JPH08116450A - Image compressor and image expander - Google Patents

Abstract

PURPOSE: To select a quantization table suitable for image property of a block and to improve the image deterioration suppressing performance and the coding efficiency. CONSTITUTION: An image property of a noticed block is predicted by a block prediction section 6 based on the image property of a surrounding block arranged around the noticed block of a coding object or a decoding object and a quantization table selection section 7 selects a quantization table for quantization or a quantization table for inverse quantization based on the image property of the predicted noticed block.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image compression apparatus which divides image data into a plurality of blocks, orthogonally transforms the gradation data of each pixel, and quantizes and encodes the transform coefficient, and an encoding method using the same. The present invention relates to an image decompression device that obtains an original image by decoding and dequantizing the encoded data.

[0002]

2. Description of the Related Art As one of the methods for compressing multi-value gradation image data with high efficiency, a discrete cosine transform (hereinafter simply referred to as DC
Say T. There is an encoding method. In this DCT encoding method, an original image is divided into blocks made up of a plurality of pixels, and two-dimensional DCT is applied to grayscale data in the blocks.
Is applied to transform the coefficient into a spatial frequency distribution coefficient, and the high-frequency component side of the conversion coefficient is quantized more coarsely than the low-frequency component side by utilizing that human visual characteristics are insensitive to the high-frequency component. By doing so, the amount of information is reduced. Furthermore, the quantized transform coefficient is subjected to redundancy compression by entropy coding such as Huffman coding to reduce the amount of information.

FIG. 13 is a diagram for explaining a conventional example, (a) is a block diagram, and (b) is a diagram for explaining zigzag scanning. The upper block diagram in FIG. 13A shows the image compression device, and the lower block diagram shows the image decompression device. The image compression device includes a block conversion unit 1, a DCT unit 2,
The image decompressor includes a decoding unit 8, an inverse zigzag scanning unit 9, and a quantization unit 3, a zigzag scanning unit 4, and an encoding unit 5.
Inverse quantizer 10, inverse DCT (hereinafter referred to as IDCT)
The IDCT unit 11 and the inverse block conversion unit 12 that perform

The block conversion unit 1 cuts multi-value gradation image data into blocks each consisting of M × N pixels. These M and N are arbitrary natural numbers, but in the following description, M and N will be described as 8. D
The CT unit 2 formula 1 is applied to the gradation data s _{xy of} each block.
The transform coefficient S _uv is obtained by applying the DCT shown in FIG.

[0005]

[Equation 1]

Here, C _u and C _v in _Equation 1 are constants shown in Equation 2.

[0007]

[Equation 2]

The DCT unit 2 converts a block composed of 8 × 8 pixels into a matrix of 8 × 8 spatial frequency coefficients as shown in FIG. 13B. The high frequency components in this spatial frequency are concentrated in the lower right part of the matrix, and conversely, the lower frequency components are concentrated in the upper left part of the matrix. The components located in the shaded area in the figure are called DC components, and the other components are AC components.
It is called an ingredient. When a natural image such as a photograph is input to the DCT unit 2 and orthogonal transformation is performed, the transformation coefficients tend to be concentrated on the relatively low frequency side and scattered on the high frequency side. . Further, when an image such as a character image having a large density change is input to the DCT unit 2 and orthogonal transformation is performed, the transformation coefficient tends to have a large value on the high frequency side.

The quantizing unit 3 quantizes the transform coefficient output from the DCT unit 2 using a predetermined quantization table. That is, the transform coefficient is quantized in a predetermined quantization step corresponding to each matrix of the spatial frequency coefficient matrix. As described above, in the natural image, significant coefficients among the transform coefficients are concentrated on the relatively low frequency side, so that 0 coefficients are concentrated on the high frequency side in the matrix quantized by the quantization unit 3. Further, in the character image, the significant coefficient in the transform coefficient appears on the high frequency side as well, so that the matrix quantized by the quantizer 3 has 0 on the high frequency side.
Other quantized coefficients will appear.

The zigzag scanning unit 4 shows the quantized coefficient matrix of the two-dimensional array which is the output from the quantization unit 3 as shown in FIG.
(B) Scanning is performed in a zigzag direction as indicated by an arrow, and only the AC component thereof is converted into a one-dimensional array. This is because in a matrix in which 0 coefficients are concentrated on the high frequency side, the data after the final coefficient other than 0 (= significant coefficient) continues to 0 until the final data in the block, so scanning in the zigzag direction is performed. By doing so, even the last significant coefficient is encoded, and a block end code (for example, EOB) is inserted therein to omit the subsequent encoding and reduce the amount of information.

The encoding unit 5 entropy-encodes the one-dimensionally arranged quantized coefficients up to the final significant coefficient.
Huffman coding or arithmetic coding is used as the entropy coding. By the above, the DCT of the multi-value gradation image data is performed.
An image compression device for encoding is configured.

