JP2002204169A - Subband encoder, subband decoder, wavelet transformer, reverse wavelet transformer, compressor and expander - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable a parallel processing at high speed by preventing access collision while suppressing an increase in storage capacity for a buffer memory. SOLUTION: A subband encoder divides an input signal into areas and divides band in a parallel processing by band-dividing filter banks 104-0 through 104-3 corresponding to the areas. In an input buffer memory section, signals used for processing of the corresponded areas are written in memory devices 102-0, 102-2, 102-4 and 102-6 corresponding to the areas, and signals used in two areas adjacent to the vicinity of an area-dividing boundary are written in storage devices 102-1, 102-3 and 102-5 corresponding to the other area-dividing boundary.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to compression and decompression of image signals and the like, and more particularly to subband coding and wavelet transform by parallel processing.

[0002]

2. Description of the Related Art There are a large number of conventional techniques relating to compression / expansion of image signals. For example, Japanese Unexamined Patent Publication No. Hei 8-139935 discloses a wavelet transform coding device and a TS transform filter, Japanese Unexamined Patent Publication No. Hei 9-146926 discloses a technique for increasing the speed of discrete wavelet transform by parallelizing the processing, Japanese Patent Publication No. -308055 discloses efficient parallel processing techniques in a subband encoding / decoding device and a forward / inverse wavelet transform device.

In order to speed up the wavelet transform,
JP-A-9-146926 and JP-A-2000-308
As described in Japanese Patent Publication No. 055, parallel processing is generally effective. For parallel processing, the input signal is divided into a plurality of regions, and band division of the input signal is executed in parallel by a plurality of band division filter banks corresponding to the regions. For that purpose, it is necessary to buffer-store the signals of each divided area in a plurality of storage devices that can be accessed independently. However, when a filter having three or more taps is used as a low-pass filter (LPF) or a high-pass filter (HPF) of the band division filter bank, signals in adjacent regions are also required for processing near the region division boundary of each region. Simply storing the signal of the storage device in the storage device corresponding to the area causes access collision because each storage device is accessed from a plurality of band division filter banks, and the effect of the parallel processing can be sufficiently achieved by increasing the access waiting time. I can't show it. Also, access control of the storage device becomes complicated.

For example, a TS conversion (2-tap LPF / 6-tap) disclosed in Japanese Patent Application Laid-Open No. H8-139935 discloses that a one-dimensional signal consisting of 64 signals is divided into four regions of 16 signals and processed in parallel. The case of using (HPF) will be described with reference to FIG. In FIG. 16, 60
0_0 is a storage device for storing the signal of the first area, 600_1
Is a storage device for storing signals in the second area. 60
Reference numeral 2 denotes a 6-tap HPF for band division of the second area. When generating the first high-pass coefficient H1 (0) of the second area, the 6-tap HPF outputs the first four signals X1 (0), X1 (1), X1 (2), X1 (3) of the same area. And the last two signals X0 (14), X0 in the first area
(15) is required. That is, at this time, the storage device 6
At 00_0, the access for processing the first area and the access for processing the second area collide, and the processing for one area waits for access.

To avoid such access collision,
The technique of JP-A-2000-308055 is effective. According to this conventional technique, when a one-dimensional signal composed of 64 signals is divided into four regions of 16 signals and processed in parallel,
A signal as shown in FIG. 17 is written in the storage device corresponding to the area. Here, TS conversion (2-tap LPF / 6-tap HPF) is also assumed. In FIG. 17, in addition to signals 02, 12, 22, and 32 of the corresponding areas, signals 03, 11, 13, and 13 of adjacent areas necessary for processing of the respective areas are stored in four storage devices 700_0 to 700_3 corresponding to the areas. 21 and 2
3, 31 are also written. Therefore, it is possible to avoid collision of access of the storage device in the processing near the area division boundary, and it is not necessary to perform exceptional access control near the area division boundary. Further, since the mirror image inversion signals 01 and 33 at the beginning and end of the one-dimensional signal are written in the storage devices 700_0 and 700_3 of the first and second areas, respectively, It is not necessary to perform a mirror image reversal process.

According to this conventional technique, 16 × 16
A two-dimensional signal (e.g., image signal) consisting of
When the signal is divided into four regions each composed of eight signals and wavelet transform is performed by parallel processing using TS conversion, a signal as shown in FIG. 18 is stored in a buffer storage device corresponding to a region for horizontal processing, for example. Written. That is, FIG.
8, reference numerals 800_0 to 800_3 denote storage devices corresponding to areas for horizontal processing, which include signals 41,
In addition to 44, 47, and 50, signals 42, 43, 48, and 49 of adjacent areas required for processing near the area division boundary and mirror image inversion signals 40, 45, 46, and 51 are also written.

[0007]

Problems to be Solved by the Invention
According to the technology of No. 055, it is possible to avoid the stagnation of the process due to the waiting for the access of the buffer storage device, and to simplify the access control and eliminate the exception process. However,
As can be easily understood from FIGS. 17 and 18, since the signals near the region division boundary are stored twice, there is a problem that the storage capacity required for buffer storage increases. FIG.
In the example, the original number of signals is 64, but a storage capacity is required to store 80 signals (76 even excluding the mirror image inversion signal). As can be seen from the example of FIG. 18, when a two-dimensional signal is handled, the storage capacity is further increased. Therefore, an object of the present invention is to suppress an increase in the storage capacity required for buffer storage of a signal. Another object of the present invention is to provide an improved sub-band encoding device, sub-band decoding device, wavelet transform device, inverse wavelet transform device, compression device, and decompression device that can maximize the effect of parallelization.

[0008]

A subband encoding apparatus according to the present invention divides an input signal into a plurality of regions, and a plurality of band division filter banks corresponding to each region in a one-to-one correspondence. In order to buffer the input signal, the input signal has a storage device corresponding to each region of the input signal, and a storage device corresponding to each region division boundary of the input signal. A signal used only for processing of the corresponding area is written in the storage device corresponding to
A signal used for processing two adjacent areas near the corresponding area division boundary is written in the storage device corresponding to each area division boundary. According to such a configuration, while reducing the storage capacity required for buffer storage,
High-speed parallel processing can be realized by avoiding processing delay due to collision of buffer storage access.

