JP2006295954A - Device and method for encoding image, and device and method for decoding image - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve the degrees of freedom on the selection of image quality and compressibility, without increasing the hardware constitution of the wavelet transformation. <P>SOLUTION: Input image data 100 are read by lines, in the descending order from the top line or a prescribed area of the input image data 100 is read, and input into a data reading memory 1. A predetermined amount of data is accumulated in the data reading memory 1, and a wavelet transformation part 5 performs a wavelet transformation filtering process in the horizontal/vertical directions. The wavelet transformation part 5 switch-controls the data from the data reading memory 1 with a switch 2 and sends to a fixed decimal-type wavelet transformation part 3 or an integer-type wavelet transformation part 4. The output from the wavelet transformation part 5 is encoded. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to an image encoding apparatus and method, and an image decoding apparatus and method for encoding a still image or a moving image with high efficiency using wavelet transform.

  As a conventional typical image compression method, there is a JPEG (Joint Photographic Coding Experts Group) method standardized by ISO (International Organization for Standardization). This JPEG system is a system that mainly compresses and encodes still images using DCT (Discrete Cosine Transform). When relatively high bits are assigned, a good encoded / decoded image is obtained. It is known to provide However, in this method, when the number of encoded bits is reduced to some extent, block distortion peculiar to DCT becomes remarkable, and deterioration becomes conspicuous subjectively.

  Apart from this, recently, research on a method of dividing an image into a plurality of bands by a filter that combines a high-pass filter and a low-pass filter called a filter bank, and performing coding for each band has become active. ing. Among them, wavelet coding is regarded as a promising new technology to replace DCT because there is no defect that block distortion becomes remarkable due to high compression, which is a problem in DCT.

  In addition, MPEG (Moving Picture image coding Exparts Group) is used for moving picture coding, and currently MPEG-1, MPEG-2, and MPEG-4 are known. In particular, MPEG-2 is a DVD (Dgital Versatile Disc) is widely used for video compression. As these JPEG and MPEG encoding means, encoding control is performed for each macroblock (usually 16 × 16) composed of several 8 × 8 blocks which are DCT processing units.

  Currently, many products such as electronic still cameras and video movies adopt the JPEG method, the MPEG method, or the so-called DV (Digital Video) method for compression encoding, and these compression encoding methods are all conversion methods. DCT is used. In the future, it is speculated that the products based on the wavelet transform will appear on the market, but studies for improving the efficiency of the coding method are being actively conducted by each research institution. In fact, JPEG2000 (under work by ISO / IEC / JTC1SC29 / WG1, the same organization as JPEG), which is expected to be the next generation still image international standard system that can be said to be the successor of JPEG, This is the format for which standardization recommendations will be issued. In JPEG2000, wavelet transform is decided to be adopted as a conversion method that is the basis of image compression instead of the existing JPEG DCT.

  This JPEG2000 standard is disclosed in Non-Patent Document 1.

Osama K. Al-Shaykh, Iole Moccagatta, and Homer Chen, "JPEG-2000: A New Still Image Compression Standard", Conference Record of the Thirty-Second Asilomar Conference on Signals, Systems & Computers, 1998, 1 Nov. 1998, vol.1

By the way, when using wavelet transform to obtain a high-quality encoded image not only for still images but also for moving images, it is important to solve the following problems. That is,
(1) In the JPEG-2000 Part-1 FCD (Final Committee Draft), there are two wavelet transform filters defined as of July 2000. One is an integer type 5x3 filter for lossless conversion, and the other is an irreversible floating point type (Float) 7x9 filter.
(2) Compared to the integer type 5x3 filter, the floating-point type (Float) 7x9 filter has a higher complexity in construction, and there is a problem in using both in hardware. Moreover, in order to guarantee the latter floating-point precision, a dedicated floating-point arithmetic unit is required, and there is a problem that leads to an increase in hardware circuit scale.
(3) On the other hand, as a result of the experiment, the fixed-point type (Fixed-point) 5x3 filter, which is the fixed-point precision improvement of the integer type (Integer) 5x3 filter, has the coding efficiency of the floating-point type (Float) 7x9 filter. Not only does it have excellent performance, it has many common parts between the Integer 5x3 filter and the internal arithmetic unit. Therefore, providing both has the effect of minimizing the increase in hardware without sacrificing encoding efficiency.
(4) In JPEG-2000 Part-1 FCD, the arithmetic expression of the integer type 5 × 3 filter is described, and according to this procedure, wavelet transform coefficients can be generated. However, there is no description about the calculation means of the fixed-point type (fixed-point) 5 × 3 filter, and it is very important to make the both compatible with each other because the circuit is shared.

  The present invention has been proposed in view of the above circumstances, and is an image encoding device that can increase the degree of freedom in selecting image quality and compression rate without increasing the hardware configuration of the wavelet transform, and It is an object to provide a method and an image decoding apparatus and method.

  In order to solve the above-described problem, the present invention provides an image encoding apparatus that encodes input image data, a recording unit that records a plurality of lines of the image data in order from the highest line, and a recording unit. Wavelet transform means for performing wavelet transform processing for each of the plurality of read lines, and encoding means for performing coding processing on the result of the wavelet transform means, and for each of the plurality of lines read from the recording means In addition, the processing by the wavelet transform unit and the encoding unit is sequentially executed.

  Here, the wavelet transform means includes a fixed-point wavelet transform means configured to include a bit shifter and a wavelet transformer, and an integer wavelet transform means configured to include only the wavelet transformer. Preferably, the wavelet transform unit of the integer wavelet transform unit and the wavelet transform unit of the fixed-point wavelet transform unit preferably have the same configuration. In addition, when performing lossless encoding, it is preferable to select the integer wavelet transform unit, and when performing lossy encoding, it is preferable to select the fixed-point wavelet transform unit.

  In order to achieve the above object, the present invention provides an image decoding apparatus that performs decoding processing on input encoded image data, a recording unit that records the encoded image data for each of a plurality of lines, and the recording A plurality of lines read out from the recording means, the decoding means performing decoding processing for each of the plurality of lines read out from the means, and wavelet inverse transformation means for performing wavelet inverse transformation processing on the result of the decoding means The processing by the decoding processing means and the wavelet inverse transformation processing means is sequentially executed every time.

  According to the present invention, the wavelet transform process is performed for each of the plurality of lines read from the recording unit that is sequentially recorded from the highest line of the plurality of lines of the image data, the encoding process is performed on the result of the wavelet transform, By sequentially executing wavelet transform processing and encoding processing for each of a plurality of lines read from the recording means, if the number of lines necessary for filtering is accumulated, the wavelet transform filter processing can be executed immediately, and the hardware configuration Can be simplified.

  Embodiments of an image encoding device and method, and a decoding device and method according to the present invention will be described below with reference to the drawings. In the following embodiments, in particular, a wavelet transform device used in an image encoding device and a wavelet inverse transform device used in an image decoding device will be described in detail.

