JPH11103460A - Image processor, method and storage medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To efficiently compress into both a multivalued image and a binary image and to satisfactorily maintain an image quality by outputting 1st information that shows a place where a significant conversion coefficient exits and 2nd information that shows the significant conversion coefficient. SOLUTION: Input multivalued image data to a switching device 6 switches the flowing direction of multivalued image data of the device 6, a multivalued image is inputted to a discrete wavelet converter 2, a binary image is inputted to a predictive converter 3 and conversion processing is executed. Conversion coefficients after conversion of respective converters are both inputted to an area divider 4. On the other hand, a discrimination result of a discriminator in each block is also inputted to the divider 4. The divider 4 outputs division information that shows a division processing result, an actual value of the conversion coefficient and a sign as bit data. They are inputted to an entropy encoder 5 and final encoding data is outputted. Here, the divider 4 divides them into significant coefficients in part and the other insignificant coefficients and reduces bit data quantity.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to an image processing apparatus and method for efficiently compressing an image and a storage medium storing the method.

[0002]

2. Description of the Related Art In recent years, computers and networks have been remarkably developed, and various types of information such as character data, image data, voice data, and the like have been handled in computers and networks.

As for image data and audio data, those having a relatively large data amount are often handled. Therefore, processing for reducing the data amount by compressing image data and audio data has been conventionally performed.

For example, by compressing image data,
Many pieces of image information can be transmitted at high speed via a network.

[0005] In the above-mentioned image compression technology, ITU-T Recommendation T.8
A method for compressing multi-valued still images in J.1 (commonly known as JPEG) is known, and a method for compressing binary still images in ITU-T Recommendation T.82 (commonly known as JBIG) is known.

The JPEG method is suitable for compressing a natural image such as a photograph, and the JBIG method is suitable for compressing a binary image such as black and white characters.

In the conventional JPEG and JBIG systems, it is not possible to efficiently compress all images of a multi-valued image (photograph, etc.) and a binary image (character image, line image, etc.) without deteriorating the image quality. Can not.

In order to solve this problem, there is also known a method of separating a multi-value image portion and a binary image portion in one screen and compressing them by different compression methods.

[0009]

However, as described above, a method of separately compressing a multi-valued image and a binary image has not yet been established.

The present invention has been made in view of the above conventional example, and provides a method capable of efficiently compressing both multi-valued images and binary images and maintaining good image quality. The purpose is to do.

[0011]

According to the present invention, there is provided an image processing apparatus for converting an input image data by a first method to generate an M-value conversion coefficient. 1 conversion means, second conversion means for converting input image data by a second method to generate N-valued conversion coefficients such that M> N, and any one of the M-valued conversion coefficients and the N-valued conversion coefficients In the block composed of
An output unit that outputs first information indicating a position where a valid conversion coefficient exists and second information indicating the valid conversion coefficient.

[0012]

BEST MODE FOR CARRYING OUT THE INVENTION

(First Embodiment) FIG. 1 is a schematic diagram of an image processing apparatus used in a first embodiment of the present invention.

First, the processing flow of each unit will be briefly described.

In FIG. 1, image data input from an input unit in the figure is multi-valued image data having a predetermined number of bits per pixel.

In the present embodiment, the multi-valued image data having a depth of 8 bits per pixel includes not only image data indicating a photograph, a picture, and the like suitable for the multi-valued expression but also a binary expression. It is assumed that image data indicating a character, a line drawing, and the like to be used is also included.

Therefore, in the present embodiment, in order to efficiently encode image data representing an image in which these images are mixed, the former image is encoded as multi-valued image data, and the latter image is encoded as multi-valued image data. Encoding efficiency is improved by encoding as value image data.

Subsequently, the input multi-valued image data is input to the block divider 7. The block divider 7 divides the multi-valued image data into multi-valued image data corresponding to a block having a predetermined size, for example, a vertical Hb pixel and a horizontal Wb pixel as shown in FIG. Output.

The output multivalued image data is input to the discriminator 1 shown in FIG. The discriminator 1 determines whether the multi-valued image data represents a binary image or a multi-valued image by detecting the number of gradations of one pixel. The image data determined by the determiner 1 is also input to the switch 6 at the same time.

If the multi-valued image data input to the switching unit 6 is multi-valued image data to be expressed as a binary image based on the result of the discrimination by the discriminator 1, it is output to the predictive converter 3. If the data is multivalued image data to be represented as a multivalued image, the data is output to the discrete wavelet transformer 2. That is, the direction in which the multivalued image data flows in the switching unit 6 is switched based on the determination result of the determination unit 1.

The multivalued image data input to the discrete wavelet transformer 2 or the predictive transformer 3 is subjected to a predetermined conversion process. Each of the conversion processes will be described later.

The transform coefficients converted by the respective converters are input to the area divider 4 together. On the other hand, the discrimination result for each block by the discriminator 1 is also input to the region divider 4.

In the area divider 4, the input transform coefficients are compared with a predetermined threshold value, and are sequentially divided into smaller areas based on the comparison result. This division processing will also be described later.

The area divider 4 outputs, as bit data, division information indicating the result of the above-described block division processing, the actual value of the transform coefficient, and the sign. The division information and the bit data are input to the entropy encoder 5 at the subsequent stage.

