JP2004229101A - Image encoding system and its method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve image compression while maintaining the image quality. <P>SOLUTION: An image encoding system is provided with a converting means for converting an image into a plurality of subbands, a correlation calculating means for calculating a correlation between coefficients in which a position in each subband is identical with the other subband of one and the same resolution for each subband, and a coefficient operating means for changing the value of a coefficient with a small correlation. <P>COPYRIGHT: (C)2004,JPO&NCIPI

Description

式１が成立した時に相関があると判定し、式１が成立しなかった場合に相関がないと判定すれば良い。但し、式１において、ＡＢＳ（Ｘ）はＸの絶対値を表す。
【００３９】
相関判定部分に置いて、先に述べた実施例については、式１により各サブバンドデータとの差分で相関を判定していた。
【００４０】
相関を判定する際に、処理するサブバンド内のデータを微分し、微分値を比較して相関を判定してもかまわない。
【００４１】
ここで、ＨＬサブバンドとＨＨサブバンドの相関を判定する方法について説明する。ＨＬサブバンドデータの注目係数ＨＬｓ（ｑ、ｒ）を処理する場合、微分値は、下記の式により求める。
【００４２】
ｄｘＨＬｓ（ｑ、ｒ） ＝ ＨＬｓ（ｑ、ｒ）−ＨＬｓ（ｑ−１，ｒ）
ｄｙＨＬｓ（ｑ、ｒ） ＝ ＨＬｓ（ｑ、ｒ）−ＨＬｓ（ｑ，ｒ−１）
次にＨＨサブバンドの微分値も同様に求める。
【００４３】
ｄｘＨＨｓ（ｑ、ｒ） ＝ ＨＨｓ（ｑ、ｒ）−ＨＨｓ（ｑ−１，ｒ）
ｄｙＨＨｓ（ｑ、ｒ） ＝ ＨＨｓ（ｑ、ｒ）−ＨＨｓ（ｑ，ｒ−１）
さらに下記式により、相関があるか判定する。
【００４４】