A second feature of the present invention is that, in the subband encoding apparatus according to the first aspect of the present invention, the mirror image inversion signal is also written in the storage device corresponding to the first area and the last area of the input signal. is there. According to such a configuration, since the exceptional mirror image inversion processing is not required, the processing is simplified and the processing can be performed at higher speed.

According to a third aspect of the present invention, in the subband encoding apparatus according to the first aspect of the present invention, the storage devices corresponding to the first area and the last area of the input signal are each composed of two independent storage devices. The mirror image inversion signal and its underlying signal are written in one of the storage devices, and the mirror image inversion signal and its underlying signal, which are used only for processing the corresponding area, are written in the other storage device. Is written. According to such a configuration, the storage device in which the mirror image inversion signal and the base signal are written can be handled in the same manner as the storage device corresponding to the area division boundary, so that access control and processing can be simplified.

According to a fourth aspect of the present invention, in the subband encoding apparatus of the first, second or third aspect of the present invention, a specific signal in a signal written to a storage device corresponding to a region division boundary is a region-specific signal. That is, the data is also redundantly written to the storage device. According to such a configuration, even when the number of taps of the band division filter is large, it is possible to avoid collision of buffer storage access and delay of processing due to access waiting.

[0012] The sub-band decoding apparatus according to claim 5 is
The input signal is divided into a plurality of regions, and the input signals are band-synthesized in parallel by a plurality of band-synthesizing filter banks corresponding to each region in a one-to-one manner. A storage device corresponding to each region, and a storage device corresponding to each region division boundary of the input signal,
In the storage device corresponding to each area, a signal used only for processing of the corresponding area is written, and in the storage device corresponding to each area division boundary, processing of two adjacent areas near the corresponding area division boundary is performed. A signal used for writing is written. According to such a configuration, high-speed parallel processing can be realized while reducing the storage capacity required for buffer storage and avoiding processing delay due to collision of buffer storage access.

According to a sixth aspect of the present invention, in the subband decoding apparatus of the fifth aspect of the present invention, the mirror image inversion signal is also written in the storage device corresponding to the first area and the last area of the input signal. . According to such a configuration, since the exceptional mirror image inversion processing is not required, the processing is simplified and the processing can be performed at higher speed.

According to a seventh aspect of the present invention, in the subband encoding apparatus according to the fifth aspect of the present invention, the storage devices corresponding to the first area and the last area of the input signal comprise two independent storage devices. The mirror image inversion signal and its underlying signal are written in one of the storage devices, and the mirror image inversion signal and its underlying signal, which are used only for processing the corresponding area, are written in the other storage device. Is written. According to such a configuration, the storage device in which the mirror image inversion signal and the base signal are written can be handled in the same manner as the storage device corresponding to the area division boundary, so that access control and processing can be simplified.

According to an eighth aspect of the present invention, in the subband decoding apparatus according to the fifth, sixth or seventh aspect of the present invention, a specific signal in a signal written to the storage device corresponding to the region boundary is stored in the storage device corresponding to the region. Is also duplicated. According to such a configuration, even when the number of taps of the band division filter is large, it is possible to avoid collision of buffer storage access and delay of processing due to access waiting.

According to a ninth aspect of the present invention, there is provided a wavelet transform apparatus for performing horizontal processing and vertical processing.
It is characterized by including two or more of the subband encoding devices of the invention of 2, 3 or 4. According to a tenth aspect of the present invention, there is provided an inverse wavelet transform device including two or more sub-band decoding devices according to the fifth, sixth, seventh, or eighth aspects for horizontal processing and vertical processing. Such a wavelet transform device or inverse wavelet transform device can perform horizontal direction processing and vertical direction processing by high-speed parallel processing, and thus can perform high-speed wavelet transform or inverse wavelet transform.

The compression device according to the eleventh aspect of the present invention is the ninth aspect of the present invention.
The wavelet transform device and the coding device according to the present invention are characterized in that the two-dimensional signal is wavelet-transformed by the wavelet transform device, and the obtained wavelet transform coefficients are coded by the coding device. A decompression device according to a thirteenth aspect includes the inverse wavelet transform device and the decoding device according to the tenth aspect, wherein the decoding device decodes a compression-coded signal of a two-dimensional signal by the decoding device, and obtains the obtained wavelet transform coefficient. The signal is inversely wavelet transformed by the inverse wavelet transformer to restore a two-dimensional signal. According to such a compression device or decompression device, high-speed compression or decompression of an image signal or the like can be performed by high-speed processing by performing wavelet transform or inverse wavelet transform at high speed.

According to a twelfth aspect of the present invention, there is provided a compression apparatus according to the ninth aspect.
A plurality of regions obtained by dividing a two-dimensional signal into a plurality of regions, performing a wavelet transform in parallel by the plurality of wavelet transform devices, and obtaining a plurality of regions. Are encoded in parallel by the plurality of encoding devices. A decompression device according to a fourteenth aspect of the present invention includes a plurality of inverse wavelet transform devices and a plurality of decoding devices according to the tenth aspect of the present invention. And a plurality of obtained wavelet transform coefficient signals in a plurality of regions are subjected to inverse wavelet transform in parallel by the plurality of inverse wavelet transform devices to restore a two-dimensional signal. According to such a configuration, it is possible to compress or decompress an image signal or the like at a higher speed due to the effect of parallelizing the wavelet transform and the encoding or the decoding and the inverse wavelet transform.

[0019]

Embodiments of the present invention will be described below with reference to the accompanying drawings. Note that, in order to reduce repetition of the description, similar reference numerals may be used for equivalent parts or corresponding parts in a plurality of drawings.