The entropy encoder 5 entropy-encodes the input bit data and outputs final encoded data.

Next, a detailed operation of each block will be described.

The block divider 7 divides the input multi-valued image data into blocks of a predetermined size, and sequentially outputs the multi-valued image data for each divided block.

Here, as shown in FIG. 2, it is assumed that multivalued image data divided for each rectangular area of horizontal Hb pixels and vertical Wb pixels is output.

The discriminator 1 examines each bit of the input multi-valued image data, and determines whether the multi-valued image data is to be represented as a binary image (character, line drawing, etc.) or not. It is determined whether the image should be expressed as a natural image such as a photograph or a painting.

FIG. 3 is a conceptual diagram showing the operation of the discriminator 1. The input image data is compared by the comparator 101, and the result is input to the counter array 102.

Here, each reference value of the comparator 101 corresponds to the pixel value of the input multi-valued image data, and the number L
Is set equal to the maximum number of levels (for example, 256) that can be predicted for multi-valued image data.

The comparator 101 outputs 1 when the result of the input image data is equal to each reference value, and the result is input to a corresponding counter of the counter array 102. On the other hand, the image data input to the comparator 101 is counted for each pixel, and when all the pixels in the block have been compared, the contents of the counter array 102 are output to the determination unit 103 and The contents are reset. The determination unit 103 scans the content of each counter C (i) and counts the number N _C of counters where C (i) ≠ 0 (i = 0,..., L−1) *** (1). Then, if N _C = 2 *** (2), it is determined that this image data should be represented as a binary image, and if N _C > 2 *** (3), This image data is determined to be represented as a multi-valued image.

The result of this determination is output to the switch 6 and the area divider 4.

If the input multi-valued image data is to be represented as a multi-valued image based on the discrimination result input from the discriminator 1, the switcher 6 sends it to the discrete wavelet transformer 2. If the input multi-valued image data is to be represented as a binary image, the multi-valued image data is switched to be output to the predictive converter 3.

(Encoding process as multi-valued image) Next, when the input multi-valued image data is to be represented as a multi-valued image, that is, when the multi-valued image data is to be processed as it is (discrete The case where the wavelet transformer 2 is operated) will be described.

The discrete wavelet transformer 2 performs a discrete wavelet transform on the input multi-valued image data and decomposes the data into a predetermined number of frequency bands (hereinafter referred to as sub-bands).

FIG. 4 shows a schematic diagram (a) for executing a discrete wavelet transform and a conceptual diagram (b) of a subband generated by this transform process.

In FIG. 4A, the input multi-valued image data x passes through either the low-pass filter H0 or the high-pass filter H1 in each of the horizontal and vertical directions, as shown in FIG. Each time the signal passes through the filter, it is decomposed into a plurality of frequency bands by performing subsampling.

FIG. 4B shows multi-valued image data corresponding to horizontal Wb pixels and vertical Hb pixels, as shown in FIG.
It shows a processing result obtained by performing step conversion.

In this embodiment, the size of the block shown in FIG. 4B corresponds to the size (Wb × Hb) of the image divided by the block divider 7.

It is assumed that each filter used in the present embodiment is a reversible filter capable of completely reproducing data at the time of image decoding.

For example, a result r obtained by performing a low-pass filter H0 and sub-sampling on the multi-valued image data x
Is represented by the relational expression of (4), and the high-pass filter H1,
And the result d of the subsampling is expressed by the relational expression of (5).

R (n) = << (x (2n) + x (2n + 1)) / 2 >> *** (4) d (n) = x (2n + 2) -x (2n + 3) + << (-r (n) + R (n + 2) +2) / 4 >> *** (5): << X >> is the largest integer not exceeding X

The discrete wavelet transformer 2 sequentially repeats the filtering process and the sub-sampling in the horizontal and vertical directions as shown in FIG. 4 (a), and converts each input block image into a plurality of sub-bands. Divide sequentially.

FIG. 4 (b) shows the names of the sub-bands obtained in FIG. 4 (a) and the spatial positional relationship. Each sub-band has a corresponding conversion coefficient ( Frequency component). The transform coefficients obtained by the discrete wavelet transformer 2 are output to the subsequent region divider 4 for each subband.

In this embodiment, the LL sub-band is output first, HL3, LH3, HH3 is output, HL2, LH2, HH2 is output, and H
L1, LH1, and HH1 are output. By doing so, the decoding side can decode in order from the low frequency component (LL), and it is possible to recognize the outline of the image earlier.

A domain divider 4 that inputs a transform coefficient (multi-value) from the discrete wavelet transformer 2 outputs a coefficient having a value within a certain range (represents a multi-value transform coefficient by a bit plane). In this case, only the coefficient in which the largest significant coefficient bit exists in a bit plane (n _max or more to be described later) indicating a certain range of values is extracted, and information of each extracted coefficient is extracted as bit data. Output as a column.

FIG. 5 is a conceptual diagram for explaining the operation of the area divider 4.

The region divider 4 converts the transform coefficients in each sub-band into a bit data sequence according to the procedure shown in the flowchart of FIG. Hereinafter, description will be made with reference to this flowchart.

