JP2008005343A - Analysis filter bank apparatus, analysis filter bank method, analysis filter bank program, synthesis filter bank apparatus, synthesis filter bank method, synthesis filter bank program, image compression apparatus, image compression method, image decompression apparatus, and image decompression method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a filter bank apparatus capable of more suppressing end points from increasing while more reducing calculation load and memory consumption than a convolutional operation by executing the symmetrical extension of an image boundary part by using lifting. <P>SOLUTION: An extension part EA<SB>0</SB>[ ] 39 receives a down sampling signal A(z) and extends an end point on the outside of an image boundary part. After extending the end point by the extension part EA<SB>0</SB>[ ] 39, a prediction filter P(z) 40 filters the extended signal and an omission part EB<SB>0</SB>[ ] 41 executes end point omitting operation of the filtered result. A subtractor 42 finds out a difference between the omission result of the omission part EB<SB>0</SB>[ ] 41 and a down sampling signal B(z) and the subtraction result becomes a high frequency component H(z). <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to an analysis filter bank apparatus, an analysis filter bank method, and an analysis filter bank program having a lifting structure for adding or subtracting a filtering process signal from both a low frequency side and a high frequency side to a counterpart side. The present invention also relates to a synthesis filter bank apparatus, a synthesis filter bank method, and a synthesis filter bank program having a lifting structure that adds or subtracts a filtering processing signal from both the low frequency side and the high frequency side to the counterpart side. The present invention also relates to an image compression apparatus and an image compression method using the filter bank apparatus and the filter bank method. The present invention also relates to an image decompression apparatus and an image decompression method using the synthesis filter bank apparatus and the synthesis filter bank method.

  2. Description of the Related Art A filter bank device that filters a signal, particularly an image signal, and decomposes the signal into frequencies is known. The filter bank is used in the pre-stage of image analysis and compression, and is also used in the international standard JPEG and JPEG2000. As an application range of the present invention, there are an image compression device and an expansion device. Specifically, a digital camera, a video camera, a portable / mobile image transmission / reception terminal (PDA), a nonlinear editing device, and a wireless image transmission device in a broadcasting station. , A compression / decompression unit for a printer, a satellite image, a medical image, or the like, or a software module thereof, a game compression / decompression unit for use in 3D CG, or a software module thereof.

  The image signal is a finite signal, and there is no signal at the boundary portion of the image, and discontinuity always occurs at that portion. Therefore, many studies and examinations have been made as to how the filtering of the image boundary portion is less affected by the discontinuity.

  “A development of symmetric extension methods for subband image coding” published in IEEE Transactions on Image Processing., Vol.3.1 in 1994 is a method for symmetrically extending the signal at the image boundary (so that it becomes a mirror image). Has proposed.

  Also, JPEG2000, which is an international ISO standard, uses symmetric expansion when performing wavelet transform and wavelet inverse transform. This has the effect of reducing distortion at the boundary between images and at the boundary between tiles.

  By the way, as a means for filtering an image signal, the most general calculation method is a means called a convolution calculation. This is the most basic means for realizing a digital filter, and convolutionally multiplies the tap coefficient of the filter with actual input data. However, this drawback is that if the tap length is long, the calculation load increases accordingly.

  As a means to solve this problem, W. Swelden introduced his paper in Applied and Computational Harmonic Analysis 3, 186-200 in 1996, “The Lifting Scheme: A Custom Design Construction of Biorthogonal Wavelets”. There is a lifting means.

“A development of symmetric extension methods for subband image coding” IEEE Transactions on Image Processing., Vol.3.11994 "The Lifting Scheme: A Custom Design Construction of Biorthogonal Wavelets" Applied and Computational Harmonic Analysis 3, 186-200. 1996. W. Swelden.

  However, the symmetric expansion has an effect of reducing the discontinuity of the boundary portion, but there is a possibility that the end points other than the original signal increase due to the expansion.

  The present invention has been made in view of the above circumstances, and by performing symmetrical expansion of an image boundary portion using lifting, a filter bank that further suppresses an increase in endpoints while reducing calculation load and memory consumption compared to a convolution operation. An object is to provide a device, a filter bank method, and a filter bank method.

  The present invention has been made in view of the above circumstances, and by performing symmetrical expansion of the image boundary portion using lifting, a synthesis filter that further suppresses an increase in endpoints while reducing calculation load and memory consumption compared to convolution calculation. An object is to provide a bank apparatus, a synthesis filter bank method, and a synthesis filter bank method.

  In order to solve the above-described problem, the analysis filter bank apparatus according to the present invention divides an input signal into different sets of signals, and the other side from both the first set and the second set obtained by the division. An analysis filter bank apparatus having a lifting structure for adding or subtracting a filtering process signal to a prediction processing system for performing a prediction process for predicting a second set using the first set, First extension means for extending a down-sampled signal obtained by down-sampling one set, and prediction filter means for performing processing for predicting the second set using the signal extended by the first extension means; First truncation means for truncating the extended signal included in the filter processing signal output by the prediction filter means, and processing from the prediction processing system; An update processing system for performing an update process for updating the first set using the result includes a downsampling signal obtained by downsampling the second set and a truncation from the first truncation unit of the prediction processing system. An update that is provided in the update processing system so as to form a pair with the second extension means for extending a difference signal from the output and the prediction filter means and performs an update process on the signal extended by the second extension means. And a second truncation unit for truncating the extension signal included in the filter processing signal output from the update filter unit, and the truncation output from the second truncation unit is downsampled. The addition signal added to the set of 1 and the difference signal are used as an analysis filter output signal.