Embodiment 1 FIG. 1 is a schematic block diagram showing an example of a subband encoding apparatus according to an embodiment of the present invention. In FIG. 1, reference numeral 100 denotes a source of a one-dimensional input signal, for example, a storage device. This subband encoding device divides an input signal into four regions, and performs band division by parallel processing in four band division filter banks 104_0 to 104_3 associated with each region on a one-to-one basis. In order to efficiently perform such four-parallel processing, the input buffer storage unit 102 for taking in an input signal and temporarily storing the input signal includes a plurality of storage devices that can be accessed independently. The storage device constituting the input buffer storage unit 102 and its storage contents will be described later. Band division filter banks 104_0-10
4_4, the low-pass coefficient signal and the high-pass coefficient signal of each area are temporarily stored in the output buffer storage unit 106, and then a set of low-pass coefficient signals and a set of high-pass coefficients It is output to the outside in the form of a signal. However, the output buffer storage unit 106 can be omitted.

A band division filter bank is generally
It comprises a band division filter that divides an input signal into a plurality of band signals, and a downsampler that performs thinning (downsampling) on each band signal according to the bandwidth. In the sub-band encoding apparatus shown here, since the sub-band encoding is performed in two parts, the band division filter banks 104_0 to 104_104
_3 is a low-pass filter LPF as a band division filter
And a pair of high-pass filters HPF, and two 2-to-1 downsamplers (indicated by the symbol "↓ 2" in the figure)
Thus, a two-to-one downsampling of each band signal is performed.

According to one embodiment (1-1), as shown in FIG. 2, the input buffer storage unit 102 has seven independently accessible storage units 102_0 to 102_6. The input signal is composed of 64 signals, and signals X0 (0) to X0 (15) in the first area and signals X1 (0) to X1 (1) in the second area.
5), the signals are divided into signals X2 (0) to X2 (15) in the third region and signals X3 (0) to X3 (15) in the fourth region. Band division filter bank 104
_0 to 104_3 are executed in parallel. In this embodiment, the TS conversion described in Japanese Patent Application Laid-Open No. 8-139935 is adopted, and the low-pass filter LP used in the band division filter banks 104_0 to 104_3.
The number of taps of F is two, and the number of taps of the high-pass filter HPF is six.

Of the seven storage devices, the storage device 102
_0, 102_2, 102_4, and 102_6 are storage devices corresponding to the band division filter banks 104_0 to 104_3 on a one-to-one basis, into which signals used only for processing of the corresponding areas are written. Other storage devices 102_1, 10
Reference numerals 2_3 and 102_5 denote storage devices corresponding to the region division boundaries, into which signals used for processing of two regions before and after the corresponding region division boundary are written. Depends on the number. However, since one signal is written into only one storage device, in this embodiment 64 signals are distributed and stored in seven storage devices as shown in FIG. Therefore, the storage capacity required for buffer storage is reduced to a capacity corresponding to the number of input signals.

Further, it is possible to avoid collision of accesses to the storage devices in the input buffer storage unit 102. For example, when generating the first high-pass coefficient of the second area, the signals X1 (0) to X1 (3) of the second area and the signals of the first area are supplied to the high-pass filter HPF of the band division filter bank 104_1.
X0 (14) and X0 (15) must be entered. These signals are read from the storage device 102_1 or 102_2, but at this time, the band division filter bank 104 in the first area is used.
_0 executes processing of a signal near the beginning of the first area, and therefore does not access the storage device 102_1. Similarly, the band division filter bank 104_1 stores the storage devices 102_2, 102_ when processing the latter half signal of the second area.
3 is accessed but the storage device 102_1 is not accessed, so that access from the band division filter banks 104_0 and 104_1 does not collide with the storage device 102_1. As described above, since the access collision of the storage devices 102_0 to 102_3 does not occur, it is possible to avoid a delay in processing due to an access wait and to simplify access control.

According to another embodiment (1-2), as shown in FIG.
The configuration includes seven storage devices 102_0 to 102_6 similar to (1-1). As in the previous embodiment (1-1), the one-dimensional input signal composed of 64 signals is divided into four regions of 16 signals each and processed in parallel by the band division filter banks 104_0 to 104_3. Storage devices 102_0, 102_2,
102_4, 102_6 are band division filter banks 104_0
１０４104_3 are storage devices corresponding one-to-one, in which signals used only for processing of corresponding areas are written. The other storage devices 102_1, 102_3, and 102_5 are storage devices corresponding to the area division boundary, and signals in the vicinity of the area division boundary used for processing of the two preceding and succeeding areas are written therein. In this embodiment, similarly to the embodiment (1-1), TS
A conversion (2-tap LPF / 6-tap HPF) is used.

As is clear from a comparison between FIG. 3 and FIG.
In this embodiment, in the input buffer storage unit 102,
The mirror image inversion signals X0 (1), X0 of the first two signals of the one-dimensional input signal (the first two signals of the corresponding area) are stored in the storage device 102_0.
(0) is also written. Similarly, in the storage device 102_6,
Mirror image inversion signals X3 (15) and X3 (14) of the last two signals of the one-dimensional input signal (the last two signals of the corresponding area) are also written. Therefore, the storage capacity required for buffer storage is increased by four signals of the mirror image inversion signal from the embodiment (1-1).
The storage capacity is sufficiently reduced as compared with the storage capacity required in the technique of JP-A-2000-308055.

According to this embodiment, it is apparent that collision of access to the storage device in the input buffer storage unit 102 can be avoided as in the case of the embodiment (1-1). Further, according to this embodiment, the mirror image inverted signals at the beginning and end of the one-dimensional input signal are written in the storage devices 102_0 and 102_6 in advance, so that an exceptional mirror image is generated at the time of band division at the beginning and end of the one-dimensional input signal. Since the necessity of performing the inversion processing is eliminated, higher-speed processing can be performed.

Another embodiment (1-3) will be described with reference to FIG. Also in this embodiment, the embodiment (1-1),
Similarly to (1-2), the one-dimensional input signal composed of 64 signals is divided into four regions of 16 signals each and processed in parallel by the band division filter banks 104_0 to 104_3. Also,
Also in this embodiment, TS conversion (2-tap LPF / 6-tap H
PF) is used.

