-frequency component. The sixth component and the seventh component of the component are the high-frequency components, and the second component is the eighth component of the high-frequency component. By repeating this division, both components divide the low-frequency component a plurality of times, The image data compression method according to claim 1, wherein Haar transform is used for the division.