First Embodiment FIG. 1 shows a configuration example of a wavelet transform apparatus used in an image encoding apparatus according to a first embodiment of the present invention.

  In FIG. 1, a data reading memory unit 1 that reads and stores input image data 100 in a predetermined area in a memory, and a wavelet transform in the horizontal or vertical direction with respect to these image areas as soon as the data is stored in the data reading memory unit 1 And a wavelet transform unit 5 for filtering. The wavelet transform unit 5 includes a fixed-point type wavelet transform unit 3 and an integer-type wavelet transform unit 4. The output from the data read memory unit 1 is switched by the switching unit 2, and these fixed-point type wavelet transform units 3 are switched. Alternatively, the data is sent to the integer wavelet transform unit 4.

Next, the operation of the first embodiment having the configuration of FIG. 1 will be described.
First, the input image data 100 is read out line by line or in a specific area in order from the most significant line, and input to the data read memory unit 1. When predetermined data is stored in the data read memory unit 1, the wavelet transform unit 5 performs wavelet transform filter processing in the horizontal and vertical directions. Normally, the filter used for wavelet transform filtering is a multi-tap filter, and the wavelet transform filter process can be executed immediately if the number of lines necessary for this filtering is accumulated.

  When the fixed-point wavelet transform unit 3 is selected in the wavelet transform unit 5, the switch unit 2 is switched and connected to the selected terminal a, and the output data 101 from the data read memory unit 1 is input data. 102 is sent to the fixed-point wavelet transform unit 3, where wavelet transform processing is performed by the fixed-point wavelet transform unit 3, and a transform coefficient 107 is output. On the other hand, when the integer type wavelet transform unit 4 is selected, the switching unit 2 is switched and connected to the selected terminal b side, and the output data 101 from the data read memory unit 1 is used as the input data 103 for the integer type wavelet transform. The wavelet transform process is applied to the unit 4 and the transform coefficient 108 is output.

  In FIG. 1 showing the first embodiment, the fixed-point type wavelet transform unit 3 and the integer type wavelet transform unit 4 are shown as independent configurations. Many of them can be shared. The sharing of this configuration will be described later.

Second Embodiment The second embodiment of the present invention embodies the fixed-point wavelet transform unit 3 described in the first embodiment, and this fixed-point wavelet transform unit 3. An example of the configuration is shown in FIG.

That is, FIG. 2 shows a configuration example of the fixed-point type wavelet transform unit 3 in the configuration of FIG. 1, and the fixed-point type wavelet transform unit 3 in FIG. 2 includes a bit shifter 6 and a wavelet transform unit 7. And is configured. The bit shifter 6 is used once for input image data. For example, as shown in FIG. 3A, the bit shifter 6 performs a bit shift process on the input data to data of fixed-point precision. The point P in FIG. 3A indicates the position of the fixed point, the bit on the upper side (left side in the figure) from this point P is the integer part, and the bit on the lower side (right side in the figure) is the decimal part. . In the example of FIG. 3A, since the fractional part has 6 bits, the fixed point precision can be ensured with an accuracy of 1/2 ⁶ = 1/64. Further, in the example of FIG. 3A, the integer part is assumed to handle 8-bit data, and 16-bit calculation is performed as a whole (the calculation accuracy is 16 bits). 2) Guard-bit is provided on the side), 1 bit of this indicates the sign (+/-), and the other bit is prepared as a bit to avoid overflow (Overflow) Is.

  On the other hand, (B) in FIG. 3 shows an example of integer precision calculation. Since the least significant bits (right end in the figure) are used as the integer part, the point P indicating the decimal point is the right end (highest position). Exists on the lower side. In addition, in the calculation of integer precision, only the bit indicating the sign (+/−) is provided.

Returning to FIG. 2 again, the operation will be described.
The bit shifter 6 shown in FIG. 2 operates when the input data 102 is image data, that is, for image data before wavelet transformation is first performed, and does not operate thereafter. That is, when performing wavelet transform of a plurality of stages, the bit shifter is used so that the data 102 as the wavelet transform coefficient that has been subjected to at least one (one stage) wavelet transform is used as the wavelet transformer input 109 as it is. 6 is configured to pass through.

  The integer wavelet transform unit 4 in FIG. 1 is composed of only the wavelet transformer 7 without the bit shifter 6 in FIG.

  Here, an example of the configuration and operation of general wavelet transform / inverse transform will be described with reference to the drawings.

  First, FIG. 4 is given as a configuration of a normal wavelet transform unit. The example of FIG. 4 shows a configuration example in which octave division, which is the most general wavelet transform among several methods, is performed over a plurality of levels. In the case of FIG. 4, the number of levels is 3 (level 1 to level 3), the image signal is divided into a low frequency and a high frequency, and only the low frequency components are hierarchically divided. Yes. In FIG. 4, for the sake of convenience, wavelet transform for a one-dimensional signal (for example, a horizontal component of an image) is illustrated. However, by extending this to two dimensions, a two-dimensional image signal can be handled.

The input image signal 220 to the wavelet transform unit shown in FIG. 4 is obtained by first performing band division by the low-pass filter 21 (transfer function H ₀ (z)) and the high-pass filter 22 (transfer function H ₁ (z)). The low-frequency component and the high-frequency component are each thinned out by a half (level 1) by the corresponding down-samplers 23a and 23b. There are two outputs at this time: an L component 221 and an H component 226. Here, L is Low and indicates a low frequency, and H is High and indicates a high frequency. The low-pass filter 21, the high-pass filter 22, and the two downsamplers 23a and 23b in FIG. 4 constitute a level 1 circuit unit 201.

  Of the signals thinned out by the down-samplers 23a and 23b, only the low-frequency component, that is, the signal from the down-sampler 23a, is further band-divided by the low-pass filter and high-pass filter of the level 2 circuit unit 202, The corresponding downsampler reduces the resolution by half each (level 2). As the circuit unit 202 including the level 2 low-pass filter, high-pass filter, and down-sampler, the same configuration as the circuit unit 201 including the level 1 low-pass filter 21, high-pass filter 22, and down-samplers 23a and 23b is used. .

  By performing such 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 222 and the LH component 225. FIG. 4 shows an example in which the band is divided up to level 3, and the output from the downsampler on the low-pass filter side of the level 2 circuit unit 202 is sent to the level 3 circuit unit 203 having the same configuration as the circuit unit 201. Have been supplied. As a result of the band division to level 3, the LLL component 223, the LLH component 224, the LH component 225, and the H component 226 are generated.

  Next, FIG. 5 shows a specific configuration example of the wavelet inverse transform unit that performs the reverse operation to the wavelet transform unit shown in FIG.