According to this embodiment, as shown in FIG. 4, the input buffer storage unit 102 has nine independently accessible storage devices 102_0 to 102_8. As is clear from the comparison between FIG. 4 and FIG. 3, the difference from the embodiment (1-2) is that the storage device for storing the signal used only for the processing of the first area has two storage devices. 10
2_0, 102_1, and the first signal X0 of the first area
(0), X0 (1) and its mirror image inverted signals X0 (1), X0 (1) are stored in the storage device 1
02_1 is not written to the storage device 102_0. Another difference is that a storage device for storing signals used only for processing of the fourth area is divided into two storage apparatuses 102_7 and 102_8, and signals X (14) and X ( 15) and its mirror image inverted signals X (15), X (14)
Is not written to the storage device 102_7,
_8. Other storage device 10
The signals written to 2_2 to 102_6 are the same as those in the above embodiment (1-
This is the same as the signal written to the storage devices 102_1 to 102_5 (FIG. 3) in 2). Therefore, the storage capacity of the input buffer storage unit 102 is not different from that of the embodiment (1-2).

According to this embodiment, the same high-speed parallel processing as in the embodiment (1-2) is possible, but furthermore, the head or tail signal of the one-dimensional input signal and its mirror image inverted signal are stored. Storage device 102_0 or 102_8 can be handled in the same manner as the storage devices 102_2, 102_4, and 102_6 corresponding to the area division boundary, and thus there is an advantage that access control and processing can be simplified.

Another embodiment (1-4) will be described with reference to FIGS. In this embodiment, a one-dimensional input signal composed of 48 signals is divided into four regions each of 12 signals, and is processed in parallel by band division filter banks 104_0 to 104_3. In this embodiment, the high-pass filters HP of the band division filter banks 104_0 to 104_3 are used.
The number of taps of F is set to 10 taps.

According to this embodiment, the input buffer storage unit 102 has nine storage devices 102_0 to 102_1 as shown in FIG.
02_8, and a signal as shown is written in each storage device.

Under such a condition that the number of taps of the band division filter is large, according to the embodiment (1-3), a signal as shown in FIG. 6 is written in each storage device in the input buffer storage unit 102. Will be. In this case, access collision occurs for a specific signal.

In order to avoid such access collision, according to this embodiment, the input buffer storage unit 102
Is, as shown in FIG.
_0 to 104_3 in one-to-one correspondence with the storage device 102
_1, 102_3, 102_5, and 102_7 are configured such that specific signals that cause access collision among signals used for processing of adjacent areas are redundantly written. For example, the storage device 10 corresponding to the band division filter bank 104_1
In 2_3, in addition to the signal used only for the processing of the second area, a signal causing an access collision, that is, signals X1 (2) and X1 (3) written to the storage device 102_2, and a signal written to the storage device 102_4 The signals X1 (8) and X1 (9) to be written are redundantly written. Accordingly, the storage capacity required for the input buffer storage unit 102 is increased by the number of signals to be written in duplicate, but stagnation of processing due to access collision can be avoided, and high-speed parallel processing can be performed. Although the mirror image inversion processing is required, a configuration in which the mirror image inversion signals of the first area and the fourth area are not written in the buffer storage is also possible.

<Embodiment 2> As another embodiment of the present invention, there is a sub-band decoding apparatus that combines bands of sub-band coded signals by parallel processing and restores the original signal. Since the sub-band decoding device performs exactly the opposite processing to the sub-band coding device, the following description will be made with reference to the drawings related to the sub-band coding device.

First, the overall configuration of the subband decoding device will be described with reference to FIG. The input signals are a low-pass coefficient signal and a high-pass coefficient signal of subband coding, and are input from the signal source 100 to the input buffer storage unit 102. The band division filter banks 104_0 to 104_3 are replaced with band synthesis filter banks. Each band synthesis filter bank (conveniently represented by 104_0 to 104_3) is composed of an upsampler for inserting a zero value at a signal position decimated on the encoding side and a band synthesis filter. When the conversion is used, the inverse TS conversion is used in the subband decoding device.
In this case, similarly to the encoding side, each band synthesis filter bank 104_0 to 104_3 needs to refer to a signal in an adjacent area near the area division boundary, and the band synthesis filter banks 104_0 and 104_3 output the mirror image inverted signal. Need. Band synthesis filter banks 104_0 to 104_3
The output signal for each area is output to an output buffer 106.
After being stored in the buffer, they are combined into the original one-dimensional signal and output to the outside.

According to the embodiment (2-1), the input buffer storage unit 102 has seven independently accessible storage devices 102_0 to 102_6 as shown in FIG. An input signal composed of 64 signals is divided into four regions of 16 signals each, and band synthesis is performed by parallel processing by band synthesis filter banks 104_0 to 104_3 corresponding to the regions.

Of the seven storage devices in the input buffer storage unit 102, the storage devices 102_0, 102_2, and 102_
4, 102_6 are band synthesis filter banks 104_0 to 104_10
4_3 is a storage device corresponding one-to-one, and a signal used only for processing of the corresponding area is written. The other storage devices 102_1, 102_3, and 102_5 are storage devices corresponding to an area division boundary, and a signal near the area division boundary used for processing of the front and rear two areas is written in this storage apparatus.
The number of signals is determined by the number of taps of the band synthesis filter. However, since one signal is written into only one storage device, the storage capacity required for buffer storage is sufficient for the number of input signals.

According to such a configuration, as described in connection with the embodiment (1-1) of the subband encoding device, it is possible to avoid collision of accesses to the storage device in the input buffer storage unit 102. , High-speed parallel processing is possible, and access control to a storage device in the input buffer storage unit 102 is also simplified.

According to another embodiment (2-2), the input buffer storage unit 102 has seven storage devices 102_0 to 102_6 similarly to the above embodiment (2-1). Are basically the same as those in the embodiment (2-1). However, the storage devices 102_0 and 102_6 corresponding to the first area and the fourth area have the head of the one-dimensional input signal,
The tail mirror inversion signal is also written. Therefore, the storage capacity required for buffer storage is increased by the mirror image inversion signal compared with the embodiment (2-1), but the exceptional mirror image inversion processing is performed at the time of combining the head and tail of the one-dimensional input signal. Since it is not necessary to perform the processing, higher-speed processing can be performed.