The image decompression device for performing DCT decoding has the reverse structure of the image compression device. That is,
The image decompression device includes a decoding unit 8, an inverse zigzag scanning unit 9, an inverse quantization unit 10, an IDCT unit 11, and an inverse block conversion unit 12. The decoding unit 8 decodes the input coded data and outputs a one-dimensional array of quantized transform coefficients. The inverse zigzag scanning unit 9 generates a two-dimensional quantized transform coefficient matrix by sequentially arranging the one-dimensional array of quantized transform coefficients in a matrix along the arrow shown in FIG. 13B.
It should be noted that it is 0 for the block end code EOB and thereafter.
Is inserted.

Further, the inverse quantizer 10 is incorporated in the image compression device.
The same quantization table as that used in the quantization unit 3 in
2D quantized transform coefficient matrix to DCT
The conversion coefficient matrix in FIG. Further
In addition, the IDCT unit 11 outputs D which is the output of the inverse quantization unit 10.
CT conversion coefficient S _uvIDCT shown in (3) (see 1)
The gradation data s_xyTo restore.

[0014]

(Equation 3)

Note that C _u and C _v in _Equation 3 are the same as the constants shown in Equation 2. The DCT transform coefficient matrix inversely transformed by the IDCT unit 11 includes a quantization error resulting from the quantization processing by the quantization unit 3, and the DCT unit 2
It does not completely match with the transform coefficient matrix at the time of generation in step. Therefore, the tone data inversely converted by the IDCT unit 11 does not completely match the original image.

In addition, the inverse block conversion unit 12 uses the IDCT
Since the output of the unit 11 is the gradation data in block units, these are temporarily stored in the line memory, and output after the time when the gradation data for 8 lines are accumulated, and the lines of the restored multi-gradation image data are output. Sequential output is possible. The above constitutes an image decompression device that decodes encoded data and restores multi-tone image data.

In the orthogonal transform coding method such as the DCT coding, the coding efficiency and the image quality of the restored image largely depend on the degree of quantization of the transform coefficient. For example, in an image such as a halftone image in which the density change is small, significant coefficients of the transform coefficients obtained by DCT are concentrated on the low frequency side. If this part is roughly quantized, the density difference between adjacent blocks is increased. Is noticeable. On the other hand, in an image such as a character image in which the density changes drastically, the significant coefficient of the conversion coefficient obtained by DCT appears on the high frequency side as well, so if this part is roughly quantized, blurring of the edge part becomes conspicuous. Therefore, if various images are considered and finely quantized over the entire frequency band of the transform coefficient, it is possible to minimize image deterioration due to a quantization error. However, this leads to a decrease in coding efficiency.

Therefore, it is desirable to adaptively use a quantization table suitable for the property of the block by examining the statistical property of the input multi-tone image data and the distribution of the DCT transform coefficient. FIG. 14 is a diagram for explaining this, in which the broken line shows the filter characteristic due to the property of the quantization table, the upper solid line shows the distribution of the conversion variables by the DCT before quantization, and the lower solid line shows the quantum. The transform coefficient obtained by dequantizing the transformed transform coefficient, that is, the transform coefficient after passing through the filter of the broken line is shown.

FIG. 14A shows a case where a block having a halftone property is quantized by a corresponding quantization table, and FIG. 14B shows a block having a character property according to the quantization table. This is the case when quantized by. As described above, when the quantization table of the step corresponding to the distribution of each transform coefficient is used, the reproducibility of the transform coefficient distribution after the inverse quantization can be ensured. On the other hand, FIG. 14C shows a case where a block having a character property is quantized by a quantization table suitable for a block having a halftone property. In this case, it can be seen that the distribution of transform coefficients after dequantization is significantly different from the original distribution of transform coefficients, which causes deterioration of the image.

Therefore, if the quantization table is adaptively used according to the distribution state of the transform coefficient in the DCT, both the suppression of image deterioration and the improvement of the coding rate can be achieved at the same time. The quantization table used at this time must be uniquely determined at the time of expansion processing. For this reason, in many conventional techniques, the information of the quantization table used together with the transmission of the encoded data of each block is transmitted.

In Japanese Patent Laid-Open No. 4-215385, it is possible to adaptively quantize the DCT transform coefficient of each block and to transmit and restore the encoded data without including the information of the used quantization table. An image encoding device and a decoding device are disclosed. In this technique, a first quantization table that finely quantizes low-frequency components and coarsely quantifies high-frequency components so as to adapt to blocks with little density change, and a low-frequency component that coarsely quantizes high-frequency components so as to adapt to blocks with drastic density changes A second quantization table for finely quantizing the components is prepared in advance, and which quantization table to use is determined according to the density change of the block to be encoded.

At this time, the quantization step width is set so that the result quantized by the first quantization table does not appear as a significant coefficient at a predetermined frequency or higher, so that the final encoded data of each block can be obtained. It is possible to identify which quantization table is used by looking at the frequency band in which the significant coefficient appears.