  That is, when the band components 223, 224, 225, and 226, which are the outputs of the wavelet transform unit 2 described in FIG. 4, are input to the wavelet inverse transform unit in FIG. 5, first, the LLL component 223 and the LLH component 224 are input. Are upsampled to double the resolution by upsamplers 24a and 24b, respectively. Subsequently, the low-frequency component is filtered by the low-pass filter 25 and the high-frequency component is filtered by the high-pass filter 26, and the adder 27 combines both band components. By the circuit unit 206 so far, the inverse conversion as the inverse process of the conversion in the level 3 circuit unit 203 in FIG. 4 is completed, and the LL component 227 which is the band component on the low frequency side of the level 2 is obtained. It is done. By repeating this process to level 1 thereafter, the final decoded image 229 after inverse transformation is output. That is, the level 2 circuit unit 207 and the level 1 circuit unit 208 have the same configuration as the level 3 circuit unit 206, and the output of the level 3 circuit unit 206 is lower than that of the level 2 circuit unit 207. And the output of the level 2 circuit unit 207 is sent as the low frequency side input of the level 1 circuit unit 208, respectively. The above is the basic configuration of a normal wavelet inverse transform unit.

Here, FIG. 6 illustrates band components obtained as a result of band division of a two-dimensional image up to level 3. FIG. The notation of L and H in FIG. 6 is different from the notation of L and H in FIGS. 4 and 5 that deal with one-dimensional signals. That is, in FIG. 6, first, the four components LL, LH, HL, and HH are divided by level 1 band division (horizontal and vertical directions). Here, LL means that the horizontal and vertical components are both L, and LH means that the horizontal component is H and the vertical component is L. Next, the LL component is band-divided again to generate LLLL, LLHL, LLLH, and LLHH. Furthermore, the LLLL component is band-divided again to generate LLLLLL, LLLLHL, LLLLLH, and LLLLHH. Incidentally, _T LLLLLL in FIG _{6, T LLLLHL, T LLLLLH,} T LLLLHH, T LLHL, T LLLH, T LLHH, T LH, T HL, T HH represents a weight coefficient for each subband. Further, as shown in FIG. 6, in addition to dividing the low-frequency component hierarchically, the entire band is also divided equally.

Third Embodiment Next, a third embodiment of the present invention will be described. In the third embodiment, the internal wavelet transform unit 3 and the fixed-point wavelet transform unit 4 shown in FIG. 1 have the same internal wavelet transform. Hereinafter, specific operations of the integer wavelet transform and the fixed-point wavelet transform will be described.

First, an integer type wavelet transform operation by an integer precision type 5 × 3 filter (analyzing side) defined by JPEG-2000 Part-1 FCD (Final Committee Draft) will be described with reference to FIG. FIG. 7 is a diagram for explaining the operation of the integer precision type 5 × 3 filter. One-dimensional wavelet transformation is performed to convert the input image data at the left end in the figure into a low-frequency component s and a high-frequency component d. The operation is shown. The detailed operation will be described below with reference to FIG. Here, the d _m ⁿ is the level of the wavelet transform and a m-th high-frequency component coefficient of n. Similarly, ^let s _m ⁿ be the m-th low-frequency component coefficient whose wavelet transform level is n. When n = 0, as shown in FIG. 7, it is the input image itself.

As shown in FIG. 7, the formula for calculating the low frequency component coefficient s is generally:
s _{m + 1} ^{n + 1} = s _{m + 1} ⁿ + ((d _m ^{n + 1} + dm _{+ 1} ^{n + 1} +2) / 4) (1)
Given in. Here, the ¼ operation in the equation can be realized by bit shift, and when the 2-bit right shift is expressed as “>> 2”, the above equation (1) is
s _{m + 1} ^{n + 1} = s _{m + 1} ⁿ + ((d _m ^{n + 1} + dm _{+ 1} ^{n + 1} +2) >> 2) (1 ')
It can be expressed as Note that +2 in parentheses in the above formula (1) is for compensating for a rounding error that occurs during a right shift.

Also, the calculation formula of the high frequency component coefficient d is generally
d _m ^{n + 1} = d _m ⁿ − ((s _m ⁿ + s _{m + 1} ⁿ ) / 2) (2)
Alternatively, 1-bit right shift is represented by ">>1"
d _m ^{n + 1} = d _m ⁿ − ((s _m ⁿ + s _{m + 1} ⁿ ) >> 1) (2 ′)
Given in.

Here, in the example of FIG. 7, the first-stage wavelet transform coefficient, that is, the low-frequency component coefficient and the high-frequency component coefficient of level n = 1 are calculated for the input image data. s and d do not represent a low-frequency component coefficient and a high-frequency component coefficient. Further, at a position corresponding to the edge of the screen, data corresponding to the outside of the screen is used for returning data outside the screen. For example, in order to obtain the data _s ^{0 1} in FIG. _7, a _s ^{0 _0,} and _d ^{0 1,} further using the same _d ^{0 1} as wrapping _d ^{0 1.} Similarly, two s ₇ ^0s are used in order to obtain d ₇ ¹ .

A portion SP surrounded by a broken line in FIG. 7 is a portion for obtaining a low-frequency component coefficient s ₂ ¹ corresponding to n = 0 and m = 1 in the above formula (1).
s ₂ ¹ = s ₂ ⁰ + ((d ₁ ¹ + d ₂ ¹ +2) / 4)
Has been calculated. Further, a portion DP surrounded by a broken line in FIG. 7 is a portion for obtaining a high frequency component coefficient d ₀ ¹ corresponding to n = 0 and m = 0 in the above equation (2).
d ₀ ¹ = d ₀ ⁰ − ((s ₀ ⁰ + s ₁ ⁰ ) / 2)
Has been calculated. Here, as described above, ¼ multiplication can be realized by 2-bit right shift “>> 2” and ½ multiplication can be realized by 1-bit right shift “>> 1”.

  The amount of calculation required to calculate the above-described equations (1) and (2) is multiplication 0, addition / subtraction 5, and shift calculation 2 when multiplication is realized by the shift calculation. By performing this operation in the same manner from top to bottom, all the coefficients can be calculated. As described above, for a two-dimensional signal such as an image, the coefficient group generated by the wavelet transform in the one-dimensional direction (for example, the vertical direction) is in another direction (for example, the horizontal direction). The wavelet transform may be performed in the same manner as described above.

Next, a fixed-point wavelet transform operation on the analysis side using a 5 × 3 filter with fixed-point precision will be described with reference to FIG. FIG. 8 shows an operation of performing one-dimensional wavelet transform and converting the leftmost input image data into a low-frequency component s and a high-frequency component d. The detailed operation will be described below with reference to FIG. Here, the d _m ⁿ is the level of the wavelet transform and a m-th high-frequency component coefficient of n. In the case of n = 0, it is clear from the figure that the input image. Similarly, ^let s _m ⁿ be the m-th low-frequency component coefficient whose wavelet transform level is n.