  In order to solve the above-described problem, the synthesis filter bank apparatus according to the present invention has a lifting structure that adds or subtracts the filtering signal to or from the other side from both the first signal set and the second signal set. An update processing system for performing an update process for updating the first set using the second set, and a first extension unit that extends the second set; Update filter means for performing update processing on the signal extended by the first extension means, and first truncation means for truncating the signal for the extension included in the filter processing signal output by the update filter means, The prediction processing system that predicts the second set using the processing result of the update processing system is the first processing and the first truncation unit of the update processing system from the first truncation means. Prediction processing is performed on the signal that is provided in the prediction processing system so as to be paired with the second expansion means for extending the difference signal from the discarded output and the update filter means, and is expanded by the second expansion means. And a second truncation unit that truncates the signal of the extension included in the filter processing signal output from the prediction filter unit. The truncation output from the second truncation unit is the second truncation output. A shift signal of a signal obtained by up-sampling the addition signal added to the set and a signal obtained by up-sampling the difference signal are combined to obtain a combined filter output signal.

  According to the present invention, complete reconstruction is possible on the analysis side and the synthesis side, which is an excellent feature of the lifting means, and even if end point expansion and end point removal are performed in the lifting process, the deterioration is not affected. Therefore, when performing symmetrical extension at the image boundary, it is not necessary to increase the number of end points, so that the calculation load of filtering does not increase. For this reason, in the image compression process and the image expansion process, the calculation time can be shortened and the configuration is not complicated.

The best mode for carrying out the present invention will be described below with reference to the drawings. First, the general configuration of the filter bank will be described before the description of the present embodiment. FIG. 1 shows an example in which octave division, which is the most popular wavelet transform among several methods, is performed over a plurality of levels. In the case of the figure, the number of levels is 3, and the input signal 110 is divided into a low band and a high band by the low band analysis filter H ₀ (z) 11 and the high band analysis filter H ₁ (z) 12. In addition, only the low frequency components are hierarchically divided.

Further, FIG. 1 shows a wavelet transform for a one-dimensional signal (for example, a horizontal component of an image) for convenience, but it is possible to deal with a two-dimensional image signal by extending this to two dimensions. Next, the operation will be described. The input image signal 110 is first band-divided by the low-frequency analysis filter H ₀ (z) 11 and the high-frequency analysis filter H ₁ (z) 12, and the obtained low-frequency component and high-frequency component are obtained by the down sampler 13. , The resolution is reduced by half (level 1).

There are two outputs at this time: an L component 111 and an H component 116. Then, only the low-frequency component in the thinned out band is again divided by the low-pass filter H ₀ (z) and the high-pass filter H ₁ (z), and the down-sampler 13 reduces the resolution by half. Drawn (level 2).

  By performing this processing to a predetermined level, band components obtained by hierarchically dividing the low frequency component into bands are sequentially generated. The band components generated at level 2 are the LL component 112 and the LH component 115. FIG. 1 shows that as a result of band division to level 3, LLL component 113, LLH component 114, LH component 115, and H component 116 are generated (L is low and low, H is high and high Showing).

On the other hand, FIG. 2 shows a wavelet inverse transform configuration that performs the reverse operation of FIG. First, the LLL component 113 and the LLH component 114 of level 3 are up-sampled twice, and then band-synthesized by the low-frequency synthesis filter F ₀ (z) 15 and the high-frequency synthesis filter F ₁ (z) 16. The combined result of both is added by the adder 17 and the LL component 117 is output.

Next, the level 2 LH component 115 is up-sampled by a factor of 2, and then band-synthesized by the high-frequency synthesis filter F ₁ (z). On the other hand, the LL component 117 is upsampled by a factor of 2, and then band-synthesized by the low-frequency synthesis filter F ₀ (z). Both results are added by an adder, and the result is output as an L component 118.

  By performing the above operation up to level 1, an output signal 119 is finally generated. FIG. 3 illustrates a band generated when the above-described one-dimensional wavelet transform is performed on a two-dimensional image signal. As a result of wavelet transform up to 2 levels, it can be seen that 7 bands are generated.

  Next, a conventional example of an analysis filter bank using lifting means will be described with reference to FIG. This conventional example is given here to illustrate the course leading to the present invention. In the figure, 18 is a half downsampler, 19 is a delay unit, 20 is a subtractor, 21 is a rounder, 22 is a Predict filter, 23 is an adder, 24 is a rounder, and 25 is an Update filter. The name “lifting” is derived from the operation of adding or subtracting the filtered coefficient from both the low side and the high side to the other side as shown in the figure.

Next, the operation will be described. Of the input signal 120, even-numbered pixels are divided as X _even and odd-numbered pixels are divided as X _odd . X _odd is delayed by the delay unit 19 and the resolution is reduced by half by the down sampler 18 to generate B (z).

On the other hand, the resolution of X _even is reduced by half by the down sampler 18 to obtain A (z), and a Predict filter P (z) is applied to this. Further, after that, a rounding process is performed in a rounder, and a difference is taken between the result and B (z). This is sent out as the high frequency output H (z). In the case of an image, the above time delay means that one pixel or one coefficient is shifted.