[0023]

However, in the technique disclosed in Japanese Unexamined Patent Publication No. 4-215385, an image, that is, D in which a character portion and a photograph portion are mixed, is used.
No consideration is given to an image in which significant coefficients of transform coefficients in CT appear on both the low frequency side and the high frequency side. That is, when a block having such an image property is quantized as a block with little change in density, the edge of the character part is blurred, and on the contrary, when quantized as a block with a large change in density, the density difference in the gradation part between adjacent blocks is noticeable. The inconvenience of being lost occurs.

Therefore, it is an object of the present invention to provide an image compression apparatus and an image decompression apparatus capable of selecting a quantization table adapted to the image property of a block, suppressing image deterioration and improving coding efficiency. And

[0025]

SUMMARY OF THE INVENTION The present invention is an image compression apparatus and an image decompression apparatus which are made to achieve the above object. That is, the image compression apparatus of the present invention divides the image data into blocks each consisting of a plurality of M × N pixels and orthogonally transforms the gradation data of each pixel in the block to obtain a predetermined quantum value. The quantization table is used to perform quantization and encoding.When quantizing the coefficient after orthogonal transformation in the block of interest to be encoded, based on the image properties of the peripheral blocks arranged around the block of interest. The image property of the block of interest is predicted by the block prediction unit, and the quantization table selection unit selects the quantization table according to the image property of the predicted block of interest.

The image decompressing device of the present invention divides the image data into blocks each consisting of a plurality of M × N pixels,
The coefficient obtained by orthogonally transforming the gradation data of each pixel in the block is quantized using a predetermined quantization table, and the coded one is restored to the original image.
When dequantizing the encoded data in the target block to be restored, the block predicting unit predicts and predicts the image property of the target block based on the image property of the peripheral blocks arranged around the target block. The quantization table selection unit selects a quantization table for inverse quantization according to the image property of the block of interest.

[0027]

In the image compression apparatus of the present invention, when the coefficient after orthogonal transformation in the target block to be encoded is quantized, the image property of the peripheral blocks arranged in the periphery of the target block, that is, the peripheral block is halftone. The block predicting unit predicts the image property of the target block based on the image property of the property, the character property, or a mixture thereof. By selecting the quantization table according to the image property of the block of interest based on this prediction in the quantization table selection unit, it is possible to perform the quantization with the quantization table suitable for the image property of the block of interest. become.

Further, in the image decompressing device of the present invention, when dequantizing the encoded data in the target block to be decoded, the image property of the target block based on the same prediction as that of encoding, that is, the image property of the peripheral block Is predicted by the block predictor. In addition, since the quantization table selection unit selects a quantization table for inverse quantization according to the image property of the block of interest based on this prediction, the same quantization table used for encoding is reversed. It becomes possible to make a unique decision when quantizing. Therefore, the information of the quantization table that has been selected and used is unnecessary when transmitting the encoded data.

[0029]

Embodiments of the image compression apparatus and the image expansion apparatus according to the present invention will be described below with reference to the drawings. FIG.
FIG. 1 is a block diagram illustrating an image compression device that is a first embodiment of the present invention. The image compression apparatus according to the first embodiment includes a block conversion unit 1 that divides multi-tone image data into blocks composed of 8 × 8 pixels, a DCT unit 2 that performs two-dimensional DCT on the blocks, and a DCT. The quantization unit 3 which quantizes the transform coefficient by the predetermined quantization table, the zigzag scanning unit 4 which transforms the transform coefficient of the two-dimensional array into a one-dimensional array, and the entropy coding of the transform coefficient which is one-dimensionally arranged. An encoding unit 5 is provided, and a block prediction unit 6 that predicts the image property of a target block that is an encoding target, and a quantization table selection unit 7 that selects a quantization table to be adapted based on this prediction are provided. There is.

The block predicting unit 6 determines the image property of the peripheral blocks arranged around the block of interest, that is, whether the peripheral blocks have a halftone with little density change, a character with drastic density change, or both. The property of the target block is predicted based on the image properties such as whether they are mixed. The quantization table selection unit 7 is provided with a plurality of types of quantization tables according to the image properties of the blocks in advance, and selects the quantization table suitable for the image properties of the target block predicted by the block prediction unit 6. And output it.

Next, the principle of prediction of the image property of the block of interest by the block predictor 6 will be described with reference to FIG. Note that in FIG. 2, blocks marked with the symbol I indicate those having a strong halftone image property, blocks marked with the symbol T indicate those having a strong character image property, and the symbol T
Blocks marked with / I are those having image characteristics in which both are mixed. Also the sign? Indicates a block of interest.

As shown in FIG. 2A, when all the peripheral blocks of the block of interest are blocks having a strong halftone image property, the image property of the block of interest is also halftone with a fairly high probability. Is expected to be In addition, as shown in FIG. 2B, when all the peripheral blocks of the block of interest are blocks having a strong image property of character, the image property of the block of interest also has a property of character with a fairly high probability. Expected to be.

FIG. 2C shows a case where most of the peripheral blocks are halftone blocks having a strong image quality, and one of the peripheral blocks has a strong character property. . In such a case, the block of interest is predicted to have image characteristics in which halftones and characters are mixed. These predictions use the property that adjacent pixels have a strong correlation in image data.