As is apparent from FIG. 8, the calculation formula for the low-frequency component coefficient s is generally:
s _{m + 1} ^{n + 1} = s _{m + 1} ⁿ + β × (d _m ^{n + 1} + dm _{+ 1} ^{n + 1} +2) (3a)
Or
^{_{s m + 1 n + 1 =}} s m + 1 n + β × (d m n + 1 + d m + 1 n + 1) ··· (3b)
Given in.

Similarly, the calculation formula of the high frequency component coefficient d is generally
d _m ^{n + 1} = d _m ⁿ −α × (s _m ⁿ + s _{m + 1} ⁿ ) (4)
Given in.

Here, α = 0.5 and β = 0.25, and d corresponding to the high frequency component is Nyquist gain = 2 on the analysis side, so that the gain is adjusted so that Nyquist gain = 1. This is the reason why the high frequency component coefficient d is multiplied by S _H = −0.5. Further, in the case of the above-described integer precision 5 × 3 filter, when gain adjustment is performed, reversible conversion is not guaranteed, and thus gain adjustment is not necessary. This peripheral technology is known as a general digital signal processing technology.

When calculating the data of the position of the screen edge, when data outside the screen is necessary, the data inside the adjacent screen is used by folding (for example, d ₀ ¹ for obtaining s ₀ ^{1 in} FIG. 8). This is the same as in the case of FIG. 7 described above, and is the same in FIGS. 9 and 10 described later.

  Next, the difference between the above formulas (3a) and (3b) will be described. Since the fixed-point precision 5x3 filter has higher precision than the integer precision, the expression (3a) (compatible) for rounding +2 described in the integer precision 5x3 filter and the expression without +2 (3b) ) (Incompatible).

  When the configuration of the arithmetic means is shared between the fixed-point precision 5x3 filter and the above-described integer precision 5x3 filter, it is necessary to use the above equation (3a). That is, when using the configuration for performing the calculation of the above equation (3a) as a wavelet transformer used for a fixed-point precision 5x3 filter, it should also be used as a wavelet transformer used for the integer precision 5x3 filter. Can do.

  In such a fixed-point precision 5 × 3 filter described with reference to FIG. 8, the amount of calculation required to generate one set of coefficients of a low-frequency component and a high-frequency component is multiplication 3, addition / subtraction 5 (compatible), or addition / subtraction 4 (Incompatible).

Fourth Embodiment Next, a fourth embodiment of the present invention will be described. In the fourth embodiment, the content of the wavelet transform analysis filter is embodied in the synthesis-side wavelet inverse transform filter (synthesizer filter), whereas the third embodiment is related to the wavelet transform analysis filter. . First, the operation of the integer precision type 5 × 3 filter (combining side) defined in the FCD of JPEG-2000 Part-1 will be described with reference to FIG.

FIG. 9 is a diagram for explaining the operation of the integer precision type 5 × 3 synthesis filter. One-dimensional wavelet transform is performed, and the low-frequency component s at the left end and the high-frequency component d at the center of FIG. It shows how the output image data at the middle right end is calculated. In FIG. 9, the d _m ⁿ is the level of the wavelet transform and a m-th high-frequency component coefficient of n. Similarly, ^let s _m ⁿ be the m-th low-frequency component coefficient whose wavelet transform level is n. Here, when n = 0, an output image is obtained as shown in FIG.

As shown in FIG. 9, the calculation formula for the low-frequency component coefficient s of one lower level is generally:
s _{m + 1} ⁿ = s _{m + 1} ^{n + 1} − ((d _m ^{n + 1} + dm _{+ 1} ^{n + 1} +2) / 4) (5)
Given in. Here, when 1/4 multiplication in the equation (5) is realized by a bit shift and a 2-bit right shift is expressed by “>> 2”, the above equation (5) is
s _{m + 1} ⁿ = s _{m + 1} ^{n + 1} − ((d _m ^{n + 1} + dm _{+ 1} ^{n + 1} +2) >> 2) (5 ′)
It can be expressed as +2 in parentheses in the expression (5) or (5 ′) is for the rounding error that occurs during the right shift. This is the same as the rounding process on the analysis side already described in the third embodiment. belongs to. Of course, it is necessary for the rounding process to coincide on the analysis / synthesis side in order to maintain reversibility.

Similarly, the calculation formula of the high frequency component coefficient d is generally
d _m ⁿ = d _m ^{n + 1} + ((s _m ⁿ + s _{m + 1} ⁿ ) / 2) (6)
Alternatively, 1-bit right shift is represented by ">>1"
d _m ⁿ = d _m ^{n + 1} + ((s _m ⁿ + s _{m + 1} ⁿ ) >> 1) (6 ′)
Given in.

Here, a portion SP surrounded by a broken line in FIG. 9 is a portion for obtaining a low-frequency component coefficient s ₂ ⁰ corresponding to n = 0 and m = 1 in the above equation (5).
s ₂ ⁰ = s ₂ ¹ − ((d ₁ ¹ + d ₂ ¹ +2) / 4)
Has been calculated. Further, a portion DP surrounded by a broken line in FIG. 7 is a portion for obtaining a high frequency component coefficient d ₀ ⁰ corresponding to n = 0 and m = 0 in the above formula (2).
d ₀ ⁰ = d ₀ ¹ + ((s ₀ ⁰ + s ₁ ⁰ ) / 2)
Has been calculated. Here, as described above, ¼ multiplication can be realized by 2-bit right shift “>> 2” and ½ multiplication can be realized by 1-bit right shift “>> 1”.

  The above-described equations (5) and (6), in particular, the amount of calculation required to calculate the equations (5 ′) and (6 ′) using the shift operation, that is, one set of even-numbered and odd-numbered coefficients are calculated. The amount of calculation required for generation is multiplication 0, addition / subtraction 5, and shift calculation 2. By performing this operation in the same manner from top to bottom, all the coefficients can be calculated. As described above, for a two-dimensional signal such as an image, another direction is applied to a coefficient group generated by inverse wavelet transform (synthesis filter processing) in a one-dimensional direction (for example, vertical direction). The wavelet inverse transformation may be performed in the same manner as described above (for example, in the horizontal direction).

Next, the synthesis-side fixed-point wavelet inverse transform operation using a 5 × 3 filter with fixed-point precision will be described with reference to FIG. FIG. 10 shows an operation of performing one-dimensional wavelet transform (inverse transform) and converting the low-frequency component s and the high-frequency component d at the left end in the figure to the output pixel at the right end in the figure. The detailed operation will be described below with reference to FIG. Here, the d _m ⁿ is the level of the wavelet transform and a m-th high-frequency component coefficient of n. Similarly, ^let s _m ⁿ be the m-th low-frequency component coefficient whose wavelet transform level is n. When n = 0, s and d are odd-numbered and even-numbered pixels, respectively.