First, each input sub-band is shown in FIG.
Although the size is not always the same as in (b), the input subband itself is used as the target block T as shown in FIG.
Is determined.

Next, the maximum value c _max for all the transform coefficients c _ij in the target block T is obtained. And _{further, n max = log 2 (|} c max |) *** seeking to become n _max (6), sets the initial value of the target bit plane number n.

In S500, the currently set number n
The threshold value Th for the target block T is obtained by the equation (7) using the value of.

Th = 2 ⁿ *** (7)

In step S501, if the current number n is smaller than the predetermined minimum bit plane number n _min , the processing of the target block T ends.

S502, S503, S504, S505
, If the maximum value c _max of the coefficient in the target block T is | c _max | ≧ Th *** (8) for the current number n, it is determined that the target block T has a significant conversion coefficient, and S506 is performed. Proceed to.

On the other hand, if the equation (8) is not satisfied, S50
Proceeding to 3, the bit data “0” is output to indicate that no significant transform coefficient exists in the target block T.

Subsequently, in steps S504 and S505, the value of the threshold value Th is set to Th / 2, and the value of n is reduced by one (the height of the bit plane serving as the threshold value is reduced by one), and the process returns to step S501. .

In S506, bit data "1" indicating that a significant transform coefficient exists in the target block T is output.

In step S507, it is determined whether or not the current target block T is smaller than a predetermined size (in this embodiment, two pixels both vertically and horizontally). If the result of this determination is that the target block T is smaller than or equal to the predetermined size, the flow proceeds to S508,
If it is larger than the predetermined size, the process proceeds to S509.

In S508, the n-th coefficient corresponding to all the coefficients in the target block T (four coefficients in the present embodiment)
The bit data from the bit plane to the _nmin bit plane and the sign bit of the coefficient are output.

[0060] In S509, divided into four target block the T _m of a smaller size target block T, repeated S500 subsequent processing for the target block the T _m of each of them.

The above procedure will be described with reference to an actual example. FIG.
(A) is a target block T (FIG. 4
(B). The target block T in this case is composed of 8 × 8 transform coefficients.

In this target block T, one significant conversion coefficient is indicated by hatching (the value is assumed to be “40”), which is the maximum value in the block. The other conversion coefficients are not significant, and the value is “0”.

[0063] Further, when the sequential splitting process the target block T to the target block T _m according to the procedure of FIG. 5 (e), the would be divided in the order of (a) → (b) → (c) As a result, a bit data string as shown in (d) is output.

In the above processing, n = 5 and n
_This is the case where _min = 3.

By executing this processing, a bit data string as shown in (d), which can represent the position and coefficient value of each coefficient, can be output from a block composed of a plurality of multi-valued coefficients.
The arrows shown in (d) indicate the order in which the bit data is generated by this processing and the order in which the bit data is output to the subsequent stage. Bits are output in the order from the arrow.

Hereinafter, this processing will be described in detail.

At the time point (a), first, it is detected whether or not there is a coefficient having a significant bit on the fifth bit plane. At this time, it can be seen that there is only one such coefficient (significant coefficient).

Subsequently, in the state (a), since the position of the significant coefficient is not determined, T is set to 4 as shown in step S509.
To divide. As a result, the target blocks T0 to T3 shown in FIG. At this time, bit data “1” indicating that the target block T has been divided into four is output.

Subsequently, the same processing is performed for each target block from T0 to T3. That is, the processes from step S500 are also performed for T0 to T3.

First, the above processing is performed on the target block T0. After it is determined in step S502 that "there is a significant coefficient", the target block T0 is further divided into four parts T00 to T03. At this time, bit data “1” indicating that the target block T0 has been divided into four is output.

Since T0 is further divided into four parts, the processing from S500 is first performed on each block of T00 to T03 before processing T1 to T3.

When the above processing is performed on the target block T00, the process proceeds to S508 because the size of the target block T00 is 2 × 2 in S507. Therefore, for T00, the data of the fifth to third bits corresponding to each coefficient in the block, that is, 3-bit data “000”, “000”, “00” indicating the presence or absence of a significant coefficient
1-bit data "0", "0", "0", "0" indicating the sign of each coefficient are output in the order of FIG. The above corresponds to the arrow.

The processing of T00 is completed as described above.
01 is performed. For each transform coefficient in T01, there is no significant coefficient from the third bit to the fifth bit. Therefore, for n = 3, 4, and 5, step S502 and steps S503 to S505 are repeated, and the number n becomes 3
"0" indicating "no significant coefficient" is output until the value becomes less than "0". Subsequently, the processes of T02 and T03 are performed in the same manner, so that “000”, “000”, and “000” are output. The above corresponds to the arrow.

Since the processing of T0 (T00 to T03) is completed, T1 to T3 are subsequently processed. Since there is no significant coefficient in the third to fifth bits in any of the conversion coefficients in T1 to T3, "000", "00"
"0" and "000" are output. The above corresponds to the arrow.

The bit data output by the above processing is output to the entropy encoder 5 at the subsequent stage.

(Encoding Processing as Binary Image) Next, the discriminator 1 is to input the multi-valued image data to be represented as a binary image, that is, to process it as binary image data. The case (when the predictive converter 3 is operated) will be described.