  On the other hand, addition is performed between the signal obtained by rounding the output obtained by applying the Update filter U (z) to H (z) and the A (z), and the addition result is the low-frequency side. Is output as L (z). The above is the basic operation by lifting.

Here, the filter coefficients of the above P (z) and U (z) will be described below. For example, when realizing a 9-tap filter for the low-pass filter and a 7-tap filter for the high-pass filter, the filter coefficient (z conversion expression) can be expressed by the following equations (1) and (2).
P (z) = (-1 + 9z ^-1 + 9z ^-2 -z ^-3 ) / 16 (1)
U (z) = (1 + z ^-1 ) / 4 (2)
Therefore, P (z) has 4 taps, U (z) has 2 taps, and both have an excellent feature that division can be performed by bit shift (because it is a power of 2). At this time, the respective filter coefficients (z conversion expression) of the low-pass filter L (z) and the high-pass filter H (z) are expressed by the following equations (3) and (4).
L (z) = (1-8z ^-2 + 16z ^-3 + 46z ^-4 + 16z ^-5 -8z ^-6 + z ^-8 ) / 64 (3)
H (z) = (1-9z ^-2 + 16z ^-3 -9z ^-4 + z ^-6 ) / 16 (4)
However, since the filter coefficient is an integer, it is common to use rounding to increase the calculation accuracy in filtering with actual pixels. For example, when the pixel array (a, b, c, d) is filtered with the filter of P (z), it is called round (a × (-1) + b × 9 + c × 9 + d × (-1)) Perform the operation (round is rounded off). Accordingly, the rounding rounder of FIG. 1 performs the rounding operation described above, which may cause a rounding error due to rounding.

  Next, an embodiment of a synthesis filter bank using lifting means will be described with reference to FIG. The operation of the filter bank is to input the low-frequency component L (z) and the high-frequency component H (z) and output Y (z).

  The high-frequency component H (z) is filtered by the Update filter U (z), and then rounded, and the result is compared with L (z). Therefore, it can be seen from the relationship with FIG. 4 that this difference result completely matches A (z).

  Next, after filtering A (z) with the Predict filter P (z), rounding is performed, and the result is added to H (z). From the relationship with FIG. 4, it can be seen that this addition result completely matches B (z).

  The signal Y (z) obtained by adding the signals generated by upsampling A (z) and B (z) twice each is the same as the input X (z) in FIG. It becomes complete reconstruction.

  The above is the configuration and the realization means of the analysis filter bank and the synthesis filter bank using the lifting means, and as described above, it is based on the prior art. In this conventional configuration, even if distortion due to round (rounding processing) occurs in the lifting means in the lifting calculation process, this is canceled on the analysis side and the synthesis side.

  This means that even if processing such as truncation of some signal components is performed instead of Round in FIGS. 4 and 5, canceling it inside the lifting can remove the adverse effects such as distortion and deterioration. I can expect.

  Therefore, in the symmetrical extended filtering at the boundary portion of the image described in the object of the present invention, even if the signal component increased by extending the number of end points is discarded, the advantage of lifting is achieved without degradation. A complete reconfiguration is possible.

  Further, by utilizing the advantage of the lifting according to the present invention, it is possible to reduce the calculation load due to the increase in the number of end points, which has been a problem in the past, when performing symmetrical extended filtering at the image boundary.

FIG. 6 is a configuration diagram of an embodiment of the present invention. The difference from FIG. 4 is that the extension unit EA ₀ [] 39 is provided at the front stage of the prediction filter P (z) 40, and the truncation unit EB ₀ [] is provided at the rear stage of the prediction filter P (z) 40 instead of Round. 41, and an extension 46EA ₁ [] is provided at the front stage of the update filter U (z) 45, and a truncation part EB ₁ [] 44 is provided at the rear stage of the update filter U (z) 45 instead of Round. It is a prepared point.

  That is, the analysis filter bank device divides the input signal into different signal sets, and adds the filtering signal to the other party from both the first set and the second set obtained by the division or An analysis filter bank apparatus having a lifting structure for subtraction, wherein a prediction process for predicting a second set (odd signal Xodd (z)) using the first set (even signal Xeven (z)) is performed. The prediction processing system to be performed is expanded by a first expansion unit 39 that expands a down-sampling signal A (z) obtained by down-sampling the resolution of the first set by the down-sampler 36 of 1/2. A prediction filter 40 that performs the process of predicting the second set using the processed signal, and a first cutoff that truncates the extended signal included in the filter processing signal output by the prediction filter 40. An update processing system that includes a discarding unit 41 and performs an update process for updating the first set using a processing result from the prediction processing system down-converts the second set with a ½ downsampler 38. The prediction filter 40 is paired with the second extension unit 46 that extends the difference signal between the sampled down-sampled signal B (z) and the output of the truncation from the first truncation unit 41 of the prediction processing system. An update filter 45 that is provided in the update processing system and performs an update process on the signal extended by the second extension unit 46, and a second signal that truncates the extension signal included in the filter processing signal output by the update filter 45 And a sum signal L (z) obtained by adding the truncation output from the second truncation unit 44 to the down-sampled first set, and the difference And No. H (z) and the analysis filter output signal.