As is apparent from FIG. 10, the odd-numbered pixel s at the right end of the figure, or the formula for calculating the low-frequency component coefficient s is generally:
_{^{s m + 1 n = s m}} + 1 n + 1 -β × (d m n + 1 + d m + 1 n + 1 +2) ··· (7a)
Or
_{^{s m + 1 n = s m}} + 1 n + 1 -β × (d m n + 1 + d m + 1 n + 1) ··· (7b)
Given in.

Similarly, the calculation formula for the even-numbered pixel d at the right end of the figure or the high-frequency component coefficient d is generally:
d _m ⁿ = d _m ^{n + 1} + α × (s _m ⁿ + s _{m + 1} ⁿ ) (8)
Given in.

Here, α = 0.5 and β = 0.25. As described in the third embodiment, since d corresponding to the high frequency component is Nyquist gain = 2 on the analysis side, the gain is adjusted so that Nyquist gain = 1, and the high frequency component coefficient d is determined on the analysis side. to, had been multiplied by S _{H =} -0.5, in contrast, the synthetic side, it is necessary to return to the 2 Nyquist gain from 1. For this purpose, after the high frequency coefficient d is multiplied by S _H = −2.0, the calculation of the above equation (8) is performed.

  Next, the difference between the above formulas (7a) and (7b) will be described. Since the fixed-point precision 5x3 filter has a higher precision than the integer precision, the expression (7a) (compatible) for +2 rounding and +2 are performed in the same manner as the rounding process in the analysis-side filter described above. Both formulas (7b) (not compatible) that are not attached are conceivable.

  When the configuration of the arithmetic means is shared between the fixed-point precision 5x3 filter and the integer precision 5x3 filter, it is necessary to use the above equation (7a).

  In such a fixed-point precision 5 × 3 filter described with reference to FIG. 10, the amount of computation required to generate one set of coefficients of a low-frequency component and a high-frequency component is multiplication 3, addition / subtraction 5 (compatible), or addition / subtraction 4 (Incompatible).

5. Fifth Embodiment A fifth embodiment of the present invention is a multiplier that is a component of the wavelet transformer of the integer wavelet transformer 4 and the wavelet transformer of the fixed-point wavelet transformer 3. Alternatively, the shift calculator, adder / subtracter, and register are all shared. Here, as described above, it is clear that the wavelet transformer is composed of a multiplier, a shift arithmetic unit, and an adder / subtracter. When actually implementing hardware, a register for temporarily storing data is used. Is required. In addition, the configuration of the integer type filter and the fixed point type filter can be made common to the wavelet inverse transformer (synthesis filter).

  11 and 12 show an example of a hardware configuration for calculating the coefficients of the high frequency and low frequency components on the synthesis side. First, a configuration for calculating a low-frequency component coefficient value or pixel value (in the case of level 0) will be described with reference to FIG. Here, for convenience of explanation, three registers Rm, Rn, and Rd, adders 61, 62, and 65 respectively indicated by +, +2, and − in the figure, a multiplier 63 that performs an operation of × 0.25, and a floor And a rounder 64 shown in FIG. Each of the Rm, Rn, and Rd registers is configured to store, for example, four values x, y, z, and w, and the predetermined filtering described above is performed on the coefficient values stored in the registers. By performing the above calculation, the coefficient value or pixel value of one lower level (in the case of level 0) is calculated.

That is, four values from the register Rm (e.g. _{^{d 0 1, d 0 1,}} d 1 1, d 2 1) and, four values from the register Rn (e.g. _{^{d 0 1, d 1 1,}} d 2 1, d ₃ ¹ ) are respectively added by the adder 61, 2 is added to each addition result by the adder 62, and 0.25 is respectively added to the four output values from the adder 62 by the multiplier 63. The four values obtained by multiplication (or 2-bit right shift) and rounding by the rounder 64 are sent to the adder 65, and these four values are sent to the four values (for example, s ₀₎ from the register Rd. ¹ , s ₁ ¹ , s ₂ ¹ , s ₃ ¹ ), by subtracting the coefficient value or pixel value (for example, s ₀ ⁰ , s ₁ ⁰ , s ₂ ⁰ , s) ₃ ⁰ ) is calculated.

  For example, if an integer register having a certain bit length is provided, the above-described filtering with integer precision can be executed, and if a fixed-point register having a certain bit length is used, the above-described filtering with fixed point precision can be performed. In addition to the register accuracy, the hardware configuration need not be changed at all. In this way, common hardware components can be realized.

  In the configuration of the specific example of FIG. 11, the low frequency component is stored in the register Rd, the high frequency component is stored in the registers Rm and Rn, and the operation result is newly stored in the register Rd. By doing so, the coefficient that has been used and overwritten is overwritten, so that it is not necessary to use an extra register, which is efficient. It also contributes to hardware reduction.

FIG. 11 is a specific example of the rounding process of the rounder 64 indicated by floor. In this example, truncation by floor is not rounded down to the decimal point, but is defined to round to the smaller integer as a value. . According to this definition, for example,
+2.3 → +2 (positive value)
-2.3 → -3 (negative value)
become.

  Next, a configuration for calculating the coefficient value or pixel value (in the case of level 0) of the high frequency component will be described with reference to FIG. Here, for convenience of explanation, there are three registers Rm, Rn, and Rd, adders 71 and 75, a multiplier 72 that performs an operation of × 0.5, and a rounder 73 indicated by floor. It shall be configured. Each of the Rm, Rn, and Rd registers is configured to store, for example, four values x, y, z, and w, and the predetermined filtering described above is performed on the coefficient values stored in the registers. By performing the above calculation, the coefficient value or pixel value of one lower level (in the case of level 0) is calculated.

That is, four values from the register Rn (for example, s ₀ ⁰ , s ₁ ⁰ , s ₂ ⁰ , s ₃ ⁰ ) and four values from the register Rm (for example, s ₁ ⁰ , s ₂ ⁰ , s ₃ ⁰ , s ₄ ⁰ ) are respectively added by the adder 71, 0.5 is multiplied (or 1-bit right-shifted) by the multiplier 72 for each addition result, and each rounding process is performed by the rounder 73. By sending the four values obtained to the adder 74 and subtracting these four values from the four values from the register Rd (eg, d ₀ ¹ , d ₀ ¹ , d ₁ ¹ , d ₂ ¹ ), The coefficient value or pixel value (for example, d ₀ ⁰ , d ₁ ⁰ , d ₂ ⁰ , d ₃ ⁰ ) of the high-frequency component of one lower level is calculated. Here, among the four values of the registers ^{_{Rd (d 0 1, d 0}} 1, d 1 1, d 2 1), the uses two of _d ^{0 1,} as described above, the ends of the screen This is because the data located outside the screen is supplemented by the adjacent data inside the screen (see d ₀ ^{1 in} FIG. 9 and the broken-line circle that is turned back).