In this way, the image property of the target block is predicted from the image property of the peripheral block, and the prediction result is output to the quantization table selection unit 7 shown in FIG. In the case of an image decompression device, the peripheral block described here is an already-encoded block, and thus the same quantization table used for encoding is used as the quantization table for inverse quantization. You will be able to choose. That is, it is possible to select the same quantization table that was used for encoding during decoding, which means that even if multiple quantization tables are adaptively switched and used during encoding, that information is encoded. This will eliminate the need to leave it as data.

FIG. 3 is an internal block diagram of the block predictor 6. The block prediction unit 6 includes an average value calculation unit 61,
A variance calculation unit 62, a variance evaluation unit 63, a storage unit 64,
It is composed of a plurality of latches 65a to 65g and a focused block state prediction unit 67. This average value calculation unit 61
Is the sum of the gradation data s _xy input to the block prediction unit 6 in block units and divides by the total number of pixels (M × N = 64) belonging to the block to obtain the average gradation value s _{av for} each block.
Is calculated.

Further, the variance calculating section 62 determines the gradation data s _xy belonging to the block and the average gradation value s in the block.
_The sum of squares of the difference from _av , that is, the variance value σ is calculated. Here, the variance value σ takes a small value because the density change is smaller in a block having a stronger halftone image property, and on the contrary, the background density and the character part density are large in a block having a strong character image property. Since they can be distinguished, the variance value σ becomes a large value. The variance evaluation unit 63 uses this property to output the output of the variance calculation unit 62 to a predetermined threshold value (for example, Ti and T
(2 thresholds with t) to classify the image properties of the block.

For example, if Ti <Tt, then σ ≦ Ti
If it is, it is determined that the block has a strong halftone image property, and block feature information Ci (j) = 00 is output. Here, the block characteristic information Ci (j) is block characteristic information in the j-th block of the i-th block line. If Ti <σ ≦ Tt, it is determined that the block has an image property including a mixture of halftones and characters, and block characteristic information Ci (j) = 10 is output. Further, when Tt <σ, it is determined that the block has a strong character image property, and block feature information Ci (j) = 11 is output.

Here, an example in which only the variance value σ is used when determining the image property of the block has been shown, but the determination may be performed by referring to the average gradation value s _av . Also,
As other determination methods, pay attention to the property that the character image has many edge parts, determine the image property of the block from the number of edge parts detected using the Laplacian filter, and determine the density of the character part and the background. The image property of the block may be determined by noting the magnitude of the change and comparing the density change between adjacent pixels with a predetermined threshold value.

Further, as described above, when the density change of the gradation data is small, the significant coefficient in the DCT is concentrated on the low frequency side, and conversely, when the density change of the gradation data is severe, the significant coefficient in the DCT is large. It is also possible to use the fact that it appears on the high frequency side. That is, the image property of the block of interest is classified by examining the distribution status of the significant coefficients in the DCT. In this case, in the image compression apparatus shown in FIG. 1, the transform coefficient data output from the DCT unit 2 is input to the block prediction unit 6, and based on this, the block feature information Ci (j) is input to the quantization table selection unit 7. It may be configured to output to.

The storage unit 64 in the block prediction unit 6 is
The memory circuit is capable of storing block feature information for at least one block line. For example, 1 line 2048
A storage capacity of at least 256 words × 2 bits is required to handle the gradation data of pixels. In the storage unit 64, the new block feature information Ci (j) from the variance evaluation unit 63 is added.
Every time the block characteristic information is obtained, the block characteristic information Ci-1 (j) obtained on the i-1th line is read from the address j, and the current block characteristic information Ci (j) is written at the same address.

The latches 65a to 65g are provided in the dispersion evaluation unit 63.
The feature information template as shown in FIG. 4 is configured by temporarily holding the output from the storage unit 64. The output of the latch 65e corresponds to Ci-1 (j + 1), the output of the latch 65f corresponds to Ci-1 (j), the output of the latch 65g corresponds to Ci-1 (j-1), and the output of the latch 65c corresponds to The output of the latch 65d corresponds to Ci (j-1) and Ci (j-2). The template for predicting the block feature information of the block of interest is not limited to this size. In other words, the number of reference peripheral blocks may be increased to improve the predictive predictive value,
On the contrary, in order to suppress the circuit scale and the storage capacity of the storage unit 64, only the block adjacent to the target block may be limited.

The focused block state prediction unit 4 is a look-up table which receives the block feature information of the peripheral blocks as an address and outputs the predicted feature information PBC of the focused block. In predicting the image property of the block of interest, block feature information of the peripheral blocks used in the template has not yet been obtained at the initial stage of encoding. Further, in the block located at the end of the original image, the block used for the template may not exist in the original image. Therefore, it is assumed that blocks having strong halftone image characteristics are spread around the original image, and the blocks located at the edge of the original image are encoded in the initial stage of encoding. Also makes it possible to predict the block characteristic information of the block of interest.