If it is determined that the input multi-valued image data is to be represented as a binary image, the multi-valued image data is output to the predictive converter 3 by the switch 6. The predictive converter 3 converts multi-valued image data input for each pixel into prediction error data as a binary image.

FIG. 6 is a conceptual diagram illustrating the operation of the predictive converter 3. In the figure, the multi-valued image data x is input to the pixel unit from the switching device 6, and converted into binary image data x ₁ for each pixel in the binary conversion unit 300, and outputs.

[0079] Pattern matching unit 301 extracts the pixel indicated by the x ₁ peripheral pixels (10 pixels shown in FIG. 6 to be described later (b)) of (a coded pixel). At the same time,
The template 302 storing all possible combinations of these peripheral pixels and information indicating the combinations (corresponding to a state S (i) described later) is referred to.

As a result, the state S (i) corresponding to the extracted actual peripheral pixels is determined and output to the subsequent stage. This is used for generating a predicted value when encoding the input encoding target pixel.

[0081] FIG. 6 (b) shows the positional relationship between the peripheral pixels is extracted to generate a coded pixel x ₁ and state S (i) to be the current coded. A0 in the figure
To a9 are pixels that have already been encoded.

The state S (i) will be specifically described. Since the input binary image data x ₁ is image data of 1 bit for each pixel, each pixel value is either 0 or 1. Therefore, there are 1024 possible combinations of values of the surrounding pixels a0 to a9 of the encoding target pixel.

This is the state S (i) indicating the state of the pixel values of the surrounding pixels of the actual pixel to be encoded, where i is 1
It means that it corresponds to the ith combination in 024 ways.

In this embodiment, since x ₁ is input to the pattern matching unit 301 in the form of a block of Hb × Wb, pixels to be coded at the left end, the top end, and the right end of this block are input. In some cases, the peripheral pixels of the ten pixels do not exist. In the present embodiment, processing is performed assuming that “0” exists in peripheral pixels that do not exist in order to cope with such a situation.

The state S (i) is input to the predictor 303,
Predictor 303 outputs one of the predicted value x ₂ which is stored in association with each state S input (i) with respect to comparator 304. That is, the predictor 303 is a look-up table. Note that the predicted value x2 associated with each state S (i) is an optimum value obtained by analyzing a binary sample image in advance.

On the other hand, the binary image data x ₁ is
Is also entered. That is, the comparator 304 stores the data x ₁ of a certain encoding target pixel and this encoding target pixel as 1
0 predicted value x ₂ which is predicted based on the surrounding pixels is to be entered simultaneously.

The comparator 304 compares the data x _{1 of the} pixel to be coded with the predicted value x ₂ , outputs “0” as the prediction error e when they match, and outputs the prediction error e when they do not match. "1" is output as the error e. This process is sequentially performed on the data x ₁ (x) of all the input pixels.

Since the block divider 7 outputs image data to the discrete wavelet transformer 2 or the predictive converter 3 in units of blocks of a predetermined size, the processing of the predictive converter 3 is performed in the predetermined size. (Hb × W
It is completed for each predetermined size of b). Therefore, by configuring the prediction error e output for each block of each predetermined size in the same relationship as the actual position of each pixel to be encoded, a map E of Hb × Wb size is configured.

Next, the map E obtained by the predictive converter 3
Is output to the region divider 4. The region divider 4 treats the map E in the same manner as the sub-band output from the discrete wavelet transformer 2, and is processed using substantially the same steps, and indicates a division result when the map E is divided. The information, the actual prediction error value of the transform coefficient, and the sign are output as bit data.

However, unlike the transform coefficients in the subband output from the discrete wavelet transformer 2, the prediction error value in the map E is limited to 0 or 1. Accordingly, the region divider 4 determines whether the result of the determination input from the discriminator 1, that is, whether the input transform coefficient constitutes a subband generated from the discrete wavelet transformer 2, is a map generated from the predictive converter 3. Based on the information indicating whether or not E is included, the area division processing and the switching of the bit data string output method are performed.

More specifically, in the case where the sub-bands whose transform coefficients are input from the discrete wavelet transformer 2 are divided into regions and a bit data string as shown in FIG. When the map E input from 3 is divided into regions to output bit data strings, the following changes are made. (1) At the time of initialization, the division target area T is set to the whole map E, n _max = 0 *** (9), and the target bit plane number n and the minimum bit plane number n _min are set to 0. (2) Since it is sufficient to divide a block consisting of a 1-bit prediction error coefficient instead of dividing a block consisting of multi-valued transform coefficients, the presence or absence of a significant coefficient is indicated as shown in FIG. Instead of outputting 3 bits and 1 bit indicating the sign, data indicating 1 bit of the prediction error is output. For example, if T shown in FIG. 5A is a map E (a hatched area c3 indicates a prediction error value “1” and other areas indicate “0”), the output bit data strings are the same. In FIG. 4D, c0 = c1 = c2 = 0 c3 = 1 T01 = T02 = T03 = 0 T1 = T2 = T3 = 0. That is, the bit data string “11” is arranged in the order of the arrows.
1000100000 "is output.

Except for the above two points, it is basically the same as the case where the multi-valued conversion coefficient is divided into regions, and the description is omitted.