According to another embodiment (2-3), as shown in FIG. 4, the input buffer storage unit 102 has nine independently accessible storage units 102_0 to 102_8. You. The difference from the embodiment (2-2) is that
A storage device for storing a signal used only for processing the region is divided into two storage devices 102_0 and 102_1,
The mirror image inversion signal at the head of the first area and its original signal are not written to the storage device 102_1, but are stored in another storage device 102_1.
_0. Another difference is that a storage device for storing a signal used only for processing of the fourth region is divided into two storage devices 102_7 and 102_8, and a mirror image inversion signal at the end of the fourth region and its base. The signal is not written to the storage device 102_7, and another storage device 1
02_8. Other than the above, it is the same as the embodiment (2-2).

According to this embodiment, the same high-speed parallel processing as in the embodiment (2-2) is possible, but the head or tail signal of the one-dimensional input signal and its mirror image inverted signal are also stored. Since the storage device 102_0 or 102_8 to be processed can be handled in the same manner as the storage devices 102_2, 102_4, and 102_6 corresponding to the area division boundaries, access control and processing can be simplified.

Another embodiment (2-4) corresponds to the above-described embodiment (1-4) of the sub-band encoding apparatus, and
The one-dimensional input signal composed of eight signals is divided into four regions each of twelve, and the band synthesis filter banks 104_0 to 104_3 are divided.
To perform band synthesis for each area by parallel processing.

According to this embodiment, the input buffer storage section 102 has nine storage devices 102_0 to 102_0 to
102_8, and the band division filter banks 104_0 to 104_1.
Storage device 102_1, 1 associated with the storage device 04_3 on a one-to-one basis
Signals used only for processing of the corresponding areas are written in 02_3, 102_5, and 102_7. In addition, specific signals that cause access collision are written in duplicate. Therefore, the input buffer storage unit 10
Although the storage capacity required for 2 is increased by the number of signals written in duplicate, high-speed parallel processing becomes possible because stagnation of processing due to access collision can be avoided. It is also possible to adopt a configuration in which the mirror image inversion signal is not written in the buffer storage.

Embodiment 3 As another embodiment of the present invention, a wavelet transform apparatus which applies the above-described subband encoding apparatus of the present invention to horizontal processing and vertical processing of wavelet transform. explain.

FIG. 7 is a schematic block diagram showing an example of the wavelet transform device of the present invention. The input signal is, for example, a two-dimensional signal (for example, an image signal) composed of 16 × 16 signals as shown in FIG.
00 is stored in the buffer. This input signal is basically divided into four regions each consisting of 8 × 8 signals as shown in FIG. 8, and the signals in each region are horizontally processed by four band division filter banks 202_0 to 202_3 by parallel processing. Is divided into two subbands. That is, in the horizontal two-division sub-band encoding, 16 signals of each row in the upper half of the 16 × 16 input signals are divided into eight by two.
It is divided into two regions, and the sub-band coding of each region is performed in parallel by two band division filter banks 202_0 and 202_1. Similarly, the 16 signals in each row of the lower half of the 16 × 16 input signal are divided into two regions by eight, and the subband coding of each region is performed in parallel by the two band division filter banks 202_2 and 202_3. Done.
The input buffer storage unit 200 and the band division filter banks 202_0 to 202_3 for such horizontal processing.
Constitutes the subband encoding apparatus of the present invention as described above.

By such horizontal processing, FIG.
Are obtained and stored in the intermediate buffer storage unit 204. The low-pass coefficient signal Lxx and the high-pass coefficient signal Hxx shown in FIG. The 16 × 16 coefficient signals basically correspond to FIG.
As shown by 0, the signal is divided into four regions each composed of 8 × 8 signals, and the signals in each region are vertically divided into two sub-bands by parallel processing by four band division filter banks 206_0 to 206_3. That is, in this vertical processing, the 16 signals in each column of the left half of the 16 × 16 coefficient signals are divided into two regions by eight, and the subband encoding of each region is performed by two band divisions. The processing is performed in parallel by the filter banks 206_0 and 206_1. Similarly,
The 16 signals in each column of the right half of the 16 × 16 coefficient signals are divided into two regions of eight each, and the sub-band coding of each region is performed by two band division filter banks 206_2, 20
6_3 in parallel. Intermediate buffer storage unit 20
4 and the band division filter banks 206_0 to 206_3 constitute the subband encoding apparatus of the present invention as described above. By such vertical processing, FIG.
Thus, four sets of coefficient signals corresponding to the four regions are obtained. The coefficient signals are buffer-stored in the output buffer storage unit 208, and are then combined into a set of signals and output to the outside. According to one embodiment (3-1), as shown in FIG. 9, the input buffer storage unit 200 is configured to include ten independently accessible storage devices 200_0 to 200_9. In the storage devices 200_0 and 200_1, signals used only for processing of the first region of each row in the upper half of the input signal shown in FIG. 8 (in other words, only the band division filter bank 202_0 corresponding to the region is referred to) Is written. In this embodiment, the mirror image inversion signal at the head of each row is also written in the storage device 200_0. Storage device 20
0_3 and 200_4 are signals used only for processing of the second region of each row in the upper half of the input signal shown in FIG. 8 (in other words, the band division filter bank 202 corresponding to the region).
_1 is written. In this embodiment, the mirror image inversion signal at the end of each row is also written in the storage device 200_4. The storage device 200_2 stores signals near the region division boundary (in other words, of the band division filter banks 200_0 and 202_1) used for processing the first region and the second region of each row in the upper half of the input signal shown in FIG. The signal referred to by both is written. Storage devices 200_5 to 20
0_9 corresponds to the lower half of the signal shown in FIG. 8, and the storage devices 200_5, 200_6 and the storage devices 200_8, 200_9 have signals (only used for processing the first area or the second area in each row of the lower half). A signal which is referred to only the band division filter bank 202_2 or 200_3) is written. In this embodiment, the mirror image inversion signals at the beginning and end of each row in the lower half are also written in the storage devices 200_5 and 200_9. In the storage device 200_7, signals near the area division boundary used for processing the first area and the second area of each row in the lower half are written. The number of signals written to the storage devices 200_2 and 200_7 is determined by the number of taps of a filter used in the band division filter bank. Here, TS conversion (2-tap LPF / 6-tap HPF) is assumed.