In this case, since the block characteristic information indicating the halftone image property is "00", a gate circuit is provided at the output of the storage unit 64 and the output is forcedly performed during the processing of the first block line. It is sufficient to output "00". Further, for the latches 65a to 65g, a clear signal may be input before the processing of each block line starts.

FIG. 5 is an internal block diagram of the quantization table selection unit 7 of the image compression apparatus in FIG. Since the block feature information Ci (j) and the feature information prediction value PBC in the present embodiment take three values, the quantization table selection unit 7 has three types of quantization having different quantization characteristics corresponding to each case. Table 71 to quantization table 7
Equipped with 3. These quantization tables 71 to 73 may be those in which predetermined values are written in advance such as ROM, or may be configured by RAM and rewritable from the outside.

The quantization table 71 is adapted to a block where the characteristic information prediction value PBC = 00, that is, a block having a strong halftone image property. That is, since the transform coefficients in the DCT are concentrated on the low frequency side in a block having a strong halftone image property, the quantization threshold value is set so that the low frequency side is finely quantized and the high frequency side is roughly quantized. It has been done.

Further, the quantization table 73 is adapted to the block when the characteristic information prediction value PBC = 11, that is, the block having a strong character image characteristic. That is, in a block having a strong character-like image property, the transform coefficient in the DCT also appears on the high frequency side. In such a block in which the density changes drastically, the human visual sensitivity becomes relatively dull with respect to the density information, so the quantization threshold for finely quantizing the high frequency side of the transform coefficient and coarsely quantizing the low frequency side is set. It has established.

The quantization table 72 is adapted to a block in which the characteristic information prediction value PBC = 10, that is, a block in which characters and halftones are mixed. In other words, in such a block, both the conversion coefficient on the low frequency side that responds to a gradual change in density and the conversion coefficient on the high frequency side that responds to a drastic change in density are important. Quantization threshold is set so that

The multiplexer 74 is a selection circuit that selects and outputs one of the three quantization tables 71 to 73 described above according to the feature information prediction value PBC of the block of interest. With the configuration described above, it becomes possible to predict the image property of the target block based on the image property of the peripheral block and select the quantization table adapted to it. As a result, even when the block of interest is halftone, characters, or both are mixed, gradation data can be efficiently coded by selecting and using the optimum quantization table.

Next, a second embodiment of the present invention will be described. FIG. 6 is a block diagram for explaining the image decompression device in the second embodiment. The image decompression device according to the second embodiment is for decoding the encoded data generated by the image compression device according to the first embodiment described above to restore the original image data. The image expansion apparatus includes a decoding unit 8, an inverse zigzag scanning unit 9, an inverse quantization unit 10, and an IDCT unit 2.
And an inverse block conversion unit 12, and a block prediction unit 13 and a quantization table selection unit 14.

The decoding section 8 decodes input coded data and expands it into quantized transform coefficients which are one-dimensionally arranged. The reverse zigzag scanning unit 9 uses the ID in the subsequent processing.
Two-dimensionally quantized transform coefficients are used to perform CT.
It is transformed into a dimensional array to form a quantized transform coefficient matrix. The quantized transform coefficient matrix obtained here is generated by adaptively switching a plurality of quantization tables 71 to 73 (see FIG. 5) at the time of encoding.

The inverse quantizing unit 10 converts the quantized transform coefficient matrix into a transform coefficient matrix in DCT using the quantization table selected by the quantization table selecting unit 14. The IDCT unit 11 inversely transforms the 8 × 8 transform coefficient matrix into an 8 × 8 pixel block by performing an inverse discrete cosine transform. Also,
The inverse block conversion unit 12 is for outputting the restored multi-gradation image data line-sequentially by temporarily accumulating the gradation data generated in block units in the line memory.

Block predictor 13 in the image decompressor
Requires that the feature information prediction value of the block of interest to be decoded be the same as the prediction result performed at the time of encoding, so the block prediction unit 6 (in the image compression device described in the first embodiment The same as the one shown in FIG. 3) is used. The block prediction unit 13 uses the grayscale data inversely orthogonally transformed by the IDCT unit 11 to obtain the block feature information Ci (j) of the peripheral blocks.

Further, it is necessary to prepare the same quantization table used for the image compression as the quantization table provided in the quantization table selection unit 14. That is, the quantization table selection unit 14 finely quantizes the low frequency side of the transform coefficient and roughly quantizes the high frequency side of the transform coefficient, as shown in FIG. A quantization table 73 for roughly quantizing the side;
And a quantization table 72 for finely quantizing the transform coefficient over the entire frequency band.

Block predictor 13 in the image decompression device
Since it has the same configuration as the block prediction unit 6 (see FIG. 3) in the image compression device, the prediction of the image property of the target block based on the image property of the peripheral block should be the same result as at the time of encoding. You can That is, it is possible to select the quantization tables 71 to 73 to be used in the inverse quantization based on the same prediction as that used in encoding, and it is possible to minimize image deterioration. Further, since the prediction result of the same image property as that at the time of encoding can be obtained at the time of decoding, it is not necessary to transmit the information regarding the selection of the quantization table 71 to the quantization table 73 at the time of encoding. The conversion efficiency can be improved.