式２又は式３が成立した場合に相関があると判定する。但し、式２、式３において、ＡＢＳ（Ｘ）はＸの絶対値を表す。
【００４５】
上記のように、注目係数の周辺係数を用いて演算を行い、サブバンド間の相関があるかどうかを判定する事ができる。
【００４６】
また、上記のしきい値はサブバンド毎に切り替えてもよい。
【００４７】
【発明の効果】
本発明では、ＪＰＥＧ２０００の算術符号化において、解像度レベル１（ＮＬ＝１）以上のサブバンドデータの画質劣化を抑えて量子化を行うことができ、ノイズ成分を低減する事により圧縮性能を向上させることができる。
【図面の簡単な説明】
【図１】本発明の実施形態による画像符号化システムの構成を示すブロック図である。
【図２】従来例による画像符号化システムの構成を示すブロック図である。
【図３】画像信号のウェーブレット変換後のサブバンド構成を示す図である。
【図４】本発明の実施形態による画像符号化方法のうちの相関判定部が行うものを示す第１のフローチャートである。
【図５】本発明の実施形態による画像符号化方法のうちの相関判定部が行うものを示す第２のフローチャートである。
【図６】本発明の実施形態による画像符号化方法のうちの相関判定部が行うものを示す第３のフローチャートである。
【符号の説明】
１ ＤＣレベル変換、色変換部
２ ウェーブレット変換部
３ 相関判定部
４ 係数モデリング部
５ 算術符号化部
６ 符号順序制御部
１０ 量子化部[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to an image encoding system and method, and more particularly, to an image encoding system and method for converting an image into a plurality of subbands and encoding the image.
[0002]
[Prior art]
In JPEG2000 (ISO / IEC 15444), which is an international standard of ISO / IEC, image quality at a low bit rate is improved as compared with conventional JPEG (ISO / IEC 10918). JPEG2000 allows image compression while allowing image distortion, and it is possible to add distortion to an image while adjusting image quality on the encoder side, and effectively compress the image.
[0003]
[Patent Document 1]
JP 2002-135594 A [Patent Document 2]
Japanese Patent Application Laid-Open No. 2002-252759
[Problems to be solved by the invention]
Looking only at the compression performance, there is not much improvement over JPEG.
[0005]
An object of the present invention is to provide an image encoding system and method for improving image compression performance while maintaining image quality in JPEG2000.
[0006]
[Means for Solving the Problems]
According to the present invention, a correlation between a coefficient having the same position in each sub-band between another sub-band having the same resolution for each sub-band and a conversion unit for converting an image into a plurality of sub-bands is calculated. An image coding system is provided, comprising: a correlation calculating unit that performs a correlation calculation; and a coefficient operating unit that changes a value of the coefficient having a small correlation.
[0007]
In the above image encoding system, the correlation calculation means may calculate that the correlation between the coefficients is small when the absolute value of the difference between the coefficients is equal to or greater than a predetermined value.
[0008]
In the above-described image encoding system, the correlation calculation means may calculate that the correlation between the coefficients is small when the absolute value of the difference between the derivatives regarding the coefficient positions of the coefficients is equal to or greater than a predetermined value.
[0009]
In the above-described image encoding system, the coefficient operation unit may set the value of the coefficient having a small correlation to zero.
[0010]
In the above image encoding system, the coefficient operation unit may set the value of the coefficient having a small correlation to a value calculated from coefficients around the coefficient.
[0011]
BEST MODE FOR CARRYING OUT THE INVENTION
The present invention quantizes subband data after wavelet transform has been performed in JPEG2000 encoding with reference to subband data of the same resolution level, thereby suppressing image quality degradation and efficiently quantizing data. It is characterized by the following.
[0012]
FIG. 2 shows the procedure of the JPEG2000 encoding method. DC level conversion (level shift) is performed on input image data, and color conversion processing is performed on a color image. Wavelet transform processing is performed on the converted image to convert the image into coefficients. The transformed coefficients are quantized by a quantizer. After performing the coefficient modeling process on the quantized coefficients from multi-valued data to binary data, the arithmetic encoder performs bit plane coding. Finally, the code order is rearranged and output.
[0013]
Next, FIG. 3 shows coefficient data after the wavelet transform. Since the wavelet transform is a sub-band transform that reduces the resolution to １ ／, it has a hierarchical structure as shown in FIG. In the wavelet transform, data having a resolution level of 1 or more has subband data of HL, LH, and HH. The subband data is composed of an m × n data sequence, and each coefficient data is k-bit data.
[0014]
If the resolution levels are the same, the subband data has a data array of HL size = LH size = HH size = m × n.
[0015]
As shown in FIG. 1, a correlation determination unit determines whether or not there is a correlation between subband data of the same resolution level among HL, LH, and HH subbands, and performs a quantization process based on the determination result.
[0016]
This quantization process removes noise components in the sub-band. Even if the noise component in the sub-band is removed, the influence on the quality deterioration of the image is small. Further, since the noise component is removed, there is a feature that the compression rate of the encoded data can be improved.
[0017]
A configuration of an embodiment of a JPEG2000 quantizer using the present invention will be described.
[0018]
FIG. 1 shows a JPEG2000 encoder of the present invention. DC level conversion (level shift) and color conversion processing 1 are performed on the input image data. Wavelet transform processing 2 is performed on the converted image to convert the image into coefficients. The correlation determination unit 10 determines the correlation of the subband data after the wavelet conversion on the converted coefficients, and quantizes the coefficients. Next, quantization is performed by the quantizer 3. After performing the coefficient modeling process 4 on the quantized coefficients from multi-valued data to binary data, the arithmetic encoder 5 performs bit plane coding. Finally, the code order control unit 6 rearranges and outputs the code order.
[0019]
An embodiment of the present invention will be described with reference to FIGS.
[0020]
The wavelet transform unit 2 shown in FIG. 1 performs sub-band transform for reducing the resolution of the input image to １ ／ as shown in FIG. It has a hierarchical structure as shown in FIG.
[0021]
LL (NL = 0) 7 in FIG. 3 is a coefficient of the resolution level 0.
[0022]
HL (NL = 1) 8, LH (NL = 1) 8 and HH (NL = 1) 8 in FIG.
[0023]
HL (NL = 2) 9, LH (NL = 2) 9, and HH (NL = 2) 9 in FIG.
[0024]
As shown in FIG. 3, the data of the resolution level 1 or higher has subband data of HL, LH, and HH. The subband data is composed of an m × n data sequence, and each coefficient data is k-bit data.
[0025]
For the coefficients subjected to the wavelet transform, subband data for each resolution level is input to the correlation determination unit 10. The correlation determination unit 10 determines whether there is a correlation between HH, HL, and LH subband data for each resolution level for the subband data having a resolution level of 1 or more.