Further, according to this embodiment, as shown in FIG. 11, the intermediate buffer storage unit 204 has ten independently accessible storage devices 204_0 to 204_9. In the storage devices 204_0 and 204_1, FIG.
A signal used only for processing of the first region of each column of the left half of the coefficient signal indicated by 0 (in other words, a signal referred to only the band division filter bank 206_0 corresponding to the region) is written. In this embodiment, the storage device 204
In _0, the mirror image inversion signal at the head of each column is also written. In the storage devices 204_3 and 204_4, signals used only for processing of the second region of each column of the left half of the coefficient signal shown in FIG. 10 (in other words, only the band division filter bank 206_1 corresponding to the region is referred to) Is written). In this embodiment, the mirror image inversion signal at the end of each column is also written in the storage device 204_4. The first region and the second region of each column in the left half of the signal shown in FIG.
Signals in the vicinity of the region division boundary (in other words, band division filter banks 206_0, 206_1) used for region processing.
) Are written. Storage device 2
04_5 to 204_9 correspond to the right half of the coefficient signal shown in FIG. 10, and the storage devices 204_5 and 204_6 and the storage devices 204_8 and 204_9 have only the first region or the second region of each column of the right half. Signals used (signals referenced only by band division filter banks 206_2 or 206_3)
Is written. In this embodiment, the storage devices 204_5,
In 204_9, the mirror image inversion signals at the beginning and end of each column in the right half are also written. In the storage device 204_7, signals near the region division boundary used for processing the first region and the second region of each column in the right half are written. Storage device 204_2,2
The number of signals to be written to 04_7 is determined by the number of taps of filters used in the band division filter banks 206_0 to 206_3.
/ 6 tap HPF).

Because of the configuration described above, the embodiment
As described in connection with (1-1) and (1-2),
The storage capacity required for the input buffer storage unit 200 and the intermediate buffer storage unit 204 can be reduced as compared with the case of the technique of No. 308055 (see FIG. 18). Further, the access waiting time of the storage device in the input buffer storage unit 200 and the storage device in the intermediate buffer storage unit 204 can be reduced. Further, since the mirror image inversion signal is written in the input buffer storage unit 200 and the intermediate buffer storage unit 204 in advance, the process for the exceptional mirror image inversion becomes unnecessary. Therefore, high-speed parallel processing can be performed in both the horizontal and vertical directions.

Although not shown, according to another embodiment (3-2), mirror image inversion processing is required, but in the above embodiment (1-2)
As in 1), the mirror image inversion signal is not written in the storage device inside the input buffer storage unit 200 and / or the intermediate buffer storage unit 204. Although not shown, according to another embodiment (3-3), similarly to the embodiment (1-4), only in each area inside the input buffer storage unit 200 and / or the intermediate buffer storage unit 204. In a storage device for storing used signals, a specific signal which may cause an access collision in a signal used for processing of an adjacent area is redundantly written.

In the above-described wavelet transform apparatus of the present invention, the horizontal processing and the vertical direction are executed in this order. However, the vertical processing may be performed first, and then the horizontal processing may be performed.

Normally, the wavelet transform is recursively repeated for LL coefficients. By cascading the wavelet transform devices of the present invention described above,
Wavelet transform of multiple layers is possible. As an example, FIG. 13 shows a configuration for performing a three-layer wavelet transform. In FIG. 13, 400_1, 400_
2, 400_3 are the wavelet transform devices of the present invention described above. Wavelet transformer 4
00_1 performs a first-layer wavelet transform on the image signal x. Among the obtained coefficient signals, LL
The coefficient signal is input to the wavelet transform device 400_2, and the second-layer wavelet transform is performed. Among the obtained coefficients, the LL coefficient signal is input to the wavelet transform device 400_3, and the wavelet transform of the third hierarchy is performed. Thus, 1HH, 1HL, 1LH,
2HH, 2HL, 2LH, 3HH, 3HL, 3LH, 3
LL coefficient signals are obtained. FIG. 14 shows such a 2
The state of band division by three-layer wavelet transform for a dimensional signal is shown. Such an apparatus for wavelet transform of two or more layers is also included in the present invention.

Embodiment 4 As one embodiment of the present invention, a wavelet transform coefficient signal by a wavelet transform device as shown in FIG.
There is an inverse wavelet transform device for restoring to a dimensional signal. Such an inverse wavelet transform device can be realized by using the above-described sub-band decoding device for vertical and horizontal processing. Next, an example of an inverse wavelet transform device of 4-parallel processing will be described with reference to FIGS.

The input signals are LL and L as shown in FIG.
H, HL and HH coefficient signals. In the inverse wavelet transform, processing in the vertical direction is performed first, and then processing in the horizontal direction is performed. The band division filter bank 202 in FIG.
_0 to 202_3 and 206_0 to 206_3 are replaced with band synthesis filter banks. The input signal is written into the input buffer storage unit 200, and a subband decoding process in the vertical direction is performed by a band synthesis filter bank (represented by 202_0 to 202_3 for convenience). In this process, FIG.
Of the left and right halves of the
Coefficients are divided into two regions, subband decoding is performed by band synthesis filter banks 202_0 and 202_1 for each left half region, and subband decoding is performed by band synthesis filter banks 202_2 and 202_3 for each right half region. Is performed. As a result, the L and H coefficients as shown in FIG. 10 are restored on the intermediate buffer storage unit 204.

When the inverse TS conversion is used for the vertical direction processing, the number of taps of the band synthesis filter exceeds three. Therefore, as described in relation to the sub-band decoding device, each of the band synthesis filter banks 202_0 to 202_3 has It is necessary to refer to the coefficient of the adjacent area near the area division boundary of each column in the vertical direction, and also needs a mirror image inverted signal of the coefficient signal at the beginning and end of each column. Therefore, one embodiment (4-1)
According to the embodiment, as described with reference to FIG. 11, the input buffer storage unit 200 includes a storage device in which a coefficient signal and a mirror image inversion signal used only for processing of each area are written,
The storage device is configured to include a storage device in which coefficient signals near the area division boundary used in processing of two adjacent areas are written, thereby avoiding access collision and reducing required storage capacity.