  The configuration shown in FIG. 6 is used as the circuit portion 201, 202, or 203 for one level shown in FIG. In FIG. 6, each pixel of the input signal 120 is divided into an even-numbered pixel Xeven and an odd-numbered pixel Xodd. The even-numbered pixel Xeven is reduced in half by the ½ downsampler 36 to be a downsampled signal A (z), and is supplied to the extension unit 39 and the adder 43 of the prediction processing system. On the other hand, the odd-numbered pixel Xodd is time-delayed by the delay unit 37, and further the resolution is reduced by half by the 1/2 downsampler 38 to be the downsampled signal B (z). Supplied.

The operation of the analysis filter bank apparatus shown in FIG. 6 will be described below with reference to the flowchart of FIG. First, in the extension unit EA ₀ [] 39 for the down-sampling signal A (z), for example, the end point outside the image boundary part is extended when filtering by the prediction filter P (z) 40 (step) S1). Here, the symmetric expansion means already described may be used. Figure 8 is a depiction of symmetric extension method, the boundary of the image R _T of the solid line region becomes actually subject to filtering, the dotted line portion is the outer circumference of the range R _F filtering spans. In the horizontal and vertical directions, an operation for compensating external coefficients using coefficients inside the image is performed. At this time, symmetrical expansion is performed so that the coefficients inside and outside the image have a mirror image relationship with the boundary as the center. That is, in FIG. 8, a, b, c, d, e, and f all represent pixels (coefficients). For example, when the pixel a is filtered in the horizontal direction, the pixels a, b, and c in the original image _RT to be filtered are symmetrically extended to the filtering range R _F so as to have a mirror image relationship with the solid line as a reference. .

After the end point extension is performed by the extension unit EA ₀ [] 39, the extended signal is filtered by the prediction filter P (z) 40 (step S2), and the filtering result is obtained by the truncation unit EB ₀ [] 41. An end point truncation operation is performed (step S3). In FIG. 8, this means that the coefficients in the outer peripheral area (a, b, c in the horizontal direction and d, e, f in the vertical direction) are discarded.

The result of the truncation by the truncation unit EB ₀ [] 41 is subtracted by the subtractor 42 from the downsampling signal B (z), and this result becomes the high frequency component H (z) (step S4). Next, the high-frequency signal H (z) is an extension of the end point outside the image boundary portion, which is necessary when the update filter U (z) 45 is filtered by the extension unit EA ₁ [] 46 of the second signal processing system. (Step S5). After the end point extension is performed by the extension unit EA ₁ [] 46, the update signal U (z) 45 is filtered for the extended signal (step S 6), and the filtering result is obtained by the truncation unit EB ₁ [] 44. An end point truncation operation is performed (step S7). The truncation processing result of the truncation unit EB ₁ [] 44 is added to the down-sampling signal A (z) by the adder 43 to output a low-frequency signal L (z) (step S8).

The above operation is shown in the following mathematical expression. First, when the input signal X (z) is distributed to the even point A (z) and the odd point B (z), the following equation (5) is obtained.
X (z) = A (z ² ) + z ⁻¹ B (z ² ) (5)
Next, as a result of the first stage operation of the lifting means being performed on A (z) and B (z), the output H (z) is given by the following equation (6).
H (z) = B (z)-EB ₀ [EA ₀ [A (z)] P (z)] (6)
Further, as a result of the second stage operation being performed on H (z), the output L (z) is given by the following equation (7).
L (z) = A (z) + EB ₁ [EA ₁ [H (z)] U (z)] (7)
The above is the realization means by the forward lifting configuration.

After all since the equation (6) in EA ₀ There are endpoints increases in [A (z)], discarded endpoints truncation is performed by _{EB 0 [EA 0 [A (} z)] P (z)], B (z ) And the number of endpoints is the same. Similarly, after all because it is the end point increase in EA ₁ In Equation (7) [H (z) ], discarded endpoints truncation is performed by _{EB 1 [EA 1 [H (} z)] U (z)], A The number of end points is the same as (z). Therefore, the end points of L (z) and H (z), which are the final outputs, are the same as A (z) and B (z), respectively.

  Next, another embodiment will be described. FIG. 9 is a configuration diagram of synthesis filtering using lifting. The reverse processing of the analysis filter bank apparatus shown in FIG. 6 is performed. The output signal L (z) on the low frequency side and the output signal H on the high frequency side which are outputs from the analysis filter bank apparatus of FIG. (z) is supplied.

This configuration includes one differencer 47, one adder 54, one delay unit 57, two upsamplers 55 and 56, an update filter U (z) 49, a prediction filter P (z) 52, And an extension part EA ₁ [] 50 in the front part of the update filter U (z) 49, a truncation part EB ₁ [] 48 in the rear part of the update filter U (z) 49, and a part in the front part of the prediction filter P (z) 52. An extension unit EA ₀ [] 51 is provided, and a truncation unit 53 EB ₀ [] is provided in the subsequent stage.

  The low frequency output signal L (z) is supplied to the subtractor 47. The high-frequency output signal H (z) is supplied to the expansion unit 50 and the adder 54.