Next, a third embodiment of the present invention will be described. FIG. 7 is a block diagram illustrating an image compression apparatus according to the third embodiment. The image compressing apparatus according to the third embodiment includes a block transforming unit 1, a DCT unit 2, an adaptive quantizing unit 31, a standard quantizing unit 32, a first zigzag scanning unit 41 and a second zigzag scanning unit 41.
It comprises a zigzag scanning unit 42, a first coding unit 51 and a second coding unit 52, a block prediction unit 6, a quantization table selection unit 7, a prediction result comparison unit 15 and a selection unit 16.

The difference between the image compression apparatus according to the first embodiment and the image compression apparatus according to the third embodiment is that, in the third embodiment, first, there are two quantizations, an adaptive quantization unit 31 and a standard quantization unit 32. The first zigzag scanning unit 41 and the second unit.
First, a zigzag scanning unit 42 and two scanning units are provided.
The difference from the first embodiment is that it includes two coding units, a coding unit 51 and a second coding unit 52. Furthermore, the third embodiment differs from the first embodiment in that the prediction result comparison unit 15 and the selection unit 16 are provided. The first zigzag scanning unit 41 and the second zigzag scanning unit 42, and the first coding unit 51 and the second coding unit 52 may be the same.

The points different from the first embodiment will be described in detail below. First, the standard quantizer 32
A standard quantization table is provided in which quantization thresholds are set so that the transform coefficients obtained from the DCT unit 2 are finely quantized over the entire frequency band. The transform coefficient input by the standard quantization unit 32 is always quantized by the standard quantization table regardless of the prediction result of the image property of the block of interest. The quantized transform coefficient quantized by the standard quantization table is entropy-coded via the second zigzag scanning unit 42 and the second coding unit 52.

The prediction result comparison unit 15 obtains the feature information prediction value PBC of the target block obtained from the block feature information of the peripheral blocks (encoded) of the target block and the block feature information Ci (j) of the target block itself. For example, if the feature information prediction value PBC and the block feature information Ci (j) match (prediction is correct), "1" is output, and otherwise (0), "0" is output. It is a thing.

When the prediction result is correct on the basis of the output from the prediction result comparing unit 15, the selecting unit 16 uses the quantization table selected by the adaptive quantizing unit 31 according to the image property. The quantization is performed and the first zigzag scanning unit 41
And the encoded data encoded via the first encoding unit 51 is selected and output. On the other hand, if the prediction result is incorrect, the standard quantization unit 4 performs quantization using the standard quantization table and encodes it via the second zigzag scanning unit 42 and the second encoding unit 52. The encoded data is selected and output.

If the prediction of the image property of the block of interest is wrong, the optimum quantization table cannot be selected, which causes the deterioration of the restored image and the like. In the image compression apparatus according to the third embodiment, the selection unit 1 selects the quantized image using the standard quantization table in such a case.
The feature is that it can be selected in 6. Since the standard quantization table is capable of finely quantizing the transform coefficient in the DCT over the entire frequency band, it is possible to minimize deterioration caused in the restored image regardless of the image property of any kind. .

For this reason, when the prediction of the image property of the target block is incorrect, it is not the one quantized by the non-adaptive quantization table but the one quantized and encoded by using the standard quantization table at the same time. Select in the selection unit 16,
I am trying to output. This makes it possible to minimize image deterioration in the restored image even if the prediction of the image property of the block of interest is erroneous.

As an example, the transform coefficient quantized by using the quantization table suitable for the block having the halftone image property or the quantization table suitable for the block having the character image property is inversed by the standard quantization table. The case of quantization will be described. In the quantization table considering the image property, the quantization step in the frequency band where significant coefficients are considered to be concentrated is set to be relatively small. Therefore, even if the inverse quantization is performed using the standard quantization table, the reproducibility of the significant coefficient portion is not impaired.

On the other hand, the quantization step is set relatively large in the frequency band in which it is considered that the significant coefficients are not concentrated, but the standard quantization table sets the quantization step finely over the entire frequency band. Even if inverse quantization is performed using this, the component of the insignificant coefficient will be further attenuated. Therefore, the multi-gradation image data obtained by dequantizing the transform coefficient in the DCT using the standard quantization table and further performing the inverse discrete cosine transform emphasizes the character property or halftone property of the original image. Becomes

In the image compression apparatus according to the third embodiment shown in FIG. 7, there are two zigzag scanning units (first zigzag scanning unit 41 and second zigzag scanning unit 42) and two coding units ( The first encoding unit 51 and the second encoding unit 52) are provided, but the same applies to configurations that are common to each. That is, as in the modification shown in FIG. 8, after the adaptive quantizing unit 31 and the standard quantizing unit 32, the selecting unit 1
By disposing 6 and arranging the zigzag scanning section 4 and the encoding section 5 after the selecting section 16, the circuit scale can be reduced.