  In the case of the specific example of FIG. 12, as in the specific example of FIG. 11, the hardware configuration is completely different only by using each register Rm, Rn, Rd as an integer register or a fixed-point register. The configuration of the integer type filter and the fixed point type filter can be made common without changing. In addition, the rounding process of the rounder 73 indicated by “floor” in FIG. 12 is also defined so as to perform the process of rounding to the smaller integer as a value instead of truncation after the decimal point, similarly to the rounder 64 of FIG. Keep it. In the specific example of FIG. 12, the low frequency components are stored in the registers Rm and Rn, the high frequency components are stored in the register Rd, and the calculation result is newly stored in the Rd register. In this way, since the coefficient already used and overwritten is overwritten, it is not necessary to use an extra register, which is efficient. It also contributes to hardware reduction. It is self-evident that there are other uses for these registers. If the register is large, the amount of data that can be stored increases. Therefore, the amount that can be simultaneously filtered increases, which can contribute to speeding up the processing. .

  By the way, the multipliers 63 and 72 shown in FIGS. 11 and 12 can be replaced by shift arithmetic units. Specifically, it is obvious that the × 0.25 multiplier 63 in FIG. 11 corresponds to a right 2-bit shift, and the × 0.5 multiplier 72 in FIG. 12 corresponds to a right 1-bit shift. In general, it is known that the shift calculator is less hardware than the multiplier, and is more efficient.

Sixth Embodiment The sixth embodiment of the present invention performs lossless encoding in a configuration in which an integer precision wavelet transform and a fixed-point precision wavelet transform can be switched as shown in FIG. In this case, the integer wavelet transforming unit is selected, and when performing lossy encoding, the fixed-point wavelet transforming unit is selected.

  That is, as described in the fifth embodiment, the integer precision wavelet transform and the fixed-point precision wavelet transform can be shared except for the bit precision. However, since the fixed-point precision type requires a bit for the precision of the decimal part as shown in FIG. 3, a register larger than the integer precision type is required. This leads to an increase in hardware. Here, when the lossless wavelet transform necessary for lossless coding (Lossless), which is also supported by JPEG-2000 Part-1 FDC, is to be performed, as described above, it is combined with the analysis side. If the rounding process is unified on both sides, the integer precision wavelet transform is selected in terms of less hardware, considering that both the integer precision wavelet transform and the fixed-point precision wavelet transform can be realized without problems. It is preferable to do this. On the other hand, since lossy encoding (Lossy) often requires high precision, it is a good idea to select a fixed-point precision wavelet transform.

  Considering the above points, switch to the integer wavelet transform when lossless lossless encoding is selected, and the fixed-point wavelet when lossy lossy encoding is selected. By switching to the conversion, the optimum wavelet transform for each encoding can be realized.

Seventh Embodiment As a seventh embodiment of the present invention, an image quality-oriented code is used in a configuration in which an integer precision wavelet transform and a fixed-point precision wavelet transform can be switched as shown in FIG. When selecting the fixed-point type wavelet transform unit, select the integer type wavelet transform unit when performing hardware reduction, power saving, and low bit rate encoding. The thing which was done is mentioned.

  This is because, when aiming at high accuracy (high image quality), it is usual to select a fixed-point wavelet transform that can maintain the accuracy higher than the integer accuracy. Further, when emphasizing hardware reduction, power saving, and high compression (low bit rate coding), it is better to select integer wavelet transform means. Naturally, the fixed-point wavelet transform that requires a register with a long bit length in order to increase the bit accuracy will increase the hardware scale, and the scale of the arithmetic units such as adders will increase accordingly. Because it does.

Further, the compression rate, that is, the bit rate, is greatly affected by the quantization processing as the processing after the wavelet transform. As a quantization method in this quantization processing, the simplest is as shown in the following equation (9), the wavelet transform coefficient value x is divided by the quantization index value Δ, and the obtained value is used as the quantization coefficient. Q. That is,
Q = x / Δ (9)
x is a wavelet transform coefficient value, and Δ is a quantization index value.

  Here, if the quantization index value or the quantization step size is increased, the quantization coefficient obtained by dividing the wavelet transform coefficient value is decreased, so that the compression rate is increased and the image quality is deteriorated. Conversely, if the quantization step size is reduced, the quantization coefficient obtained by dividing the wavelet transform coefficient value is increased, so that the compression rate is lowered and the image quality is improved. Therefore, a predetermined compression rate or image quality can be obtained by controlling this quantization index value.

  In high compression (low bit rate encoding), as can be seen from Equation (9), the wavelet transform coefficient value x is divided by a larger quantization index value Δ, so that the output quantization coefficient is the original value. The difference in the accuracy of the wavelet transform coefficients is reduced. That is, the difference between the fixed-point precision wavelet transform coefficient and the integer precision wavelet transform coefficient becomes very small.

  For the above reasons, it is better to select the integer wavelet transform means for high compression (low bit rate encoding).

Eighth Embodiment As an eighth embodiment of the present invention, in contrast to the above-described wavelet transform unit on the image coding device side or analysis side, wavelet inverse transform on the decoding device side or the synthesis side is used. The part is presented. That is, FIG. 13 is a diagram showing a configuration of a wavelet inverse transform device of an image decoding device corresponding to the wavelet transform device of the image coding device shown in FIG.

  In FIG. 13, a wavelet inverse transform unit 8 corresponds to the wavelet transform unit 5 in FIG. 1, and the transform coefficient 107 from the fixed-point wavelet transform unit 3 in FIG. 1 is used as the transform coefficient 111 in FIG. 1 is supplied to the fixed-point wavelet inverse transform unit 9, and the transform coefficient 108 from the integer wavelet transform unit 4 in FIG. 1 is supplied to the integer wavelet inverse transform unit 10 as the transform coefficient 112 in FIG. The output 113 from the fixed-point wavelet inverse transform unit 9 is sent to the selected terminal a of the switching unit 11, and the output 114 from the integer type wavelet inverse transform unit 10 is sent to the selected terminal b of the switching unit 11. An output 115 from the switching unit 11 is sent to the data writing memory unit 12.

  Next, the operation will be described. The wavelet inverse transform unit 8 includes a fixed point type wavelet inverse transform unit 9 and an integer type wavelet inverse transform unit 10. When a wavelet transform coefficient 111 having a fixed point precision is inputted, the fixed point type wavelet inverse transform unit 8 is provided. The wavelet inverse transformer 9 performs inverse transform and outputs a decoded image 113.

  On the other hand, when the integer precision wavelet transform coefficient 112 is input, the integer wavelet inverse transformer 10 performs inverse transform and outputs a decoded image 114. Either one of the decoded image 113 and the decoded image 114 is selected by the switch of the switching unit 11 and is input to the data writing memory unit 12 as an output 115. In the data writing memory unit 12, an image corresponding to the area of the decoded image is written in the memory.