  That is, this synthesis filter bank is a filtering signal from the low frequency side (first signal set) L (z) and the high frequency side (second signal set) H (z) to the other side. Is a synthesis filter bank apparatus having a lifting structure for adding or subtracting, an update processing system for performing update processing for updating the first set using the second set is a high-frequency side (second Set) a first extension unit 50 that extends the input signal H (z), an update filter 49 that performs an update process on the signal extended by the first extension unit 50, and a filter processing signal output by the update filter 49 1 includes a first truncation unit 48 that truncates the signal included in the extension, and generates a filtering signal for the high frequency side H (z). The prediction processing system includes the low frequency side input signal L (z) and the first The first truncation of one signal processing system The second signal processing system forms a pair with the update filter 49 and the second extension unit 51 that extends the difference signal (subtracted output signal of the subtractor 47) A ′ (z) from the rounded down output from 48. A prediction filter 52 that performs prediction processing on a signal that is provided and expanded by the second expansion unit 51, and a second truncation unit 53 that truncates the signal for expansion included in the filter processing signal output by the prediction filter 52. Thus, the addition signal (addition output of the adder 54) B ′ (z) obtained by adding the truncation output from the second truncation unit 53 to the high frequency side input signal is upsampled by the double upsampling unit 56 The signal obtained by the signal delayer 57 and the signal obtained by upsampling the difference signal A ′ (z) twice by the upsampling unit 55 are combined by the combining unit 58 to obtain a combined filter output signal Y (z). .

  The configuration shown in FIG. 9 is used as the circuit portion 206, 207 or 208 for one level in FIG.

Hereinafter, these operations will be described with reference to the flowchart of FIG. First, the extension unit EA ₁ [] 50 extends the end point with respect to the high frequency side input signal H (z) (step S11). Next, the update filter 49 performs an update process on the expanded signal (step S12). In step S13, the truncation unit 48 truncates the end points with respect to the update filtering result. Next, in step S14, the difference signal (subtracted output signal of the subtractor 47) A between the low-frequency side input signal L (z) and the truncated output from the first truncating unit 48 of the first signal processing system. Take '(z) and upsample.

  Further, in step S 15, the end point of the difference signal (subtraction output signal of the subtractor 47) A ′ (z) is extended by the second extension unit 51. In step S <b> 16, a prediction process is performed on the signal expanded by the second expansion unit 51 using the prediction filter 52. In step S <b> 17, the extended signal included in the filter processing signal output by the prediction filter 52 is discarded by the second truncation unit 53. In step S18, the addition signal (addition output of the adder 54) B ′ (z) obtained by adding the truncation output from the second truncation unit 53 to the high-frequency side input signal is upsampled by the double upsampling unit 56. To do. In step S19, the upsampled signal is shifted by the delay unit 57. Finally, in step S20, the shift signal by the delay unit 57 and the signal obtained by up-sampling the difference signal A ′ (z) by the up-sampling unit 55 by the up-sampling unit 55 are synthesized by the synthesizing unit 58 and synthesized filter output signal Let Y (z).

According to the lifting configuration of FIG. 9, the inputs L (z) and H (z) are expressed as A ′ (z) and B ′ (z) in FIG. can get.
A '(z) = L (z)-EB ₁ [EA ₁ [H (z)] U (z)] (8)
B '(z) = H (z) + EB ₀ [EA ₀ [A (z)] P (z)] (9)
On the other hand, it is clear from FIG. 9 that the output result Y (z) of the synthesis filtering by lifting is the following expression (10).
Y (z) = A ′ (z ² ) + z ⁻¹ B ′ (z ² ) (10)
In the embodiment shown in FIG. 6, analysis filtering using lifting means has been described. An image compression apparatus can be configured by providing wavelet transform using this analysis filtering and performing quantization and encoding in the subsequent stage.

  FIG. 11 is a configuration diagram of the image compression apparatus. The wavelet transform unit 60 performs wavelet transform on the input signal 120 by a predetermined number of levels. The resulting wavelet transform coefficient 121 is quantized by the quantization unit 61.

As the quantization means, a commonly used scalar quantization (below: Equation 11) may be used.
Q = x / Δ (11)
Here, x is a wavelet transform coefficient value, and Δ is a quantization index value.

  The quantization coefficient 122 obtained by the above means is entropy-encoded in the entropy encoding unit 62, and an encoded bit stream 123 is output. As means used in the entropy coding unit 62, there are arithmetic coding means in addition to variable length coding means, and research results of these entropy coding means have been reported by each research institution. Should be used. For example, a method called EBCOT adopted in JPEG2000 and Huffman coding used in JPEG are examples of application.

  In the embodiment shown in FIG. 9 described above, the synthesis filtering using the lifting means has been described. If this synthesis filtering is used to provide wavelet inverse transformation, and decoding and inverse quantization are performed in the previous stage, an image expansion device can be configured.

  FIG. 12 is a configuration diagram of the image decompression apparatus. A quantized coefficient 124 is output from the input encoded stream 123 by entropy decoding means. Here, as means used in the entropy decoding unit 63, there are arithmetic decoding means in addition to variable length decoding means, as in the case of the entropy encoding means.

The quantization coefficient 124 is subjected to an inverse quantization operation in the inverse quantization unit 64 (see Expression 8), and a wavelet transform coefficient 125 is output.
x = Q × Δ (12)
Here, x is a wavelet transform coefficient value, Δ is a quantization index value, and Q is a quantization coefficient.

  The wavelet transform coefficient 125 is subjected to wavelet inverse transform for a predetermined number of levels in the wavelet inverse transform unit 65, and a final image 126 is output.

  In the analysis bank filter device shown in FIG. 6 and the synthesis filter device shown in FIG. 9, the lifting stage M is set to 2. This is related to the downsampler resolution magnification (1 / M = 1/2) and the upsampler resolution magnification (M = 2).