Next, an image decompression device according to the fourth embodiment of the present invention will be described. The image decompression device in the fourth embodiment is for decoding the encoded data generated by the image compression device in the third embodiment to restore the original multi-gradation image data. This image expansion device includes a decoding unit 8, an inverse zigzag scanning unit 9, a standard inverse quantization unit 102, an adaptive inverse quantization unit 101, a first IDCT unit 111, and a second IDCT unit 111.
The IDCT unit 112, the block prediction unit 13, the quantization table selection unit 14, the block analysis unit 17, the prediction result comparison unit 18, the selection unit 19, and the inverse block conversion unit 12 are included.

The decoding section 8 decodes the input coded data and decompresses it into quantized transform coefficients which are one-dimensionally arranged. Further, the inverse zigzag scanning unit 9 converts the one-dimensionally arranged quantized transform coefficients into a two-dimensional array to perform IDCT in the subsequent processing, and forms a quantized transform coefficient matrix. The adaptive dequantization unit 101 dequantizes the quantized transform coefficient matrix using the quantization table selected by the quantization table selection unit 14, and transforms the quantized transform coefficient matrix into a transform coefficient matrix.

Further, the first IDCT unit 111 and the second ICT
The DCT unit 112 outputs 8 × 8 which is the output of the inverse quantization unit 10.
Inverse discrete cosine transform is applied to the transform coefficient matrix of
It is the inverse conversion into an 8 × 8 block. The inverse block conversion unit 12 is for outputting the restored gradation data line-sequentially by temporarily storing the expanded gradation data generated in block units in the line memory. The block prediction unit 13 obtains block feature information of peripheral blocks (decoded) arranged around the target block to be decoded, and outputs the feature information prediction value PBC of the target block based on the block feature information. is there.

The standard dequantization unit 102 dequantizes a quantized transform coefficient matrix decoded using a standard quantization table provided in advance, and transforms it into a transform coefficient matrix. As described above, the restored gradation data obtained by performing the inverse conversion using the standard quantization table emphasizes the character property or halftone property of the original image.

The block analysis unit 4 analyzes the block in which the image property is emphasized based on the statistical property, and outputs the block feature information Ci (j). This has an internal configuration as shown in FIG. 3, finds the average value and the variance value of the gradation data of the pixels belonging to the block, and classifies the image characteristics of the block from the magnitude relation between the predetermined threshold value and the variance value. ing.

Further, the prediction result comparison unit 18 converts the feature information prediction value PBC in the target block estimated from the block feature information in the decoded peripheral block and the code data of the target block into transform coefficients using the standard quantization table. The block characteristic information Ci (j) based on the gradation data obtained by the reverse inverse discrete cosine transform is compared to check whether or not they match.

If the prediction result comparison unit 18 determines that the feature information prediction value PBC and the block feature information Ci (j) match, the selection unit 19 selects the output from the first IDCT unit 111, The output from the second IDCT unit 112 is selected when it is determined that the output is not done.
With such a configuration, the image property of the target block is predicted based on the image property of the peripheral blocks arranged around the target block to be decoded, and the quantization table for dequantization adapted to this is selected. In addition, it becomes possible to select the grayscale data obtained by inverse quantization with the standard quantization table when the prediction is incorrect.

In decoding, the image property of the target block is predicted and the prediction results are compared based on the same standard as in encoding, so that no information about the quantization table is given. Inverse quantization can be performed using the same quantization table as when encoding. As a result, the coding efficiency can be improved.

Next, a fifth embodiment of the present invention will be described. FIG. 10 is a block diagram for explaining the image compression apparatus in the fifth embodiment. In determining the quantization table for quantizing the transform coefficient of the target block to be encoded, the image compression apparatus in the fifth embodiment determines the feature of the target block from the block feature information of the peripheral blocks (encoded). The process of obtaining the information prediction value PBC and selecting an adaptive one from a plurality of prepared quantization tables in accordance therewith is the same as that of the image compression apparatus in the first and third embodiments. In the fifth embodiment, when the block characteristic information Ci (j) regarding the target block obtained by the block predicting unit 6 and the characteristic information predicted value PBC do not match, the quantization table selecting unit 7 The feature is that the quantization table is selected again according to the image property of the block of interest and the encoding is repeated.

Further, the encoding section 5 recognizes whether or not the prediction of the image property of the target block in the block prediction section 6 is correct based on the output from the prediction result comparison section 15, and the prediction is incorrect. In this case, the quantization table selection unit 7 adds the information of the quantization table finally determined and the fact that the quantization table has been switched to the encoded data as header information.

FIG. 11 is a diagram showing an example of encoded data output from the encoding unit 5. If the prediction is correct, the same quantization table as that used in the encoding can be specified in the decoding as described in the first and second embodiments, so that the header information about it can be specified. Is unnecessary (see the first block, the second block, the n-2th block, and the nth block in the figure).