In the restored L and H coefficient signals as shown in FIG. 10, the upper half signal and the lower half signal are each 8 signals in the horizontal direction.
The coefficients are divided into two regions, and the sub-band decoding process is performed by the band synthesis filter banks 206_0 and 206_1 for each upper half region, and the sub-band decoding process is performed by the band synthesis filter banks 206_2 and 202_3 for each lower half region. Is performed. As a result, the original 2 shown in FIG.
The dimensional signal (for example, image signal) is stored in the output buffer storage unit 20.
8 is restored. When the inverse TS conversion is used, since the number of taps of the band synthesis filter exceeds 3, the region division of each row in the horizontal direction is performed in each band synthesis filter bank 206_0 to 206_3 as described in relation to the subband decoding device. It is necessary to refer to coefficients in an adjacent area near the boundary, and a mirror image inverted signal of the coefficient signal at the beginning and end of each row is required. Therefore, according to this embodiment (4-1), as described with reference to FIG. 9, the intermediate buffer storage unit 204 is written with the coefficient signal and the mirror image inverted signal used only for processing of each area. The storage device is configured to include a storage device and a storage device in which coefficient signals near an area division boundary used in processing of two adjacent areas are written, so that a necessary storage capacity is reduced and a processing stagnation due to an access wait is prevented.

Although not shown, according to another embodiment (4-2), mirror image inversion processing is required, but as in the embodiment (3-2) of the sub-band decoding device, the input buffer storage section is used. 20
The configuration is such that the mirror image inversion signal is not written in the 0 and / or the intermediate buffer storage unit 204. Although not shown, according to another embodiment (4-3), similarly to the embodiment (3-3), only in each area inside the input buffer storage unit 200 and / or the intermediate buffer storage unit 204. A specific coefficient signal that may cause an access collision among the coefficient signals near the area division boundary, which is also used for processing of an adjacent area, is written in a storage device for storing used signals in an overlapping manner. Configuration.

It should be noted that the inverse wavelet transform device described above is replaced with the wavelet transform device 400_ in FIG.
It is apparent that three layers of inverse wavelet transform are possible if used instead of 1,400_2,400_3. The direction of the signal is opposite to that in FIG. An apparatus for such inverse wavelet transform of two or more layers is also included in the present invention.

<Embodiment 5> As another embodiment of the present invention, a plurality of sets of the above-described parallel processing wavelet transform device and encoding device are used,
There is a compression device that performs compression encoding of a dimensional signal by parallel processing. FIG. 15 shows an example of such a compression device.
The compression device shown here divides the image signal 500 into four, and performs the wavelet transform of each region in parallel by the four wavelet transform devices 502_1, 502_2, 502_3, and 502_4. Each wavelet transform device 502_1,
502_2, 502_3, and 502_4 are shown in FIGS.
Alternatively, a wavelet transform device for parallel processing as described with reference to FIG. Therefore, the image signal 500
High-speed wavelet transform is possible. Each wavelet transform device 502_1, 502_2, 502_3, 502_
The wavelet coefficient signals of the quadrant output from 4 are divided into four encoding devices 504_1, 504_2, and 504_.
3, 504_4, and entropy-encoded in parallel, and a compressed encoded signal 506_ for each region of the image signal.
1, 506_2, 506_3, and 506_4 are output.

Embodiment 6 As another embodiment of the present invention, a plurality of sets of the inverse wavelet transform device and the decoding device for parallel processing as described above are used, and compression coding of an image or the like is performed by parallel processing. There is a decompression device for restoring an image or the like from a signal. With reference to FIG. 15, an example of such a decompression device will be described. Encoding devices 504_1, 504_2, and 504
_3 and 504_4 are decoding devices (for convenience, 504_1 and 504_4).
2, 504_3, 504_4) and the wavelet transform devices 502_1, 502_2, 502_3, 5
02_4 is the inverse wavelet transform device for parallel processing of the present invention described above (for convenience, 502_1, 502_2, 502_3, 50
2_4). The input signals are compression-encoded signals 506_1 and 506 of the quadrant of the image signal 500.
_2, 506_3, and 506_4, which are
4_1, 504_2, 504_3, and 504_4 are decoded in parallel, and the wavelet transform coefficient signals of the four divided regions are restored. For these wavelet transform coefficient signals, four inverse wavelet transform devices 502_1, 50
The inverse wavelet transform is executed in parallel by 2_2, 502_3, and 502_4. Each inverse wavelet transformer 5
Since parallel processing is also performed inside 02_1, 502_2, 502_3, and 502_4, high-speed inverse wavelet transform is possible. Thus, the original image signal 500 is restored.

It should be noted that a compression device / decompression device including one wavelet transform device / inverse wavelet transform device and one encoding device / decoding device is naturally included in the present invention. Such a configuration has a lower processing speed than the above-described parallel configuration of the compression device / decompression device, but the wavelet transform device / inverse wavelet transform device according to the present invention is capable of high-speed parallel processing. Processing speed can be achieved.

[0062]

As is apparent from the above description, according to the present invention, a sub-band encoder, a sub-band decoder,
Wavelet transform device, inverse wavelet transform device,
In a compression device or a decompression device, it is possible to eliminate the processing delay caused by the collision of the access to the buffer storage of the signal, to maximize the effect of the processing acceleration by the parallelization, and to reduce the storage capacity required for the buffer storage. Effects such as reduction, exclusion of exceptional access control, and elimination of mirror image inversion processing can simplify control and processing.

[Brief description of the drawings]

FIG. 1 is a schematic block diagram illustrating an example of a configuration of a subband encoding device according to the present invention.

FIG. 2 is a diagram illustrating an example of an internal configuration of an input buffer storage unit in FIG.