  FIG. 13 illustrates a configuration of an analysis filter bank apparatus that generalizes the lifting stage M. The input signal is divided into different M sets. This division is performed by delay units 72 and 74. In each stage, down samplers 71, 73, and 75 are provided. The configuration related to the first set and the second set is the same as the configuration shown in FIG. The prediction processing system includes an extension unit 76, a prediction filter 77, and a truncation unit 78. The truncation output from the truncation unit 78 is supplied to the subtractor 79. The update processing system of the second set and the first set includes an extension unit 84, an update filter 85, and a truncation unit 86. The truncation output from the truncation unit 86 is supplied to the adder 87.

  Similarly, the prediction processing system between the first set and the Mth set includes an expansion unit 80, a prediction filter 81, and a truncation unit 82. The truncation output from the truncation unit 82 is supplied to the subtractor 83. The update processing system for the Mth set and the first set includes an extension unit 88, an update filter 89, and a truncation unit 90. The output of the truncation unit 90 is supplied to the adder 91. The same applies to the prediction processing system between the second set and the third set, and the update processing system.

  The present invention is not limited to the hardware configuration shown in FIGS. 6, 9 and 13, but the software program of the processing procedure shown in FIGS. It can also be realized by executing the program on a computer connected to a ROM, a RAM, and a communication I / F. In that case, the software program is stored in an HDD, a ROM, or a removable storage medium.

It is a figure explaining a one-dimensional analysis filter bank (wavelet transform). It is a figure explaining the one-dimensional synthetic | combination filter bank (wavelet inverse transformation). It is a figure which shows the band division of a two-dimensional image (division level = 2). It is a block diagram of the analysis filter by a lifting means. It is a block diagram of the synthetic | combination filter by a lifting means. It is a block diagram of the analysis filter by the lifting means by this invention. It is a flowchart which shows the process sequence of an analysis filter. It is a figure explaining the filtering by a symmetrical expansion means. It is a block diagram of the synthetic | combination filter by the lifting means by this invention. It is a flowchart which shows the process sequence of a synthetic | combination filter. It is a block diagram of an image compression apparatus. It is a block diagram of an image expansion device. It is a block diagram of an analysis filter when the stage of lifting is M.

Explanation of symbols

36 downsampler, 37 delay unit, 38 downsampler, 39 extension unit, 40 prediction filter, 41 truncation unit, 42 subtractor, 43 adder, 44 truncation unit, 45 update unit, 46 extension unit

Claims (19)

An analysis filter bank having a lifting structure that divides an input signal into different sets of signals and adds or subtracts a filtering signal to or from the other side from both the first set and the second set obtained by the division A device,
A prediction processing system for performing a prediction process for predicting the second set using the first set is as follows.
First extension means for extending a down-sampled signal obtained by down-sampling the first set; and prediction filter means for performing processing for predicting the second set using the signal extended by the first extension means;
First truncation means for truncating the extension signal included in the filtered signal output by the prediction filter means;
An update processing system for performing an update process for updating the first set using a processing result from the prediction processing system,
Second expansion means for extending a difference signal between a down-sampled signal obtained by down-sampling the second set and a truncation output from the first truncation means of the prediction processing system;
Update filter means that is provided in the update processing system so as to form a pair with the prediction filter means, and performs update processing on the signal extended by the second extension means;
Comprising a second truncation means for truncating the extended signal included in the filter processing signal output by the update filter means;
An analysis filter bank apparatus characterized in that an addition signal obtained by adding the rounded down output from the second rounding down means to the downsampled first set and the differential signal as an analysis filter output signal.   2. The analysis filter bank device according to claim 1, wherein the analysis filter bank device is used as wavelet transform means for performing filtering on the image signal in the horizontal and vertical directions and frequency-resolving the image signal into a plurality of subbands.   2. The analysis filter bank apparatus according to claim 1, wherein the analysis filter bank apparatus is used as wavelet transform means for hierarchically dividing low-frequency components.   2. The analysis filter bank apparatus according to claim 1, wherein filtering is performed on a portion where the signal becomes discontinuous at the boundary portion of the image or subband.   The first extension means and the second extension means are within the boundaries of the image or subband by the number of coefficients that are insufficient when performing analysis filtering on the downsampled first set and the difference signal. 2. The analysis filter bank apparatus according to claim 1, wherein the coefficients are actually symmetrically expanded.   The first and second truncating means remove coefficients outside the boundary portion of the image or subband from the coefficients generated after performing analysis filtering. Analysis filter bank device. An analysis filter bank having a lifting structure that divides an input signal into different sets of signals and adds or subtracts a filtering signal to or from the other side from both the first set and the second set obtained by the division An analysis filter bank method for an apparatus, comprising:
A prediction processing system for performing a prediction process for predicting the second set using the first set is as follows.
A first extension step of extending a down-sampled signal obtained by down-sampling the first set; a prediction filter step of performing a process of predicting the second set using the signal extended by the first extension step;
A first truncation step of truncating the extension signal included in the filtered signal output by the prediction filter step;
An update processing system for performing an update process for updating the first set using a processing result from the prediction processing system,
A second extension step of extending a difference signal between a down-sampled signal obtained by down-sampling the second set and a truncation output from the first truncation step of the prediction processing system;
An update filter step for updating the signal provided in the update processing system to be paired with the prediction filter means and extended by the second extension step;
A second truncation step of truncating the extension signal included in the filtered signal output by the update filter step;
An analysis filter bank method characterized in that an addition signal obtained by adding a truncated output from the second truncation step to the first sample that has been down-sampled and the difference signal are used as an analysis filter output signal. An analysis filter bank having a lifting structure that divides an input signal into different sets of signals and adds or subtracts a filtering signal to or from the other side from both the first set and the second set obtained by the division An analysis filter bank program executed by the device,
A prediction processing system for performing a prediction process for predicting the second set using the first set is as follows.
A first extension step of extending a down-sampled signal obtained by down-sampling the first set; a prediction filter step of performing a process of predicting the second set using the signal extended by the first extension step;
A first truncation step of truncating the extension signal included in the filtered signal output by the prediction filter step;
An update processing system for performing an update process for updating the first set using a processing result from the prediction processing system,
A second extension step of extending a difference signal between a down-sampled signal obtained by down-sampling the second set and a truncation output from the first truncation step of the prediction processing system;
An update filter step for updating the signal provided in the update processing system to be paired with the prediction filter means and extended by the second extension step;
A second truncation step of truncating the extension signal included in the filtered signal output by the update filter step;
An analysis filter bank program characterized in that an addition signal obtained by adding a truncated output from the second truncation step to the first sample that has been down-sampled and the difference signal are used as an analysis filter output signal. A synthesis filter bank apparatus having a lifting structure for adding or subtracting a filtering process signal to or from a counterpart side from both a first signal set and a second signal set,
An update processing system for performing an update process for updating the first set using the second set is as follows.
First extension means for extending the second set;
Update filter means for performing update processing on the signal extended by the first extension means;
First truncation means for truncating the extension signal included in the filtered signal output by the update filter means;
The prediction processing system for predicting the second set using the processing result of the update processing system is:
Second extension means for extending a difference signal between the first set and the truncation output from the first truncation means of the update processing system;
Prediction filter means provided in the prediction processing system to form a pair with the update filter means, and performing prediction processing on the signal extended by the second extension means;
Comprising a second truncation means for truncating the extension signal included in the filtered signal output by the prediction filter means;
The shift signal of the signal obtained by up-sampling the sum signal obtained by adding the truncation output from the second truncation means to the second set and the signal obtained by up-sampling the difference signal are combined to obtain a combined filter output signal. A synthesis filter bank device characterized by that.   10. The synthesis filter bank apparatus according to claim 9, wherein the combined filter bank apparatus is used as wavelet inverse transform means for performing image filtering on the frequency-resolved coefficients in the horizontal and vertical directions.   10. The synthesis filter bank apparatus according to claim 9, wherein filtering is performed on a portion where the signal becomes discontinuous at the boundary of the subband.   The first extension means and the second extension means generate symmetrically extended coefficients existing in a subband boundary by the number of coefficients that are insufficient when performing synthesis filtering. Item 10. The synthesis filter bank device according to Item 9.   10. The first truncation means and the second truncation means remove coefficients outside the boundary portion of the image or subband from coefficients generated after performing synthesis filtering. Synthetic filter bank device. A synthesis filter bank method relating to a synthesis filter bank apparatus having a lifting structure for adding or subtracting a filtering processing signal to or from a counterpart side from both a first signal set and a second signal set,
An update process step for performing an update process for updating the first set using the second set is as follows:
A first extension step of extending the second set, an update filter step of performing an update process on the signal extended by the first extension step;
A first truncation step of truncating the extension signal included in the filtered signal output by the update filter step;
The prediction processing step of predicting the second set using the processing result of the update processing system includes:
A second expansion step of extending a differential signal between the first set and the truncation output from the first truncation means of the update processing step;
A prediction filter step for performing a prediction process on the signal provided in the prediction processing step to be paired with the update filter step and expanded by the second expansion step;
A second truncation step of truncating the signal for extension included in the filtered signal output by the prediction filter step;