  FIG. 14 shows a configuration example of the fixed-point wavelet inverse transform unit 9, which is composed of a wavelet inverse transformer 13 and a bit shifter 14. Regarding the fixed-point precision, the difference from the integer precision has already been described with reference to FIG. 3. However, since the analysis side performs a bit shift to the left, the composition side of the present embodiment performs a right bit shift, To return to the original level. The bit shift is necessary only after the final decoded image (level 0) is obtained. That is, when the output 116 from the wavelet inverse transformer 13 is a transform coefficient up to level 1, the output 116 from the wavelet inverse transformer 13 is the final decoded image (level 0) without going through the bit shifter 14. Only when this occurs, the bit shifter 14 performs bit shift processing.

  Note that the inverse transform means in the fixed-point wavelet inverse transform unit and the integer wavelet inverse transform unit have already been described with reference to FIGS. 9 and 10, and thus the description thereof is omitted in this embodiment.

Ninth Embodiment The ninth embodiment of the present invention presents a specific example of the wavelet inverse transformer 13 shown in FIG. 14, and the wavelet inverse transformer 13 is a multiplier or a shift calculator. , An adder / subtracter, and a register. The hardware configuration of such a wavelet inverse transformer has already been described with reference to FIGS. 11 and 12 in the fifth embodiment. In addition, as described above, the wavelet inverse transformer of the integer-type wavelet inverse transform unit on the synthesis side and the wavelet inverse transformer of the fixed-point wavelet inverse transform unit can be shared. Furthermore, as already described, the calculation means on the synthesis side does not differ between fixed-point precision and integer precision.

Tenth Embodiment A tenth embodiment of the present invention relates to transmission / reception between a wavelet transform unit on the encoding side and a wavelet inverse transform unit on the decoding side, and is an integer type wavelet transform unit and / or When decoding an encoded bitstream generated by an encoding device including a fixed-point wavelet transform unit using a decoding device including an integer wavelet inverse transform unit and a fixed-point wavelet inverse transform unit, the decoding device Then, it is provided with means for detecting information indicating whether the encoded bitstream is converted by an integer type or a fixed-point type wavelet transform means. When the decoded image is output without performing a bit shift after the inverse transform and is a fixed-point type, the high-frequency component coefficient gain is An in-adjustment unit and a bit shift unit after inverse conversion are performed to provide a unit for outputting a decoded image.

  FIG. 15 shows a schematic configuration of a main part of the tenth embodiment, which includes a bit accuracy detecting unit 31, a switch 32, a gain adjusting unit 33, a wavelet inverse transformer 34, and a bit shifter 35. Has been.

  That is, in FIG. 15, the bit accuracy detection unit 31 that has received the encoded bit stream 130 has used the encoded bit stream 130 to which wavelet transform has been performed in the encoder wavelet transformer. Detect code or information indicating. A control signal 137 is sent from the bit precision detection unit 31 to the switch 32 based on the detected information.

  At this time, when it is detected that the conversion is of integer precision, the switch 32 is switched to the selected terminal a, and conversely, when it is detected that the precision is fixed point, the selected terminal is selected. Switching connection is made on the b side. In the former case, the decoded image 134 is output through the wavelet inverse transformer 34.

  On the other hand, in the latter case (fixed-point precision), the gain adjustment unit 33 performs gain adjustment processing on the high-frequency component coefficients among the conversion coefficients, and passes the other low-frequency coefficients as they are to output 133. Get. This output 133 is then inversely transformed by the wavelet inverse transformer 34, and the inversely transformed output 135 is sent to the bit shift unit 35 and the final decoded image (level 0) is sent to the bit shift unit 35, where it is finally bit-shifted. A decoded image 136 is obtained.

Eleventh Embodiment The eleventh embodiment of the present invention presents an example of information for performing bit precision detection in the tenth embodiment described above. That is, in the tenth embodiment described in conjunction with FIG. 15, the bit precision detection unit 31 in the decoding apparatus performs wavelet transform from the encoded bit stream 130 with any bit precision in the wavelet transform apparatus on the encoding side. Although means including a means for detecting a code or information indicating whether or not it has been described has been described, in practice, there must be a mechanism for embedding the code or information in the encoded bitstream 130.

  FIG. 16 shows an example of information indicating the type of wavelet transform or bit precision in an encoded bit stream. The table shown in FIG. 16 is the above-described JPEG-2000 Part-1 FDC Table A- A code of “0000 0010” (5-3 irreversible wavelet transform) is newly added to what is described as 19. As a result, it is possible to distinguish between the integer precision type 5 × 3 filter described in the above embodiments and the fixed point precision type 5 × 3 filter.

  That is, in FIG. 16, as the wavelet transform type (Transform Type), 9-7 irreversible wavelet transform (floating point precision type 9x7 filter) of 8-bit code “0000 0000” and 5-3 reversible wavelet of “0000 0001” In addition to transform (integer precision type 5x3 filter), the code of 5-3 irreversible wavelet transform (fixed point precision type 5x3 filter) of "0000 0010" is added. The 8-bit code representing the type of wavelet transform is detected and identified by the bit precision detection unit 31 in FIG. 15, so that the decoding process or wavelet inverse transform process suitable for the bit precision after the switch 32 as described above. Can be performed.

Twelfth Embodiment According to a twelfth embodiment of the present invention, in the case of a decoding apparatus or wavelet inverse transformation apparatus having only integer precision type wavelet inverse transformation means, a fixed-point precision wavelet transformation is performed on the encoding side. Means for interrupting all decoding operations when it is detected that an encoded bitstream having been subjected to input has been input, or means for issuing an undecoded state display to the outside Is provided.

  That is, as described in the tenth embodiment, if the decoding device side is provided with a high-frequency component coefficient gain adjusting unit and a bit shifting unit after inverse transformation in advance, inverse transformation and decoding can be performed correctly. However, if these means are not provided as in the twelfth embodiment, all decoding operations are interrupted to make the decoding impossible state, or the decoding impossible state display is made. It is realistic to issue to the outside.

  In the first to twelfth embodiments of the present invention described above, the problem of providing a high-quality encoded image not only for a still image but also for a moving image using wavelet transform is as follows. The main focus is to solve the problem. That is, the Fixed-point 5x3 filter, which is a fixed-point precision improvement of the Integer 5x3 filter specified by the above-mentioned JPEG-2000 Part-1 FCD (Final Committee Draft), is the same as the Float 7x9 filter specified by the FDC. In comparison, it not only has excellent performance in terms of coding efficiency, but also has many common parts between the Integer 5x3 filter and the internal computing unit. Therefore, by sharing both circuits or calculation means, the increase in hardware is minimized without sacrificing encoding efficiency. Further, in the embodiment of the present invention, for example, an Integer 5x3 filter and a Fixed-point 5x3 filter in JPEG-2000 are shared and realized, but the present invention can be applied to other than JPEG-2000. Needless to say.

  The configuration of the embodiment of the present invention includes means for reading out and buffering an input image as necessary for wavelet transform, fixed-point precision wavelet transform means, and integer precision wavelet transform means. The fixed-point precision wavelet transform means further includes a wavelet transformer that can be shared with the integer precision wavelet transform means and a bit shifter. The sharable wavelet transformer further includes a multiplier or shift calculator, an adder / subtractor, and a register.

  According to the embodiment of the present invention having such a configuration, it is possible to suppress an increase in the overall hardware configuration by realizing the fixed-point wavelet transform unit and the integer precision wavelet transform unit with a common configuration. effective. Further, by controlling the selection of both in accordance with the image quality or the compression rate, it is possible to always realize an optimal wavelet transform unit. Therefore, mobile terminals such as mobile phones and PDAs are forced to transmit images at a low bit rate using a narrowband line, so if an integer precision wavelet transform means that is superior in terms of power saving is used, the compression rate There is an effect that it can be operated for a long time without sacrificing.

  According to the embodiment of the present invention described above, input image data is read and stored in a memory in a predetermined region, and image coding is performed by applying a horizontal or vertical wavelet transform filter to these image regions as soon as the data is stored. In this case, the above wavelet transform can be either a fixed-point wavelet transform or an integer wavelet transform. Wavelet transform can be realized.

  Further, by sharing the configuration of the fixed-point wavelet transform unit and the integer precision type wavelet transform unit, it is possible to suppress an increase in the overall hardware configuration.

  Further, by controlling the selection of both according to the image quality or the compression rate, it is possible to always realize an optimal wavelet transform unit. For example, in mobile terminals such as mobile phones and PDAs, image transmission at a low bit rate using a narrowband line is unavoidable, so if an integer precision wavelet transform means that is superior in terms of power saving is used, compression is performed. Long-term operation is possible without sacrificing rate.

  The present invention is not limited to the above-described embodiment. For example, the number of taps of a wavelet transform filter is not limited to 5 × 3, and the applied standard is not limited to JPEG-2000. . Of course, various modifications can be made without departing from the scope of the present invention.

It is a block diagram which shows schematic structure of the wavelet transformation apparatus used for the image coding apparatus used as embodiment of this invention. It is a block diagram which shows the structural example of the fixed point type | mold wavelet transformation part 3 of FIG. It is a figure which shows the example of the calculation word length of fixed point precision and integer precision. It is a block diagram which shows schematic structure (until level 3) of a normal wavelet transformation part. It is a block diagram which shows schematic structure (until level 3) of a normal wavelet inverse transformation part. It is a figure for demonstrating the zone division (division level = 3) of a two-dimensional image. It is a figure for demonstrating the calculation of the integer precision type wavelet transform filter by the side of analysis. It is a figure for demonstrating the calculation of the fixed-point precision type wavelet transform filter by the side of analysis. It is a figure for demonstrating the calculation of the integer precision type | mold wavelet inverse transform filter by the side of a synthesis | combination. It is a figure for demonstrating the calculation of the fixed point precision type | mold wavelet inverse transform filter of the synthetic | combination side. It is a figure which shows an example of the hardware constitutions for calculating the coefficient or pixel value (in the case of level 0) of the low frequency component by the side of composition. It is a figure which shows an example of the hardware constitutions for calculating the coefficient or pixel value (in the case of level 0) of the high frequency component by the side of composition. It is a block diagram which shows an example of a structure of the wavelet inverse transformation apparatus of an image decoding apparatus. It is a block diagram which shows an example of a structure of the fixed point type | mold wavelet inverse transformation part 9 of FIG. It is a block diagram which shows the other structural example of the wavelet inverse transformation apparatus of an image decoding apparatus. It is a figure for demonstrating an example of the information which shows the type of wavelet transformation in an encoding bit stream.

Explanation of symbols

  1 data read memory unit, 2,11 switching unit, 3 fixed-point wavelet transform unit, 4 integer wavelet transform unit, 5 wavelet transform unit, 6, 14, 35 bit shifter, 7 wavelet transformer, 8 wavelet inverse transform 9, fixed point type wavelet inverse transform unit, 10 integer type wavelet inverse transform unit, 12 data write memory unit, 13, 34 wavelet inverse transform unit, 31 bit accuracy detection unit, 33 gain adjustment unit

Claims (8)

In an image encoding device that encodes input image data,
Recording means for recording a plurality of lines of the image data in order from the highest line;
Wavelet transform means for performing wavelet transform processing for each of a plurality of lines read from the recording means;
Coding means for performing coding processing on the result of the wavelet transform means;
An image encoding apparatus, wherein the processing by the wavelet transform unit and the encoding unit is sequentially executed for each of a plurality of lines read from the recording unit.   The wavelet transform unit includes a fixed-point wavelet transform unit configured to include a bit shifter and a wavelet transformer, and an integer type wavelet transform unit configured to include only the wavelet transformer. The image coding apparatus according to claim 1, wherein   3. The image coding apparatus according to claim 2, wherein the wavelet transform unit of the integer type wavelet transform unit and the wavelet transform unit of the fixed-point type wavelet transform unit have the same configuration.   2. The image according to claim 1, wherein when performing lossless encoding, the integer wavelet transform unit is selected, and when performing lossy encoding, the fixed-point wavelet transform unit is selected. Encoding device. In an image encoding method for encoding input image data,
A recording step of sequentially recording a plurality of lines of the image data from the top line;
A wavelet transform process for performing a wavelet transform process for each of a plurality of lines read from the recording process;
An encoding process for performing an encoding process on the result of the wavelet transform process;
An image encoding method comprising sequentially executing the wavelet transform step and the encoding step for each of a plurality of lines read from the recording step. In an image decoding apparatus that performs decoding processing on input encoded image data,
Recording means for recording the encoded image data for each of a plurality of lines;
Decoding means for performing decoding processing for each of a plurality of lines read from the recording means;
Wavelet inverse transformation means for performing wavelet inverse transformation processing on the result of the decoding means,
An image decoding apparatus characterized by sequentially executing the processing by the decoding processing means and the wavelet inverse transformation processing means for each of a plurality of lines read from the recording means. The wavelet inverse transform means includes a fixed-point wavelet inverse transform means and an integer wavelet inverse transform means,
The fixed-point wavelet inverse transform means includes a bit shifter and a wavelet inverse transformer,
7. The image decoding apparatus according to claim 6, wherein the integer type wavelet inverse transforming means comprises only a wavelet inverse transformer. In an image decoding method for performing decoding processing on input encoded image data,
A recording step of recording the encoded image data for each of a plurality of lines;
A decoding step for performing decoding processing for each of the plurality of lines read from the recording step;
A wavelet inverse transform process for performing a wavelet inverse transform process on the result of the decoding step;
An image decoding method, wherein the decoding processing step and the wavelet inverse transformation processing step are sequentially executed for each of a plurality of lines read from the recording step.

Images