FIG. 3 is a diagram for explaining another example of the internal configuration of the input buffer storage unit in FIG. 1;

FIG. 4 is a diagram for explaining another example of the internal configuration of the input buffer storage unit in FIG. 1;

FIG. 5 is a diagram for explaining another example of the internal configuration of the input buffer storage unit in FIG.

FIG. 6 is a diagram for comparison with the internal configuration of the input buffer storage unit shown in FIG. 5;

FIG. 7 is a schematic block diagram illustrating an example of a configuration of a wavelet transform device according to the present invention.

FIG. 8 is a diagram illustrating an example of a two-dimensional input signal and its area division.

9 is a diagram for explaining an example of an internal configuration of an input buffer storage unit in FIG.

FIG. 10 is a diagram showing coefficient signals obtained by horizontal processing in the wavelet transform device shown in FIG. 7;

11 is a diagram for explaining an example of an internal configuration of an intermediate buffer storage unit in FIG.

12 is a diagram showing coefficient signals obtained by vertical processing in the wavelet transform device shown in FIG. 7;

FIG. 13 is a schematic block diagram of an apparatus for three-layer wavelet transform.

FIG. 14 is a diagram illustrating a state of band division by three-layer wavelet transform.

FIG. 15 is a block diagram showing an example of a compression device according to the present invention.

FIG. 16 is a diagram for explaining an access collision occurring when each area of a one-dimensional input signal is simply stored in a storage device corresponding to the area;

FIG. 17 is an explanatory diagram of buffer storage of a one-dimensional input signal in the related art.

FIG. 18 is an explanatory diagram of buffer storage of a two-dimensional input signal in the related art.

[Explanation of symbols]

Reference Signs List 102 Input buffer storage unit 102_0 to 102_8 Storage device in input buffer storage unit 104_0 to 104_3 Band division filter bank 200 Input buffer storage unit 200_0 to 200_9 Storage device in input buffer storage unit 202_0 to 202_3 Band division filter bank 204 Intermediate buffer storage unit 204_0 To 204_9 Storage device in the intermediate buffer storage unit 206_0 to 206_3 Band division filter bank 400_1 to 400_3 Wavelet transform device 502_1 to 502_4 of the present invention Wavelet transform device 504_1 to 504_4 of the present invention Encoding device

Claims (14)

[Claims] 1. A sub-band encoding apparatus that divides an input signal into a plurality of regions, and performs band division of the input signal in parallel by a plurality of band division filter banks corresponding to the respective regions on a one-to-one basis. For buffer storage of input signal,
A storage device corresponding to each region of the input signal and a storage device corresponding to each region division boundary of the input signal, and the storage device corresponding to each region has a signal used only for processing of the corresponding region. Is written in a storage device corresponding to each region division boundary, and a signal used for processing of two adjacent regions near the corresponding region division boundary is written. 2. The subband encoding apparatus according to claim 1, wherein the mirror image inversion signal is also written in the storage device corresponding to the first area and the last area of the input signal. 3. A storage device corresponding to a first region and a last region of an input signal is composed of two independent storage devices, and a mirror image inversion signal and a base signal are written in one of the storage devices. 2. The sub-band encoding apparatus according to claim 1, wherein a signal excluding a mirror image inversion signal and a signal based on the mirror image inversion signal, which is used only for processing of the corresponding area, is written in the other storage device. . 4. A sub-device according to claim 1, wherein a specific signal in a signal written to a storage device corresponding to the region division boundary is also written in the storage device corresponding to the region in an overlapping manner. Band coding device. 5. A subband decoding apparatus for dividing an input signal into a plurality of regions and performing band synthesis of the input signal in parallel by a plurality of band synthesis filter banks corresponding to each region in a one-to-one manner. For buffer storage of signals, the memory device has a storage device corresponding to each region of the input signal and a storage device corresponding to each region division boundary of the input signal, and the storage device corresponding to each region has a corresponding region. And a signal used for processing two adjacent areas near the corresponding area division boundary is written to a storage device corresponding to each area division boundary. Subband decoding device. 6. The sub-band decoding apparatus according to claim 5, wherein the mirror image inversion signal is also written in the storage devices corresponding to the first area and the last area of the input signal. 7. A storage device corresponding to a first region and a last region of an input signal is composed of two independent storage devices, and a mirror image inversion signal and a base signal are written in one of the storage devices. 6. The sub-band encoding apparatus according to claim 5, wherein a signal excluding the mirror image inversion signal and its base signal, which is used only for processing of the corresponding area, is written in the other storage device. . 8. A sub-device according to claim 5, wherein a specific signal in a signal written to a storage device corresponding to a region division boundary is also written to a storage device corresponding to a region in an overlapping manner. Band decoding device. 9. A wavelet transform device comprising two or more sub-band coding devices according to claim 1, 2, 3 or 4 for horizontal processing and vertical processing. 10. An inverse wavelet transform apparatus comprising two or more sub-band decoding apparatuses according to claim 5, for horizontal processing and vertical processing. 11. The wavelet transform device according to claim 9, further comprising an encoding device, wherein the two-dimensional signal is wavelet transformed by the wavelet transform device, and the obtained wavelet transform coefficient signal is encoded by the encoding device. A compression device. 12. A plurality of wavelet transform devices according to claim 9, and a plurality of encoding devices, wherein the two-dimensional signal is divided into a plurality of regions, and the two-dimensional signals are subjected to wavelet transform in parallel by the plurality of wavelet transform devices. A plurality of obtained wavelet transform coefficient signals in a plurality of regions, which are encoded in parallel by the plurality of encoding devices. 13. An inverse wavelet transform apparatus comprising: an inverse wavelet transform apparatus according to claim 10; and a decoding apparatus. A decompression device for performing inverse wavelet transform by a device to restore a two-dimensional signal. 14. A plurality of inverse wavelet transform devices according to claim 10, and a plurality of decoding devices, wherein the plurality of decoding encoded signals of a plurality of regions of the two-dimensional signal are decoded in parallel by the plurality of decoding devices. A decompression device, wherein the two-dimensional signals are restored by performing inverse wavelet transform on the obtained wavelet transform coefficient signals of the plurality of regions in parallel by the plurality of inverse wavelet transform devices.