The bit data string output in this manner is encoded by the entropy encoder 5 and output as final encoded data.

(Data Form) FIG. 7 shows a conceptual diagram of the data form of the coded data generated by the encoding as the multi-valued image or the encoding as the binary image described above.

FIG. 7A shows the entire structure of encoded data finally output by the image processing apparatus according to the present embodiment, and is composed of header information and a code stream.

The header information includes the size of the coded image, the block size (Hb × Wb) obtained by dividing the image, and other information necessary for decoding, as shown in FIG. It consists of geometry information and the number of levels of discrete wavelet transform (DWT).

The code stream is composed of divided blocks (Hb × Wb) as one unit, as shown in FIG. 9C, and the unit includes a block header. I have.

The block header includes a flag indicating whether the corresponding block has been encoded as a multi-valued image or a binary image.

Further, when this block is coded as a multi-valued image, the block header contains the highest bit plane of the multi-valued bits indicating the transform coefficients in the block, where the significant coefficient exists. The number n _max is included. That is, n _max may be different for each block encoded as a multi-valued image. In this embodiment, n _min is common to each block to be encoded as a multi-valued image. However, if n _min information is also included in encoded data as in n _max , multi-valued image it is possible to change the n _min for each block to be encoded as. If this block is encoded as a binary image, n
_No information on _max and n _min is required.

FIGS. 7 (d) and 7 (f) show the block divider 7
3 shows a configuration of a code stream corresponding to each block divided by.

FIG. 7D shows the structure of a code stream when this block is encoded as a multi-valued image. The code stream is further formed for each sub-band of the discrete wavelet transform.

FIG. 7E shows a data format corresponding to one sub-block among the sub-blocks shown in FIG. 7D, and all the sub-blocks have the same data format. Note that LL to HH1 have the same names as the subbands shown in FIG.

In FIG. 7 (e), the contents are such that two types of bit streams of divided bits and coefficient bits appear alternately as shown in the figure. Here, the division bits are the bits output in S503 and S506 in the flowchart of FIG. 5, and the coefficient bits are the bits output in S508.

FIG. 7F shows the structure of a code stream when this block is encoded as a binary image. This code stream has a form in which divided bits and prediction error bits are alternately arranged.

(Decoding Process) Next, a method of decoding encoded data generated by the encoding method of the present embodiment will be described.

FIG. 8 is a block diagram showing a decoding device for decoding the encoded data generated by the encoding method according to the present embodiment. The details will be described below with reference to FIG.

The encoded data input via the transmission path or the storage medium is decoded by the entropy decoder 13 and the header information analyzer 8 determines the size of the original image,
Necessary information such as a block size is read and stored as information necessary for decoding.

The block header (see FIG. 7 (c)) in the coded data is read, and it is determined whether the sequentially read blocks are coded as multi-valued images or coded as binary images. , And outputs the result of the determination to the switch 6. Further, for a block encoded as a multi-valued image, n _max is read and stored.

Next, the code stream read in units of blocks (see FIGS. 7C to 7F) is input to the area divider 9. The region divider 9 converts the divided bits and the coefficient bits or the prediction error bits obtained by dividing the block at the time of encoding, into data immediately before being input to the region divider 4 in FIG. 1 (a sub-block or a map E composed of transform coefficients). It is for restoring.

The operation of the area divider 9 when a certain block indicated by the input coded data is coded as a multi-valued image will be described below with reference to the flowchart of FIG. In the present embodiment, it is assumed that the order of encoding (region division) a plurality of subbands on the encoding side is determined.

First, an area of a corresponding subband is secured from a code stream corresponding to a certain subband.
, The coefficients in the area are all initialized to 0, and n _max obtained from the block header is set as the initial value of the target bit plane n.

Next, in S900, if n is smaller than the predetermined minimum bit plane number n _min , the target block T
Is completed.

In step S901, one divided bit is read from the code stream.

In steps S902 and S903, the value of the read divided bit is checked. If the value is 0, n is decremented by 1 and the process returns to step S900.

In S904 and S905, when the current target block T has a predetermined size, for example, 2 pixels or less, coefficient bits are read from the code stream, and all the coefficients in the area are restored.

In S906, the current target block T
Is divided into four small areas Tm.
The same processes as those after 0 are sequentially performed.

The above procedure will be described in detail with reference to actual examples. FIGS. 9A to 9E show a restored state of the target block T and a code stream (bit data string) relating to the target block T sequentially input.

Here, the arrow "" shown in FIG.
Represents the order in which the bit data string is actually input, and is input bit by bit in the order of ',', '.

In the initial state, as shown in (a), all the target blocks T are initialized to 0 and are not divided. Here, in step S901, the first 1-bit division information “1” of the arrow ′ is input. This gives T
Are recognized as having significant coefficients for the current bit plane n, and in step S906 block T
It is divided into two small blocks T0 to T3. This is Figure 9
This is the state shown in FIG.

Next, restoration processing of T0 is performed. First, in step S901, since the second bit (divided bit “1”) of the arrow ′ is input, it can be determined that a significant pixel exists in the transform coefficient in T0. Therefore, block T
0 is further divided into four small blocks T00 to T03 (see FIG. 9C).