The shift signal of the signal obtained by up-sampling the sum signal obtained by adding the truncation output from the second truncation step to the second set is synthesized with the signal obtained by up-sampling the difference signal to obtain a synthesized filter output signal. And a synthesis filter bank method. A synthesis filter bank program executed by a synthesis filter bank apparatus having a lifting structure for adding or subtracting a filtering process signal to or from a partner side from both a first signal set and a second signal set,
An update process step for performing an update process for updating the first set using the second set is as follows:
A first extension step of extending the second set, an update filter step of performing an update process on the signal extended by the first extension step;
A first truncation step of truncating the extension signal included in the filtered signal output by the update filter step;
The prediction processing step of predicting the second set using the processing result of the update processing system includes:
A second expansion step of extending a differential signal between the first set and the truncation output from the first truncation means of the update processing step;
A prediction filter step for performing a prediction process on the signal provided in the prediction processing step to be paired with the update filter step and expanded by the second expansion step;
A second truncation step of truncating the signal for extension included in the filtered signal output by the prediction filter step;
The shift signal of the signal obtained by up-sampling the sum signal obtained by adding the truncation output from the second truncation step to the second set is synthesized with the signal obtained by up-sampling the difference signal to obtain a synthesized filter output signal. A synthesis filter bank program characterized by that. An analysis filter bank having a lifting structure that divides an input signal into different sets of signals and adds or subtracts a filtering signal to or from the other side from both the first set and the second set obtained by the division A prediction processing system for performing a prediction process for predicting a second set using the first set is a first extension means for extending a down-sampled signal obtained by down-sampling the first set. Prediction filter means for performing processing for predicting the second set using the signal extended by the first extension means, and an extended signal included in the filter processing signal output by the prediction filter means Update processing for performing update processing for updating the first set using a processing result from the prediction processing system. A second extension means for extending a difference signal between a down-sampled signal obtained by down-sampling the second set and a truncated output from the first truncation means of the prediction processing system; and the prediction filter means; An update filter means provided in the update processing system so as to form a pair and performing an update process on the signal extended by the second extension means, and an extension portion included in the filter processing signal output by the update filter means Second truncating means for truncating the signal, and adding the summed signal obtained by adding the truncation output from the second truncating means to the downsampled first set, and the difference signal as an analysis filter output signal Analysis filter bank means for filtering the image signal in the horizontal and vertical directions, Used as a frequency decomposing wavelet transform unit to command, the frequency decomposing frequency resolution means the image signals,
Quantization means for performing quantization processing on the subband frequency-resolved by the frequency resolution means to generate a quantization coefficient;
An image compression apparatus comprising: encoding means for performing entropy encoding processing on the quantized coefficient generated by the quantization means. An analysis filter bank having a lifting structure that divides an input signal into different sets of signals and adds or subtracts a filtering signal to or from the other side from both the first set and the second set obtained by the division An analysis filter bank process relating to an apparatus, wherein a prediction processing step for performing a prediction process for predicting a second set using the first set extends a down-sampled signal obtained by down-sampling the first set. Included in a first extension step, a prediction filter step that performs a process of predicting the second set using the signal extended by the first extension step, and a filter processing signal output by the prediction filter step A first truncation step of truncating the extended signal, and updating the first set using a processing result from the prediction processing system. An update processing step for performing an update process for the second extension for extending a differential signal between a downsampling signal obtained by downsampling the second set and a truncation output from the first truncation step of the prediction processing step. And an update filter step for updating the signal provided in the update processing system so as to form a pair with the prediction filter means and extending by the second extension step, and a filter output by the update filter step A second truncation step of truncating the extension signal included in the processed signal, and adding the truncation output from the second truncation step to the down-sampled first set and the difference Analyzing filter bank process using signal as analysis filter output signal, filtering horizontally and vertically with respect to image signal Performed, an image signal by using a plurality of sub-band as a frequency decomposing wavelet transform process, the frequency decomposing frequency decomposition step an image signal,
A quantization step of generating a quantization coefficient by performing a quantization process on the subbands frequency-resolved by the frequency decomposition step;
An image compression method comprising: an encoding step of performing entropy encoding processing on the quantization coefficient generated by the quantization step. Decoding means for performing entropy decoding processing on the compressed image data;
Inverse quantization means for inversely quantizing coefficient data obtained by entropy decoding processing by the decoding means;
Filtering from both the first set of signals and the second set of signals to the other side used as wavelet inverse transform means for performing wavelet inverse transform processing on the coefficients of each frequency component from the inverse quantization means. A synthesis filter bank apparatus having a lifting structure for adding or subtracting a processing signal, wherein an update processing system for performing an update process for updating the first set using the second set is the second set A first extension means for extending the set, an update filter means for performing an update process on the signal extended by the first extension means, and a signal for the extension included in the filter processing signal output by the update filter means A prediction processing system that includes a first truncating unit that truncates the second set using a processing result of the update processing system, And a second extension means for extending a difference signal between the first and second truncation outputs of the update processing system and the update filter means to be paired with the update filter means. A prediction filter unit that performs a prediction process on the signal expanded by the second expansion unit; and a second truncation unit that truncates a signal for expansion included in the filter processing signal output by the prediction filter unit. Combining a shift signal of a signal obtained by up-sampling the sum signal obtained by adding the truncation output from the second truncation means to the second set and a signal obtained by up-sampling the difference signal to form a synthesis filter output signal Filter bank means;
An image expansion apparatus comprising: A decoding step of performing entropy decoding processing on the compressed image data;
An inverse quantization step of inversely quantizing coefficient data obtained by the entropy decoding process by the decoding step;
Filtering from the first set of signals and the second set of signals to the other side, used as a wavelet inverse transform process for performing wavelet inverse transform processing on the coefficients of each frequency component from the inverse quantization process. A synthesis filter bank process related to a synthesis filter bank apparatus having a lifting structure for adding or subtracting a processing signal, wherein an update processing step for performing an update process for updating the first set using the second set is provided. A first extension step for extending the second set, an update filter step for performing an update process on the signal extended by the first extension step, and a filter processing signal output by the update filter step. A first truncation step for truncating the extended signal, and predicting the second set using the processing result of the update processing system The prediction processing step is paired with the update filter step and the second extension step of extending the difference signal between the first set and the truncation output from the first truncation means of the update processing step. A prediction filter step that is provided in the prediction processing step and performs a prediction process on the signal expanded by the second expansion step, and an extension signal included in the filter processing signal output by the prediction filter step is truncated. 2 truncation steps, a shift signal of a signal obtained by up-sampling the sum signal obtained by adding the truncation output from the second truncation step to the second set, and a signal obtained by up-sampling the difference signal An image decompression method comprising: a synthesis filter bank step that synthesizes and outputs a synthesis filter output signal.

Images