[0026]
Next, the correlation determination method will be described with reference to FIGS. 4, 5 and 6 when the resolution level is p.
FIG. 4 shows a method for determining the correlation of HL (NL = p). The data of the subband HL (NL = p) is input from the HL data input unit 11. Next, the correlation between the HL (NL = p) subband data composed of an m × n array and the LH (NL = p) subband data having the same resolution level input by the LH correlation determination unit 12 is obtained. Determine if there is. The correlation is determined for each coefficient s (a, b) of the m × n data array. Here, a and b take values in the following ranges.
[0027]
1 ≦ a ≦ m
1 ≦ b ≦ n
When there is a correlation between the noted coefficient s (a, b) of the HL subband data and the LH subband data, the process proceeds to the step of the coefficient data holding unit 14 to hold the value of the coefficient s (a, b). I do. If there is no correlation between the coefficient of interest s (a, b) of the HL sub-band data, it is determined whether there is a correlation with the HH sub-band data 13. Here, when there is a correlation with the HH sub-band data, the process proceeds to the step of the coefficient data holding unit 14 to hold the value of the coefficient s (a, b). When there is no correlation with the HH sub-band data, the process proceeds to the step of the coefficient clear unit 15. In the coefficient clearing unit 15, since there is no correlation between the attention coefficient s (a, b) of the HL subband and the subband data at the same resolution level, the attention coefficient s (a, b) in the HL subband data is a noise component. And sets s (a, b) to '0'.
[0028]
The correlation of each coefficient of the HL (NL = p) subband data is checked, and when the quantization of the coefficients is completed, the process proceeds to the step of the quantized data output unit 16 to output the HL subband data.
[0029]
Similarly, correlation determination is performed for LH subbands having the same resolution level as shown in FIG. FIG. 5 shows a method for determining the correlation of LH (NL = p). Data of the subband LH (NL = p) is input from the LH data input unit 17. Next, the correlation between the HL (NL = p) subband data of the same resolution level as the LH (NL = p) subband data composed of the m × n array input by the HL correlation determination unit 18 is determined. Determine if there is. The correlation is determined for each coefficient s (a, b) of the m × n data array. Here, a and b take values in the following ranges.
[0030]
1 ≦ a ≦ m
1 ≦ b ≦ n
When there is a correlation between the coefficient of interest s (a, b) of the LH subband data and the HL subband data, the process proceeds to the step of the coefficient data holding unit 20 to hold the value of the coefficient s (a, b). I do. If the coefficient of interest s (a, b) of the LH subband data has no correlation, it is determined whether there is a correlation with the HH subband data 19. If there is a correlation with the HH sub-band data, the process proceeds to the step of the coefficient data holding unit 20 to hold the value of the coefficient s (a, b). If there is no correlation with the HH sub-band data, the process proceeds to the step of the coefficient clear unit 21. In the coefficient clear unit 21, since there is no correlation between the attention coefficient s (a, b) of the LH subband and the subband data at the same resolution level, the attention coefficient s (a, b) in the LH subband data is a noise component. And sets s (a, b) to '0'.
[0031]
The correlation of each coefficient of the LH (NL = p) sub-band data is checked, and when the quantization of the coefficient is completed, the process proceeds to the step of the quantized data output unit 22 to output the LH sub-band data.
[0032]
Similarly, correlation determination is performed for HH subbands of the same resolution level as shown in FIG. FIG. 6 shows a method for determining the correlation of HH (NL = p). Data of the subband HH (NL = p) is input from the HH data input unit 23. Next, the correlation between the HL (NL = p) subband data of the same resolution level and the HH (NL = p) subband data composed of an m × n array input by the HL correlation determination unit 24 is determined. Determine if there is. The correlation is determined for each coefficient s (a, b) of the m × n data array. Here, a and b take values in the following ranges.
[0033]
1 ≦ a ≦ m
1 ≦ b ≦ n
When there is a correlation between the noted coefficient s (a, b) of the HH sub-band data and the HL sub-band data, the process proceeds to the step of the coefficient data holding unit 26 to hold the value of the coefficient s (a, b). I do. If there is no correlation in the coefficient of interest s (a, b) of the HH sub-band data, it is determined whether there is a correlation with the LH sub-band data 25. Here, when there is a correlation with the LH sub-band data, the process proceeds to the step of the coefficient data holding unit 26, and the value of the coefficient s (a, b) is held. If there is no correlation with the LH sub-band data, the process proceeds to the step of the coefficient clear unit 27. In the coefficient clearing unit 27, since the attention coefficient s (a, b) of the HH sub-band has no correlation with the sub-band data of the same resolution level, the attention coefficient s (a, b) in the HH sub-band data is a noise component. And sets s (a, b) to '0'.
[0034]
The correlation of each coefficient of the HH (NL = p) sub-band data is checked, and when the quantization of the coefficient is completed, the process proceeds to the step of the quantized data output unit 28 to output the HH sub-band data.
[0035]
The correlation is determined as described above. The processing order of the subbands LH, HL, HH may be performed in any order. Further, when determining the correlation between subbands having the same resolution, subband data after correlation determination may be used. Although the subband data for which no correlation is recognized is described as "0" in the coefficient clearing section here, a value other than "0" may be used. For example, a coefficient for which no correlation has been recognized may be replaced with a value calculated from surrounding coefficients of the target coefficient and quantized.
[0036]
Next, an embodiment of the correlation determination will be described. An example in which the correlation between the HL sub-band and the HH sub-band is determined will be described below.
[0037]
The correlation between the attention coefficient HLs (a, b) of the HL subband and the corresponding attention coefficient HHs (a, b) of the subband data HH of the same resolution is determined by the following equation.
[0038]