Here, since the first small block T0 is further divided into small blocks, the restoration processing of T00 to T03 is preferentially performed.

First, for block T00, step S
Processing starts at 900. Here, in step S901, the divided bit indicated by the third bit (divided bit “1”) of the arrow ′ is input, and it can be determined that a significant transform coefficient exists in T00. However, since the size of T00 corresponds to two pixels in the vertical and horizontal directions, the process proceeds to step S905 without further division, and the remaining bit data string indicated by the arrow 'is input.

Here, since n _max = 5 is known from the information included in the encoded data and the value of n _min is also known in advance, it is possible to determine how many bits each transform coefficient is represented. I have. Therefore, the first three bits of the input bit data string are input as values indicating the third to fifth bits of c0, and the next one bit is input as the sign of the conversion coefficient. Thus, the conversion coefficient “40” of c0 is restored. Similarly, data is input for c1 to c3 to restore the conversion coefficient “0” (see FIG. 9D).

Now that block T00 has been restored,
Subsequently, the bit data string input by the arrow 'is T01
It can be seen that it is for restoring from.

Since the first three bits input for restoring the block T01 are "000", it can be seen that there is no transform coefficient having "1" in the fifth to third bits of T01. Therefore, the conversion coefficient in block T01 is restored as 0. Similarly, T02 and T03 are processed to restore the transform coefficient 0 in the block.

Further, from the bit data string indicated by the arrow ',
T1 to T3 are also restored, and the transform coefficient “0” is restored in the block.

The above operation is the same when restoring the map E coded as a binary image. However, in this case, the difference is that the initial value of T is the entire block, and that n _max is initialized to 0 as in the encoding, and no bit of the sign is included. As for the restoring process, only the reverse process to the encoding process needs to be performed, so the details are omitted.

The transform coefficients corresponding to each block restored as described above are output to the switch 7.

The switching unit 6 shown in FIG.
The restored discrete transform coefficient block is converted into an inverse discrete wavelet transform based on information indicating whether the restored transformed coefficient is encoded as a multi-valued image or a binary image. 11 or to the inverse prediction converter 11.

Hereinafter, a case where the restored transform coefficients are encoded as a multi-valued image will be described.

The inverse discrete wavelet transformer 10 combines the restored and input subbands into the original block configuration shown in FIG. 4B. Then, it is inversely converted to a value before being subjected to the filtering processing in FIG. This inverse transformation is performed by repeatedly performing up-sampling and low-pass and high-pass filters, and the filter coefficients at this time satisfy the reconstruction conditions with respect to those used at the time of encoding. For example, for the filters shown in equations (4) and (5), the image data sequence x is p (n) = d (n-1)-<< (-r (n-1) + r (n + 1)) / 4 >> *** (10), x (2n) = r (n) + << (p (n) +1) / 2 >> **** (11) x (2n + 1) = r (n)-<< (p (N)) / 2 >> *** (12): << X >> is calculated from the largest integer not exceeding X. As a result, the transform coefficients in the block are inversely transformed into multi-valued image data, and are sequentially output to the block synthesizer 12.

On the other hand, a case where the restored transform coefficients are encoded as a binary image will be described.

In this case, each of the restored 1-bit transform coefficients is input as a prediction error map E.
In the inverse prediction converter 11, a prediction error e forming the map E
Are sequentially input, and the pixel values are restored using the same template as that shown in FIG. 6 and the already restored pixel values.
As for the details, a process reverse to the above-described encoding process may be performed, and thus the detailed description is omitted. The restored binary image is subjected to binary / multi-level conversion to form multi-level image data.

The block combiner 12 combines the block of the multi-valued image data input from the inverse discrete wavelet transformer 10 and the block of the multi-valued image data input from the inverse predictive transformer 11 to obtain the original image. Restore and output data.

According to the above embodiment, a case where an area to be encoded as multi-valued image and an area to be encoded as binary image data are mixed in an image indicated by input multi-valued image data In addition, encoding can be performed efficiently. That is, the redundancy of the spatial region of the original image is removed, and the data is spatially biased. Since the respective transform coefficients obtained by the discrete wavelet transform as the multi-valued image and the predictive transform as the binary image can be concentrated at spatially skewed positions, in the subsequent area divider 4, It can be split into significant coefficients and most other non-significant coefficients. Therefore, the amount of bit data output from the area divider 4 can be reduced.

In the above embodiment, the resolution of the image restored from the final coded data is assumed to be the same as that of the original image. However, at the time of restoration, the original image is first restored to a low resolution. It is also possible to use so-called progressive coding, which can be restored to a high resolution later.

(Second Embodiment) In the following second embodiment, a case will be described in which the first embodiment is changed to progressive coding.

The basic configuration of the second embodiment is the same as that of the first embodiment, and coding is performed using the configuration of FIG. Therefore, detailed description of each part is omitted.

The difference from the first embodiment lies in the method of forming encoded data (bit data string) and the order of transmission.

FIG. 10 shows the configuration of a bit data string corresponding to each subband when encoding as a multi-valued image.

FIG. 14A shows 16 blocks (each Wb × H surrounded by a broken line) divided by the block divider 7.
3 (b). Each block indicated by oblique lines is a block encoded as a binary image, and the remaining blocks are blocks encoded as a multi-valued image. FIG. 11A shows a state after the discrete wavelet transform has already been performed. That is, each Wb ×
In the Hb block, the smallest block is the subband L
L.

In the first embodiment, as shown in FIG. 7D, each sub-band LL to HH1 in one block is one.
In this embodiment, each of the levels “LL, HL3, LH” after the discrete wavelet transform is performed.
3, HH3 "(level 3)," HL2, LH2,
HH2 ”(assumed to be level 2),“ HL1, LH1, HH
1 "(referred to as level 1). For example, in (a-1), the discrete wavelet transform level 3
A bit data string obtained by encoding only the sub-band is extracted and output by performing a region dividing operation P (a-
2).

The decoding side that inputs the bit data string of level 3 performs a region dividing operation P-1 at the time of decoding on each bit data string to restore the coefficients, and combines them into one to perform discrete wavelet transform. (A-3). Here, by performing inverse discrete wavelet transform for each block, the original image is
An image reduced to / 4 can be restored.

Subsequently, on the encoding side, as shown in (b-1), a bit data string obtained by encoding only the subband of level 2 is extracted and output by performing a region division operation P (b-2). ). The decoding side that receives the level 2 bit data string forms a transform coefficient area as shown in (b-3). Here, the data of conversion level 3 is already (a-
Since it is restored as in 4), it is used together with each block consisting of level 2 (c), and inverse discrete wavelet transform is performed. As a result, an image having twice the resolution or twice the size of the image restored in (a-4) can be restored (b-4).

Subsequently, the encoding side outputs a bit data string in which only the level 1 subband is encoded, and the decoding side performs the same processing as described above, so that the same as the original image is finally obtained. An image having a resolution or size can be restored.

In the above description, the case of encoding as a multi-valued image has been described. However, the case of encoding as a binary image can also be progressively encoded.

FIG. 11 is a diagram for explaining this method.

FIG. 11A shows a map E including the prediction errors output from the prediction converter 3. The result of converting this map E into a bit data string by the region divider 4 is shown in FIG. (B) shows a transition in which the initial target block T is divided.

In FIGS. 17A and 17B, the portion where the prediction error is 1 is represented by a black area. Hereinafter, a method of generating bit data for performing progressive encoding will be described.

The target block T in the initial state is composed of 8 × 8 pixels. In step S502 in FIG. 5, it is checked whether or not a significant conversion coefficient (prediction error “1”) exists in this block T. Here, if it is determined that there is a significant conversion coefficient, a bit “1” indicating that the block T is further divided is output. In this state, T
Because it is not possible to determine where in the
At this point, it can be considered that all the prediction errors of T are 1. This is the leftmost state D0 (T) in FIG.

Next, in step S507, the size of T is checked. In S509, area division is performed.
Is divided into T0 to T3. Furthermore, these blocks T0
About T3, the processing described in the first embodiment is performed, and those determined to be significant (in this case, T0 and T3) are further divided into small blocks. By performing this division processing recursively, the state of D3 in FIGS.
That is, a bit data string for restoring the resolution or size of the original image can be generated.

Here, in the present embodiment, unlike the flowchart shown in FIG.
6, without actually outputting the bit data in S503,
Instead, a tree structure of bits as shown in FIG. 11C is created. In (c), the part written as the prediction error has no branching, because the area has reached the minimum size, so that the division is not performed and the bit data itself of the prediction error is generated.

Here, the output from the area divider 4 to the entropy encoder 5 is based on a node of a predetermined depth in the tree structure as a unit, that is, a portion surrounded by a frame in FIG. It is performed as. This is D1-D3
Then, the structure of the generated bit data string is as shown in FIG.
It becomes as shown in.

FIG. 17B shows a state in which D0 corresponding to each block divided by the region divider 7 is output to the decoding side first, and only D0 is decoded. , A map E as shown in FIG. Each block indicated by a broken line here corresponds to each block divided by the region divider 7.

When only D0 is decoded, each dashed block can express either black or white, and a very low-resolution image is restored.

Subsequently, D1 is output from the encoding side. On the decoding side, an image as shown in (c) can be restored from D1 and D0 previously input.

Similarly, by sequentially inputting D2 and D3, the restored image sequentially changes from low resolution to high resolution, and finally the original image as shown in (d) can be restored.

As described above, according to the second embodiment,
In addition to the effects of the first embodiment, it is possible to perform progressive encoding such that the restored image on the decoding side is sequentially changed from low resolution to high resolution, and there is little encoded data on the decoding side. In both cases, it is possible to grasp the outline of the original image.

The present invention is applicable not only to coding by switching between binary and multivalued, but also to coding by switching between N and M values.

The present invention includes not only the case where the image data is automatically discriminated by the discriminator 1 of FIG. 1 but also the case where the switch 6 is manually switched.

The present invention may be applied as a system that connects and uses the units required for the encoding / decoding processing, or may be applied as a single device.

Further, the present invention is not limited to only the apparatus and method for realizing the above-described embodiment, but includes a computer (CPU or MP) in the above-described system or apparatus.
U), a program code of software for realizing the above-described embodiment is supplied, and the computer of the system or the apparatus operates the various devices according to the program code to realize the above-described embodiment. It is included in the category of the present invention.

In this case, the program code itself of the software realizes the functions of the above-described embodiment, and the program code itself and means for supplying the program code to the computer, specifically, A storage medium storing the above program code is included in the scope of the present invention.

As a storage medium for storing such a program code, for example, a floppy disk, hard disk, optical disk, magneto-optical disk, CD-ROM, magnetic tape, nonvolatile memory card, ROM and the like can be used.

In addition to the case where the computer controls various devices in accordance with only the supplied program code to realize the functions of the above-described embodiment, the program code operates on the computer. Such a program code is included in the scope of the present invention even when the above-described embodiment is realized in cooperation with an OS (Operating System) or other application software.

Further, after the supplied program code is stored in a memory provided in a function expansion board of a computer or a function expansion unit connected to the computer, the function expansion board or the function is stored based on the instruction of the program code. The case where the CPU or the like provided in the storage unit performs part or all of the actual processing, and the above-described embodiment is realized by the processing is also included in the scope of the present invention.

[0167]

As described above, according to the present invention, both an image to be encoded as a multi-valued image and an image to be encoded as a binary image can be efficiently compressed and the image quality can be reduced. Can be kept good.

In particular, according to the present invention, efficient compression is possible by providing an image compression method completely different from the conventional one.

Further, in the apparatus for executing the above-mentioned image compression method, a part of the apparatus is shared, so that the cost of the apparatus can be reduced, and the binary image data is multi-valued image data. Can be compressed using almost the same algorithm.

[Brief description of the drawings]

FIG. 1 is an overall view of an image processing apparatus on an encoding side.

FIG. 2 is a diagram for explaining a block dividing method of a block divider 7;

FIG. 3 is a schematic explanatory diagram of a discriminator 1;

FIG. 4 is a schematic explanatory diagram of a discrete wavelet transformer 2.

FIG. 5 is a schematic explanatory diagram of a region divider 4;

FIG. 6 is a schematic explanatory diagram of a predictive converter 3.

FIG. 7 is a diagram showing a data format of encoded data.

FIG. 8 is an overall view of an image processing apparatus on the decoding side.

FIG. 9 is a diagram showing a state of image restoration on the decoding side.

FIG. 10 is an explanatory diagram of progressive encoding and decoding of a multilevel image.

FIG. 11 is an explanatory diagram of progressive coding of a binary image.

FIG. 12 is an explanatory diagram of progressive decoding of a binary image.

[Explanation of symbols]

 1 Discriminator 2 Discrete Wavelet Transformer 3 Predictive Transformer 4 Area Divider 5 Entropy Encoder 6 Switch 7 Block Divider

Claims (12)

[Claims] 1. A first conversion means for generating an M-value conversion coefficient by converting input image data according to a first method, and M> by converting input image data according to a second method.
A second conversion means for generating an N-valued conversion coefficient of N, and an effective conversion coefficient in a block constituted by either the M-valued conversion coefficient or the N-valued conversion coefficient input by the input means. An image processing apparatus, comprising: first information indicating a position at which an image exists, and output means for outputting second information indicating the effective conversion coefficient. 2. The image processing apparatus according to claim 1, further comprising entropy encoding means for entropy encoding the first information and the second information output from the output means in a mixed manner. . And determining whether the input image data is to be converted by the first conversion means or the second conversion means, and converting the input image data to the first or second conversion means. The image processing apparatus according to claim 1, further comprising a determination unit configured to selectively supply the image data. 4. The image processing apparatus according to claim 1, wherein said first transforming means performs a wavelet transform. 5. The image processing apparatus according to claim 1, wherein the second conversion unit performs predictive coding. 6. The image processing apparatus according to claim 1, wherein the output unit determines an effective transform coefficient by comparing a predetermined threshold value with each transform coefficient in the block. 7. The method according to claim 1, wherein the effective transform coefficient has a value within a range of a predetermined threshold value.
An image processing apparatus according to claim 1. 8. The image processing apparatus according to claim 7, wherein a lower limit value of the threshold is predetermined. 9. The image processing apparatus according to claim 4, wherein the output of the output unit is executed for each of different frequency bands obtained by performing the wavelet transform. 10. The output means outputs first information and second information capable of restoring a low-resolution image and then outputs first information and second information capable of restoring a high-resolution image. The image processing apparatus according to claim 1, wherein: 11. A first conversion step of generating an M-value conversion coefficient by converting input image data by a first method, and M> by converting input image data by a second method.
A second conversion step of generating an N-valued conversion coefficient of N; and a second position indicating a position where a valid conversion coefficient exists in a block constituted by either the M-valued conversion coefficient or the N-valued conversion coefficient. An output step of outputting one piece of information and second information indicating the effective transform coefficient. 12. A first conversion step of generating an M-value conversion coefficient by converting input image data by a first method, and M> by converting input image data by a second method.
A second conversion step of generating an N-valued conversion coefficient of N; and a second position indicating a position where a valid conversion coefficient exists in a block constituted by either the M-valued conversion coefficient or the N-valued conversion coefficient. An image processing program having: 1 information; and an output step of outputting second information indicating the effective conversion coefficient.