On the other hand, when a quantization table different from the one determined by the prediction result is used for encoding, accurate decoding is performed if there is no information about which quantization table is used. I can't. Therefore, only when the prediction is incorrect, information about the used quantization table is given as header information, and when decoding, the quantization table used is determined by decoding the header information. (See the 3rd block and the n-1th block in the figure).

FIG. 12 is a block diagram for explaining the sixth embodiment of the present invention. In order to restore the original gradation data by decoding the encoded data generated by the image compression apparatus in the fifth embodiment. 2 is an example of the image decompressing device. The decoding unit 8
When the encoded data is received and the header information (see FIG. 11) for designating the quantization table is detected from the encoded data, the quantization table selecting unit 14 is notified of that fact. Similarly to the second and fourth embodiments, the block predicting unit 13 obtains block feature information of peripheral blocks (decoded) arranged around the target block to be decoded, and based on this, The feature information prediction value PBC of the block of interest is output.

The quantization table selection unit 14 selects the quantization table based on the prediction result from the block prediction unit 13 unless it receives information about the designation of the quantization table from the decoding unit 8. This distinguishes between the case where a quantization table different from prediction is used at the time of encoding and the case where a quantization table based on prediction is used, and the quantization table used at the time of encoding is used. It becomes possible to perform decryption.

In the sixth embodiment, the example in which the header information regarding the designation of the quantization table output from the decoding unit 8 is notified to the quantization table selecting unit 14 is shown, but this is notified to the block predicting unit 13. However, the specified quantization table may be selected by the quantization table selection unit 14. Further, in the first to sixth embodiments, the case where the quantization table selection units 7 and 14 are provided with three types of quantization tables corresponding to three types of image characteristics has been described, but the present invention is not limited to this. The same is true even if a large number of quantization tables corresponding to a larger number of image types are provided.

[0080]

As described above, the image compression device and the image expansion device of the present invention have the following effects. That is, in the image compression device of the present invention, since the image property of the target block is predicted from the image property of the peripheral block arranged around the target block to be encoded, and the quantization table is adaptively switched, Even if a block has halftone image characteristics, textual image characteristics, or even image characteristics such as a mixture of these, the optimum quantization table can be selected according to it, and image deterioration can be prevented. It is possible to achieve the encoding that is minimized. Further, in the image compression apparatus of the present invention, since the reference of the image property prediction used in the image compression apparatus is adapted, the image property of the block of interest is predicted from the image property of the peripheral block, and at the time of encoding. The same quantization table used can be selected. This eliminates the need for information regarding the selection of the quantization table, and makes it possible to improve the coding efficiency.

[Brief description of drawings]

FIG. 1 is a block diagram illustrating a first embodiment of the present invention.

FIG. 2 is a diagram illustrating prediction from peripheral blocks.

FIG. 3 is an internal block diagram of a block prediction unit.

FIG. 4 is a diagram showing a feature information template.

FIG. 5 is an internal block diagram of a quantized TCP selection unit.

FIG. 6 is a block diagram illustrating a second embodiment of the present invention.

FIG. 7 is a block diagram illustrating a third embodiment of the present invention.

FIG. 8 is a block diagram illustrating a modification of the third embodiment.

FIG. 9 is a block diagram illustrating a fourth embodiment of the present invention.

FIG. 10 is a block diagram illustrating a fifth embodiment of the present invention.

FIG. 11 is a diagram showing an example of encoded data.

FIG. 12 is a block diagram illustrating a sixth embodiment of the present invention.

13A and 13B are diagrams illustrating a conventional example, FIG. 13A is a block diagram, and FIG. 13B is a diagram illustrating zigzag scanning.

FIG. 14 is a diagram illustrating an example of quantization.

[Explanation of symbols]

 1 block conversion unit 2 DCT unit 3 quantization unit 4 zigzag scanning unit 5 coding unit 6, 13 block prediction unit 7, 14 quantization table selection unit 8 decoding unit 9 inverse zigzag scanning unit 10 inverse quantization unit 11 IDCT unit 12 Inverse block converter

Claims (2)

[Claims] 1. A coefficient obtained by dividing image data into a block composed of a plurality of M × N pixels and orthogonally transforming gradation data of each pixel in the block using a predetermined quantization table. An image compression apparatus for performing quantization and encoding, wherein when quantizing a coefficient after orthogonal transformation in a block of interest to be encoded, based on image properties of peripheral blocks arranged around the block of interest, A block prediction unit that predicts the image property of the block of interest and a quantization table selection unit that selects a quantization table according to the image property of the image property of the block of interest predicted by the block prediction unit are provided. An image compression device characterized by the above. 2. The image data is divided into blocks each consisting of a plurality of M × N pixels, and the coefficient obtained by orthogonally transforming the gradation data of each pixel in the block is used for a predetermined quantization table. An image decompression device for decompressing encoded data in a target block to be decompressed by decompressing the encoded data into an original image by dequantizing A block prediction unit that predicts the image property of the block of interest based on the image property, and a quantization table for dequantization according to the image property is selected from the image properties of the block of interest predicted by the block prediction unit. An image decompression device, comprising: