JPH0251978A - Image encoder - Google Patents

Abstract

PURPOSE:To realize a speed-up with the reduction of arithmetic circuits by executing an inverted orthogonal transformation processing corresponding to the non-zero information of orthogonally transformed data. CONSTITUTION:Since parts, where a bit assignment is zero, are contained in a large quantity in the orthogonally transformed data being quantized, the non-zero formation to indicate the part, where the bit assignment is not zero, is extracted by a non-zero extracting part 111. One of plural arithmetic means 113 to execute the inverted orthogonal transformation corresponding to the non-zero information is selected by a selecting means 115 according to the non-zero information. Thus, a useless operation to the part, where the bit assignment is zero, is evaded, the high-speed operation of the inverted orthogonal transformation processing is realized, and the restoration processing of image data compressed in an orthogonal transformation encoding system can be executed at high speed.

Description

[Detailed Description of the Invention] [Table of Contents] Field of Industrial Use Prior Art (Figures 5 and 6) Problems to be Solved by the Invention (Figures 7 and 8) To Solve the Problems Means (Fig. 1) Working Examples (Figs. 2 to 4) Effects of the Invention [Summary] Regarding an image encoding device using an orthogonal transform encoding method, an inverse orthogonal transform is performed corresponding to the content of orthogonal transform data. With the aim of realizing high-speed calculations and speeding up image data restoration processing, we have developed an orthogonal transform unit that performs orthogonal transformation of the original image data, and an orthogonal transform data that is obtained by this orthogonal transformation processing. A quantization unit that performs predetermined bit allocation, quantizes, and sends the quantized orthogonal transformation data to an image storage unit and other devices, and an inverse quantization unit that takes in this quantized orthogonal transformation data and performs inverse quantization. and an inverse orthogonal transform section that performs inverse orthogonal transform on the orthogonal transform data obtained by this inverse quantization process to obtain restored image data,
The inverse orthogonal transform unit includes a plurality of inverse orthogonal transforms that perform inverse orthogonal transforms corresponding to the non-zero information. The apparatus includes a calculation means and a selection means for selecting one of the plurality of calculation means in accordance with the non-zero information.

[Industrial Application Field] The present invention relates to an image encoding device that encodes image data using an orthogonal transform encoding method and restores the image data through decoding.

In particular, the present invention relates to an image encoding device that enables high-speed restoration processing of image data.

[Conventional technology]

In order to efficiently store huge amounts of image data, it is necessary to compress the amount of data through encoding. Various methods have been proposed for this image coding method, but predictive coding, block coding, orthogonal transform coding are the best image coding methods that have a high compression rate and less deterioration of restored images. ,
Examples include vector quantization.

Among these image encoding methods, the orthogonal transform encoding method can be said to be an excellent method that efficiently utilizes the properties of images and human visual characteristics to increase the compression rate and achieve high quality. In other words, this orthogonal transform encoding method converts the original image data into spatial frequency spectrum parameters by orthogonal transform, then quantizes it, and performs the reverse process for restoration, thereby achieving high compression rate and high quality image encoding. is made possible.

FIG. 5 is a block diagram showing the configuration of a conventional image encoding device using a cosine transform method, which is one of orthogonal transform encoding methods.

In the figure, an input unit 501 inputs an original image 503 into two
It is sampled into 56×256 pixels and sent to the compression unit 510. The compression unit 510 takes the input original image data into the block division unit 511, and divides the input original image data into, for example, 32 x 32
The data is divided into (1024) blocks, and data compression is performed for each block as partial image data of 8×8 pixels via a cosine transform unit 513 and a quantizer 515.

Generally, partial image data of NXN (N=8 in the above example) pixels is expressed as f(m, n) (m, n is 0, 1.2 degrees...
, N-1), the cosine transform unit 513 performs cosine transform based on the following equation.

-...-(1) However, u and v are 0, 1, 2. -, N-1c(u,v)
= (u=v=o)=1 (Others) The NXN two-dimensional cosine transform data F(u, v) corresponding to the NXN pixels obtained in this way is the partial image data f ( The data represents a two-dimensional spatial frequency spectrum distribution for m, n). Note that the distribution has statistical bias depending on the type of image.

Next, the quantization unit 515 assigns an appropriate number of quantization bits to the cosine transformed data F(u,v), taking into consideration the statistical bias of the data after orthogonal transformation, the sensitivity of human visual characteristics to spatial frequency, etc. Data compression is performed by allocation and quantization (encoding).

The compressed data (quantized cosine transform data) obtained here is stored in the image storage section 521. Further, the accumulated data is transferred to the restoration unit 530 in response to a predetermined operation (for example, search processing).

In the restoration unit 530, an inverse quantization unit 531 that performs processing opposite to that of the quantization unit 515 converts NXN pieces of cosine transformed data F.
'(u, v) is decoded, and then the inverse cosine transform unit 5
At step 33, inverse cosine transformation is performed based on the following equation, and restored partial image data f'(m, n) of each block is obtained.

However, m and n are 0, 1, 2. -, N-1 The restored partial image data f'(m,
n) corresponds well to the original partial image data f (m, n), and by using two colors to obtain, for example, 1024 pieces of restored partial image data for each pro-nk, the display unit 505 displays the original partial image data. A restored image 507 of 256 x 256 pixels that closely corresponds to the original image can be displayed.

FIG. 6 is a block diagram showing an example of an image database system using such an image encoding device.

Components similar to those of the image encoding device shown in FIG. 5 are designated by the same reference numerals.

In the figure, the original image 503 is captured by a scanner or other input unit 5.
01 and sent to the compression unit 510, and the image data compressed here is stored in the image database (image storage unit 521) 543 via the host computer 541. When the image data stored in the image database 543 is retrieved based on a request from a user, it is sent to the host computer 541 and the communication network (LAN, etc.) 545.
The image is transferred to each restoration unit 530 in a compressed form to improve transfer efficiency, and a restored image 507 obtained by the above-described restoration processing is displayed on the display unit 505.

[Problem to be solved by the invention]

By the way, in such orthogonal transform encoding method, (1)
As shown in equations and (2), a huge number of multiplications and additions are required for two-dimensional image data, and the processing time associated with orthogonal transformation and its inverse orthogonal transformation is long. There is a need to speed up the process.

As an example of speeding up to meet this demand, for example, in inverse orthogonal transform processing of image data, the two-dimensional inverse cosine transform shown in equation (2) can be transformed into one-dimensional in the vertical direction and one-dimensional in the horizontal direction, as shown in the following equation. By decomposing into inverse cosine transformation, high-speed processing using a high-speed calculation algorithm using butterfly calculation is possible. Note that the same applies to equation (1).

By using such a high-speed calculation algorithm, the number of calculations has increased from the order of N2 to the order of N Iog2N, and a corresponding increase in speed has been achieved in the past, but especially when multiple users as shown in Figure 6 In a system in which an image database is searched by , the frequency of restoration processing is increasing, and further speeding up of the inverse orthogonal transformation processing is desired.

Here, the state of bit allocation of the orthogonal transform data quantized by the quantization unit 515 of the compression unit 510 and the high-speed calculation algorithm used in the processing of the inverse cosine transformation unit 533 will be described.

FIG. 7 is a diagram showing an example of bit allocation of quantized cosine transform data.

In the figure, bits of 0 to 8 bits are assigned to each frequency spectrum component u(0 to 7)Xv(0 to 7). For example, 8 bits are assigned to the DC spectrum component of U=O1V=0. Note that the quantized orthogonal transformation data contains many parts where the bit assignment is "0", but this is because the correlation between nearby densities in the image is generally high, and information is contained in the low frequency spectrum. This is because they tend to concentrate.

FIG. 8 is a diagram illustrating a one-dimensional fast inverse cosine transform algorithm.

In the figure, "・" is a contact terminal, "O" is an addition operation, and the sine function value, cosine function value, and each value of "-1" on the line indicate that those values are multiplied. There is.

Here, to give an example, the output X is, and the output y
becomes , and the output 2 becomes -y, and this 2 and the other route (F'1, F'5, F'
3, F'7), the restored partial image data f'3 and F'4 can be calculated (References: W, H, CHEN, C, H, S
MITH, S.C.

FRALICK,” A Fast
Computation Algorithbmf
or Discrete Co51ne
Transform I E E Ding RA
NSACTIONS ON COMMUNICAT
IONS, Vol, C0M-25°Nα9. 19
77).

Using such a high-speed calculation algorithm, eight pieces of cosine transform data F' are generated. ~F'7, eight restored partial image data f'. ~f' can be calculated at high speed, but in this high-speed calculation algorithm, it can be said that the number of multiplications greatly influences the processing speed.

Therefore, F′. ~16 if all F'7 are non-zero
For example, if F'+ is known in advance as Oj, then the cos corresponding to F'7 is
The multiplication of (7π/16) and -5in (7π/16) is no longer necessary, and the number of multiplications becomes 14, indicating that processing speed can be increased by reducing the number of multiplications.

In addition, in order to calculate the restored partial image data f' (m, n) from the two-dimensional cosine transform data F' (LIIV), a high-speed calculation algorithm is repeated for u and m for each value of v and n. Therefore, it can be said that the impact is even greater.

The present invention realizes high-speed computation of inverse orthogonal transform processing corresponding to the content of orthogonal transform data, taking into consideration the bit allocation situation of orthogonal transform data explained above and the possibility of speeding up high-speed computation algorithms. An object of the present invention is to provide an image encoding device that enables high-speed image data restoration processing.

[Means to solve the problem]

FIG. 1 is a block diagram of the principle of the present invention.

In the figure, an orthogonal transformation unit 101 is configured to perform orthogonal transformation of original image data.

The quantization unit 103 is configured to allocate predetermined bits to the orthogonal transformation data obtained by this orthogonal transformation process, quantize it, and send the quantized orthogonal transformation data to the image storage unit and others.

The inverse quantization unit 105 is configured to take in this quantized orthogonal transform data and perform inverse quantization.

The non-zero extraction means 111 is configured to extract non-zero information indicating a portion where the bit allocation is not zero from the quantized orthogonal transform data.

The inverse orthogonal transform unit 107 includes a plurality of calculation means 113 that performs inverse orthogonal transformation corresponding to non-zero information, and a selection means 115 that selects one of the plurality of calculation means according to the non-zero information. This configuration performs inverse orthogonal transformation on orthogonal transformation data obtained through quantization processing to obtain restored image data.

[For production]

In the present invention, since quantized orthogonal transform data includes many parts where the bit allocation is zero, the non-zero extraction unit 111 extracts non-zero information indicating the parts where the bit allocation is not zero, By selecting one of the plurality of arithmetic means 113 that performs inverse orthogonal transformation corresponding to non-zero information according to the non-zero information, unnecessary operations for portions where the bit allocation is zero can be avoided, and the inverse orthogonal High-speed calculation of conversion processing is realized.

That is, by performing inverse orthogonal transformation processing using arithmetic means corresponding to the content of orthogonal transformation data, it is possible to speed up the restoration processing of image data.

〔Example〕

Hereinafter, embodiments of the present invention will be described in detail based on the drawings.

FIG. 2 is a block diagram showing an example of the configuration of main parts of the image encoding device of the present invention.

Here, an example of the configuration of the restoration unit is shown, but the other input units, compression units (block division unit, cosine transformation unit, and quantization unit) and display unit that make up the image encoding device are shown in FIG. The configuration is equivalent to each part of a conventional image encoding device.

In the figure, quantized cosine transform data transferred from the image storage section is taken into an inverse quantization section 201 of the restoration section, and predetermined cosine transform data F'(u, v) is obtained.

This two-dimensional cosine transformation data F'(u + v)
are sequentially sent to each one-dimensional inverse cosine transform unit 210 and 220 that performs inverse cosine transform in each of the vertical and horizontal directions.

Each of the one-dimensional inverse cosine transform units 210 and 220 performs inverse cosine transform using the minimum number of calculations in each direction of the cosine transform data F'(u,v) in response to predetermined non-zero information, which will be described later. A plurality of calculation units 211° to 211
) l-1,221°~221 N-.

and a multiplexer (MPX) 21 that selects one of them.
3.223, and are connected via a buffer circuit 231.

Furthermore, the quantized cosine transform data transferred from the image storage section is branched to a non-zero extraction section 2 which extracts non-zero information indicating portions where the bit allocation is not zero for each direction.
41 and each obtained non-zero information is sent to the selection control unit 2.
43.

The selection control unit 243 sends selection control signals corresponding to each multiplexer 213 and 223 in accordance with this non-zero information.

In this embodiment, the non-zero extraction means (111 in FIG. 1)
A non-zero extraction unit 241 is provided as a non-zero extraction unit 241, and each calculation unit 211° to 2
1IN-7, 221° to 221N-1 are provided, and a selection control section 243 and multiplexers 213 and 223 are provided as selection means (115 in the same).

FIG. 3 is a diagram illustrating the operation of each one-dimensional inverse cosine transform unit 210 and 220.

The operation of this embodiment will be described below with reference to FIGS. 2 and 3.

First, at the beginning of processing one block of quantized cosine transformed data, the buffer circuit 231 is initialized.

FIG. 3(a) shows the state of the two-dimensional cosine transform data F'(u,v) obtained by the inverse quantization section 201.

The hatched area corresponds to the non-zero portion of the bit allocation. Here, u and v are respectively Oll,...7,
Each one-dimensional cosine transformed data is F'. ,F'
Let l,...,F' be.

The non-zero extraction unit 241 determines the number of non-zeros in the vertical direction (V direction) for each U, and further determines the maximum number of non-zeros in the horizontal direction (U direction). That is, each number of non-zeros in the V direction 5.4.3.2, ■ for U=O1■, 2.3.4 and the maximum number of non-zeros in the U direction 5 are determined, and the selection control unit selects them as non-zero information. 2
Send to 43.

The selection control section 243 creates a selection control signal for selecting a predetermined calculation section of each one-dimensional inverse cosine transformation section 210.220 based on this non-zero information, and sends it to each multiplexer 213.223.

Note that the calculation units 211° and 221° perform calculation processing of inverse cosine transformation when only the one-dimensional cosine transformation data F'o is non-zero, and the calculation units 2111 and 2211
is cosine transformed data F'. and F′+ are non-zero, and the calculation unit 21
1? , 221. is cosine transformed data F'. ~F'7
This configuration performs arithmetic processing when all of are non-zero.

Therefore, the selection control unit 243 first selects the U=0 column (F
'. The multiplexer 213 is controlled to select the arithmetic unit 2114 of the one-dimensional inverse cosine transform unit 210 as the arithmetic unit corresponding to u=1.2).
．． 3.4, each calculation unit 2113.2
11K, 211. ．． 211° is selected.

The result of this inverse cosine transformation in the vertical direction (see FIG. 3) is temporarily stored in the buffer circuit 231.

Next, the selection control unit 243 controls the multiplexer 223 to select the calculation unit 2214 of the one-dimensional inverse cosine transformation unit 220 as the calculation unit corresponding to the maximum number of non-zeros in the horizontal direction (U direction), 5, The data in the buffer circuit 231 is inversely cosine transformed in the horizontal direction, and partial image data f'(m,
n) (FIG. 3(C)) can be restored.

In this way, by selecting the arithmetic units that perform inverse cosine transform with the minimum number of computations depending on the non-zero information of each cosine transform data, it is possible to realize speeding up as a whole.

In the embodiment described above, the non-zero information is extracted in the restoration section, but the non-zero information from the quantized cosine transform data is extracted in the compression section that encodes the image data. It may be configured such that the non-zero information is added to the accumulated data and notified to the restoration unit via the image storage unit, and the non-zero extraction unit 241 extracts the information.

Furthermore, as the selection means (115 in FIG. 1), a configuration is shown in which the multiplexers 213 and 223 are switched by a selection control signal output from the selection control section 243, but the operation of a predetermined calculation section is performed according to the selection control signal. It is also possible to enable the configuration.

Here, each one-dimensional cosine transform data F'. ~F't
The following table shows the relationship between the number of non-zero elements and the number of multiplications performed in the corresponding arithmetic unit.

As can be seen from this table, for the case where a calculation section is provided that performs inverse cosine transformation with the minimum number of calculations for each non-zero information, for example, F'. The inverse cosine transform for, F′. , F′1, inverse cosine transform for F′, , ,7, and F
'. A certain degree of speed-up can be expected even if the calculation units are integrated into four calculation units that perform inverse cosine transformation on ~F'.

Table Generally, the data compression rate is improved by changing the allocation of the number of quantization bits according to the magnitude of the change in density of the image, but the state of the bit allocation for each block is classified into four classes, and each An example of the number of multiplications in the inverse orthogonal transform process corresponding to the class is shown below.

FIG. 4 is a diagram showing the status of bit allocation classified into classes.

FIG. 4(a) shows bit allocation state B (class 1) in which the non-zero portion is the hatched portion of U=O1V=O. In this case, the number of multiplications is two in total, once each in the ■ direction and the U direction.

In FIG. 4(b), the non-zero portion is u=0. l, 2, 3, V=
Bit allocation shape B (shaded part of 0.1, 2.3)
class 2). In this case, the number of multiplications is 8+6+4+1=19 times in the ■ direction, and 8 times in the U direction.
4 to 32 times, a total of 51 times.

In Fig. 4(C), the non-zero part is u = O~7, v-〇~
7 shows the bit allocation state (class 3), which is the shaded area. In this case, the number of multiplications is 16+1 for the ■ direction.
4+12+10+8+6+4+1 to 71 times, 16X 8 to 128 times in the U direction, a total of 199 times.

In Fig. 4(C), the non-zero part is u = O ~ 7, ■ = 0 ~
7 (class 4). In this case, the number of multiplications is 16 times in each direction.
16X 8X to perform 8 times each in direction and V direction
2 to 256 times.

Note that the conventional number of multiplications for each block was 256, which is the same as in class 4, since it is not affected by the state of the non-zero portion.

Therefore, if the number of blocks per class is equal, on average (2 + 51 + 199 + 256) / 4 = 127,
Since the number of multiplications is approximately half that of the conventional method, which is 256, it is possible to significantly speed up the inverse orthogonal transform processing.

〔Effect of the invention〕

As described above, according to the present invention, inverse orthogonal transform processing corresponding to non-zero information of orthogonal transform data is performed, and speeding up is achieved by reducing the number of calculations. This makes it possible to perform high-speed restoration processing of image data, which is extremely useful in practice.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of the principle of the present invention, FIG. 2 is a block diagram showing an example of the main configuration of the image encoding device of the present invention, and FIG. 3 is a diagram explaining the operation of each one-dimensional inverse cosine transform unit. FIG. 4 is a diagram illustrating the status of bit allocation divided into classes; FIG. 5 is a block diagram showing the configuration of a conventional image encoding device; FIG. 6 is a block diagram showing an example of an image database system; The figure is a diagram showing an example of bit allocation of quantized cosine transform data, and FIG. 8 is a diagram explaining a one-dimensional high-speed inverse cosine transform algorithm. In the figure, 101 is an orthogonal transformation section, 103 is a quantization section, 105 is an inverse quantization section, 107 is an inverse orthogonal transformation section, 111 is a non-zero extraction section, 113 is an operation means, 115 is a selection means, and 201 is an inverse quantization section. conversion part, 210.220 is one-dimensional inverse cosine transformation part, 211.2
21 is an arithmetic unit, 213 and 223 are multiplexers (MPX), 231 is a buffer circuit, 241 is a non-zero extraction unit, and 243 is a selection control unit. Honjo gtl Yamari 7-o-, 70 Fig. 1 Ca) 7 sura 6ゝ 1 cho 3 shaku Hi゛..., Y 劃chi 1 mountain'' (state, Fig. (U') i child Its forgiveness Figure 7

Claims (1)

[Claims] (1) Orthogonal transformation unit (1) that performs orthogonal transformation of original image data
01), and a quantization unit (1) that allocates predetermined bits to the orthogonal transformation data obtained by this orthogonal transformation process, quantizes it, and sends the quantized orthogonal transformation data to the image storage unit and other devices.
03), an inverse quantization unit (105) that takes in this quantized orthogonal transform data and performs inverse quantization, and an inverse orthogonal transform of the orthogonal transform data obtained by this inverse quantization process to restore it. In an image encoding device comprising an inverse orthogonal transform unit (107) for obtaining image data, non-zero extraction means extracts non-zero information indicating a portion where the bit allocation is not zero from the quantized orthogonal transform data. (111), the inverse orthogonal transform unit (107) includes a plurality of calculation means (113) that performs inverse orthogonal transformation corresponding to the non-zero information, and a plurality of calculation means (113) that performs the inverse orthogonal transformation corresponding to the non-zero information. An image encoding device characterized by comprising a selection means (115) for selecting one.