It is only necessary to determine that there is a correlation when Expression 1 is satisfied, and determine that there is no correlation when Expression 1 is not satisfied. However, in Expression 1, ABS (X) represents the absolute value of X.
[0039]
In the correlation determination part, in the above-described embodiment, the correlation was determined by the difference from each subband data according to Equation 1.
[0040]
When determining the correlation, the data in the sub-band to be processed may be differentiated, and the differential value may be compared to determine the correlation.
[0041]
Here, a method of determining the correlation between the HL subband and the HH subband will be described. When processing the attention coefficient HLs (q, r) of the HL subband data, the differential value is obtained by the following equation.
[0042]
dxHLs (q, r) = HLs (q, r) -HLs (q-1, r)
dyHLs (q, r) = HLs (q, r) -HLs (q, r-1)
Next, the differential value of the HH sub-band is similarly obtained.
[0043]
dxHHs (q, r) = HHs (q, r) -HHs (q-1, r)
dyHHs (q, r) = HHs (q, r) -HHs (q, r-1)
Further, it is determined whether there is a correlation by the following equation.
[0044]

When Equation 2 or Equation 3 is satisfied, it is determined that there is a correlation. However, in Equations 2 and 3, ABS (X) represents the absolute value of X.
[0045]
As described above, it is possible to determine whether or not there is a correlation between subbands by performing an operation using peripheral coefficients of a target coefficient.
[0046]
Further, the above threshold may be switched for each subband.
[0047]
【The invention's effect】
According to the present invention, in arithmetic coding of JPEG2000, quantization can be performed while suppressing image quality deterioration of subband data having a resolution level of 1 (NL = 1) or more, and compression performance is improved by reducing noise components. be able to.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a configuration of an image encoding system according to an embodiment of the present invention.
FIG. 2 is a block diagram showing a configuration of a conventional image coding system.
FIG. 3 is a diagram illustrating a subband configuration of an image signal after wavelet transform.
FIG. 4 is a first flowchart illustrating a process performed by a correlation determination unit in the image coding method according to the embodiment of the present invention;
FIG. 5 is a second flowchart illustrating a process performed by the correlation determination unit in the image coding method according to the embodiment of the present invention.
FIG. 6 is a third flowchart showing the one performed by the correlation determination unit in the image coding method according to the embodiment of the present invention;
[Explanation of symbols]
REFERENCE SIGNS LIST 1 DC level conversion, color conversion unit 2 wavelet conversion unit 3 correlation determination unit 4 coefficient modeling unit 5 arithmetic coding unit 6 code order control unit 10 quantization unit

Claims (10)

Conversion means for converting the image into a plurality of subbands;
Correlation calculating means for calculating a correlation between coefficients having the same position in each subband between other subbands having the same resolution for each subband,
Coefficient operating means for changing the value of the coefficient having a small correlation,
An image encoding system comprising: The image encoding system according to claim 1,
The image coding system according to claim 1, wherein the correlation calculating means calculates that the correlation between the coefficients is small when the absolute value of the difference between the coefficients is equal to or larger than a predetermined value. The image encoding system according to claim 1,
An image coding system according to claim 1, wherein said correlation calculating means calculates that the correlation between the coefficients is small when the absolute value of the difference between the derivatives regarding the coefficient positions of the coefficients is equal to or greater than a predetermined value. The image encoding system according to claim 1,
The image encoding system according to claim 1, wherein the coefficient operation unit sets the value of the coefficient having a small correlation to zero. The image encoding system according to claim 1,
The image encoding system according to claim 1, wherein the coefficient operation unit sets the value of the coefficient having a small correlation to a value calculated from coefficients around the coefficient. A conversion step of converting the image into a plurality of subbands;
A correlation calculation step of calculating a correlation between coefficients having the same position in each subband between other subbands having the same resolution for each subband,
A coefficient operation step of changing the value of the coefficient having a small correlation,
An image encoding method comprising: The image encoding method according to claim 6,
The image coding method according to claim 1, wherein the correlation calculation step calculates that the correlation between the coefficients is small when the absolute value of the difference between the coefficients is equal to or larger than a predetermined value. The image encoding method according to claim 6,
The image coding method according to claim 1, wherein the correlation calculating step calculates that the correlation between the coefficients is small when the absolute value of the difference between the derivatives regarding the coefficient positions of the coefficients is equal to or larger than a predetermined value. The image encoding method according to claim 6,
The image encoding method, wherein the coefficient operation step sets the value of the coefficient having a small correlation to zero. The image encoding method according to claim 6,
The image encoding method, wherein the coefficient operation step sets the value of the coefficient having a small correlation to a value calculated from coefficients around the coefficient.

Images