JPH11164150A - Image processing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To compress image data with the same code amount as in the case of compression by the current GBTC coding method, with higher image quality, at high speed, and with a high compression rate. SOLUTION: Image data read from a scanner is converted into bitmap data and stored in a frame memory (first storage unit) 101. Here, when one piece of bitmap data is completed, this data is transferred to the image area separation unit 102. The image area separation unit 102 separates the image data into an area having a large gradation difference and an area not having the large gradation difference based on the image data developed into the bitmap. Each of the blocks separated into any one of the regions is
In, the optimal quantization is performed for each region,
Compressed to a fixed length. In the case of the rotation function, the data compressed to a fixed length is further subjected to secondary compression by an entropy coding unit 104 using an arithmetic code or the like. Thereafter, these compressed data are temporarily stored in the storage unit 105.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing apparatus for compressing image data, and more particularly, to an image processing apparatus suitable for various apparatuses which need to compress data in an easily manageable state, such as an image editing function of a digital copying machine. It relates to a processing device.

[0002]

2. Description of the Related Art For example, a digital copying machine has a rotation function and an editing function. To realize these functions, image data is compressed. For this compression, GBTC (Generalized Block Trun
Cation coding is often used. This encoding method converts 8 bits / pixel data to 3/8.
This is also discussed in, for example, "Image Compression Method Suitable for Hard Copy Apparatus" (Kenichiro Oka and two others) accepted on January 9, 1990 in the Journal of the Printing Society of Japan. .

[0003]

SUMMARY OF THE INVENTION However, GBT
In the C encoding method, severe moire often occurs in a halftone image or the like depending on the quantization method. Further, when the electronic sort function is realized, when the fixed-length compressed data is further entropy-encoded to increase the number of stored data, the data that is fixed-length-compressed by the GBTC encoding method cannot be highly compressed. is there.

As another conventional example of compressing image data, there has been proposed a fixed-length compression method using a multi-valued error diffusion method as disclosed in Japanese Patent Application Laid-Open No. 9-74488.
This method has a problem that the processing load is heavy.

[0005] In view of the above-mentioned conventional problems, the present invention can compress image data with higher code quality, higher speed, and a higher compression rate with the same code amount as when compressed by the current GBTC coding method. It is an object to provide an image processing device.

[0006]

In order to achieve the above object, a first means is a block dividing means for dividing image data into blocks of 2.times.2 pixels, and a plurality of types are provided according to a gradation difference in the block. And a quantizing means for quantizing each of the separated areas to the same number of bits with a fixed length, wherein the image area separating means calculates a difference from a pixel value of input image data. It is characterized in that it is separated into a first region having a large difference and a second region not having a large difference.

The second means is the first means, wherein the image area separating means first separates the first area, and then uses the high frequency coefficient after subband conversion for the second area. It is further characterized by being separated.

[0008] A third means is the first means, wherein the quantization means is configured to determine, for the first area, data which is close to the maximum or minimum value of the pixel value among the four pixels of the block. The method is characterized in that each pixel is converted into a maximum value or a minimum value and stored, the remaining pixels are averaged, and in the second region, the high-frequency component after the sub-band conversion is vector-quantized.

A fourth means is the second or third means, wherein the quantization means applies a gray code to the low-frequency component after the sub-band conversion for the block to be subjected to the sub-band conversion. It is characterized by the following.

[0010]

Embodiments of the present invention will be described below.

[First Embodiment] FIG. 1 is a functional block diagram showing a process of signal processing of a compressor in a digital image forming apparatus as an embodiment of an image processing apparatus according to the present invention. Here, a digital copying machine is assumed as the digital image forming apparatus, and a fixed length compression for a rotation function of the digital copying machine will be described.

The compressor in this embodiment includes a frame memory 101, a block dividing unit 102, an image area separating unit 10
3. It comprises a quantization unit 104, an entropy encoding unit 105, and a storage unit 106. The image data read from the scanner is converted into bitmap data and stored in the frame memory (first storage unit) 101. Here, when one piece of bitmap data is completed, this data is divided into 2 × 2 pixels by the block dividing unit 102 and transferred to the image area separating unit 103. The image area separation unit 103 separates an area having a large gradation difference from an area where the difference is not large based on the image data divided into 2 × 2 pixels.
Each block separated into any one of the regions is
At 04, optimal quantization is performed on each region and compressed to a fixed length. In the case of the rotation function, the data compressed to the fixed length is further encoded by the entropy encoding unit 105,
Secondary compression using arithmetic codes and the like is performed. Thereafter, these compressed data are temporarily stored in the storage unit 106.

In the image area separating unit 103, the block dividing unit 1
In step 02, for example, image data having a depth of 8 bits (256 gradations) is divided into blocks each including 2 × 2 pixels a, b, c, and d (see FIG. 5). Then, a block whose input image data value is close to “0” or “255” is extracted. For example, in FIG. 2, a = 0, b = 100, c = 251, d
= 200, if there is a block as shown in FIG.
“1” is replaced by “255”. As a result, since “0” and “255” exist in the block, it is determined to be an “edge area”. Further, the block as shown in FIG. 3 has a = 25, b = 100, c = 180, and d = 200. Since neither “0” nor “255” or a value close to these exists, “ Edge region "is determined.

In the quantization section 104, the following processing is executed. Here, it is assumed that 8 bits are compressed to 3 bits as fixed-length compression. Therefore, 4
When the pixels are collectively fixed-length compressed, 8 × 4 = 32 (bits) becomes 3 × 4 = 12 (bits).

First, an "edge region" is
In the block determined as, as shown in FIG.
Data within a range of ± 10 close to “0” and “255” (a =
(0, c = 251) is converted to “0” or “255”. The other intermediate density data is averaged in the block. For example, in FIG.
Data of b = 100 and d = 200 that did not become “5” are averaged to become “150”. The bit representation compressed to the fixed length at that time is as shown in FIG. M
The 4 bits from the SB side are intermediate density (gray pixels), the next 4
The bit indicates a black pixel “0”, and the LSB side 4 bits indicate a 4-bit linear quantization of the averaged gray density. Here, the average value is “150” (1001011)
0), it becomes “144” (10010000), and “1001” is entered. This gives a total of 12 bits.

In the "non-edge area" which is not determined as the "edge area", first, sub-band conversion is performed.
Here, a 2 × 2 wavelet transform is used as one of the sub-band transforms. Wavelet transform (Ha
In rr Wavelet transform), image data divided into 2 × 2 blocks is decomposed into one low-frequency component LL and three high-frequency components HL, LH and HH as shown in FIG.

LL = (a + b + c + d) / 4 HL = (ab) / 2 + (cd) / 2 LH = (a + b) / 2− (c + d)｝ / 2 HH = (ab) − (c) −d) (1) Here, the input image data has a depth of 8 bits (256 gradations).
In the case of, the LL component can take a value from 0 to 255, so 8 bits, the HL and LH components can take a value from -255 to 255, so 9 bits, and the HH component takes a value from -510 to 510. 10 bits.

Therefore, when the wavelet transform as shown in FIG. 5 is performed on the image data of the “non-edge area” in FIG. 3, the coefficients (LL = 126,
HL = −128, LH = −48, HH = −55). The bit representation in this case is as D3 shown on the right side of FIG. As can be seen from this figure, the first 4 bits are used as the position information of the gray pixels, similarly to the “edge area”. Therefore, since all four pixels are gray pixels, all four bits on the MSB side are “1”. The remaining 5 bits are used for vector quantization of wavelet coefficients. Of these, the upper 4 bits are obtained by quantizing the LL of the low frequency component by 4 bits. Here, “126” (0
1111110) → “112” (01110000). The lower 4 bits are the above-mentioned three components (L
H, HL, HH) into a vector,
Insert up to 15. A vector code is determined by using a table that is close to a previously prepared table.
This example is shown in the figure. That is, as shown in FIG.
Up to five patterns are prepared, and the coefficient having the largest value among the current coefficients is used to select from these. In the coefficient shown in FIG. 6, HL = −128 is the largest as an absolute value, and thus the vector code 6 (0,
−128,0) is selected. Therefore, "0110" is entered in the lower 4 bits.

As described above, according to the present embodiment, the number of processing blocks is 2.times.2, so that the number of working input line memories can be reduced to two, and the cost can be reduced. In addition, since the image area is separated by directly detecting the edge from the pixel value of the input data, efficient quantization suitable for each area can be easily performed.

[Second Embodiment] FIG. 8 is a functional block diagram showing a process of signal processing of a compressor of a digital image forming apparatus according to a second embodiment.

The compressor in this embodiment includes a frame memory 801, a block division section 802, a first image area separation section 803, a subband conversion section 804, a second image area separation section 805, a quantum 806, an entropy encoding unit 807, and a storage unit 808. In the second embodiment configured as described above, the 2 × image data input in the same manner as the first embodiment and extracted by the block division unit 802 from the image data expanded into the bitmap is used.
The image data of the second block is transferred to the first image area separation unit 803. Here, a region having a large gradation difference is cut out first. Here, this area is called a “strong edge area”. Then, the unextracted block is subjected to sub-band conversion by the sub-band converter 804 to generate a low-frequency component and a high-frequency component.
At 05, it is further separated into two types of regions. Each of the blocks whose region type is determined is quantized by the quantization unit 806.
Optimum quantization is performed on each area, and the data is compressed to a fixed length. Then, as in the first embodiment, the data is further compressed by the entropy coding unit 807 and stored in the storage unit 80.
8

In this embodiment, the frame memory 8
01, block division unit 802, first image area separation unit 80
3 and the sub-band converter 804 are configured in the same manner as in the first embodiment, and operate in the same manner. Therefore, the second image area separation unit 805 will be described in detail. The block separated as the “non-edge area” by the first image area separation unit 803 performs the above-described sub-band conversion (wavelet conversion) in the sub-band conversion unit 804 to generate a low-frequency component and a high-frequency component. Is done. Then, the subband-converted block is processed by the second image area separation unit 805 to obtain a separation parameter: | LH |> 32 or | HL |> 32 or | HH |> 32 (2) If this condition is satisfied, the block is determined to be an “edge area”. Image data similar to FIG. 3 shown in FIG. 9 becomes | HL | = 128> 32 by the wavelet transform, and satisfies the separation condition. Therefore, "edge area"
Is determined. Further, the block in FIG. 10 does not satisfy any of the conditions of the expression (2), so that the “non-edge area”
Is determined. That is, the first image area separation units 103 and 8
Area with strong gradation difference extracted in 03 "Strong edge area"
, And the second image area separation unit 805 determines an “edge area” and a “non-edge area” for the other areas.

The separated image data of each block is further quantized by a quantization unit 806. here,
The "strong edge region" is fixed-length compressed as shown in FIG. 4, and the "edge region" is fixed-length compressed as shown in FIG. However, in this embodiment, “lower 2 bits of 4 bits used for H of wavelet vector quantization to distinguish from non-edge areas are“ 0
A code that is "0" cannot be used. Therefore, all 16 of the vector codes shown in FIG. 7 cannot be used, and only 16−4 = 12 can be used. Therefore, the contents of the vector are limited by narrowing down the frequency of appearance.

The bit representation of the coefficients of the block in the "non-edge area" shown in FIG. 10 is as shown in FIG. The four MSB bits indicate the position information of the gray pixels as before. Here, all four bits are "1". Remaining 8
As for the bit, since the edge has the weakest edge, only the low frequency component is used. Then, “0” is inserted to distinguish the two LSB-side bits from the “edge area”, and
A 6-bit linearly quantized LL component is used. Here, “113” (01110001) → “11”
2 "(01110000).

As described above, the data is divided into three regions, and different quantizations are performed. However, the way of discrimination on the decoding side is as follows. First, all the leading 4 bits are “1” (1111XXXX).
X), it can be determined whether or not it is a “strong edge area”. Of the lower 8 bits of the block determined not to be a "strong edge area". When the LSB side 2 bits are “00”, since the area is a “non-edge area” as described above, it can be seen that the remaining area is an “edge area”.

If the data compressed to a fixed length as described above is compressed by entropy encoding as shown in FIG. 8 for an electronic sort function or the like and is stored in a storage memory, it becomes executable. In order to increase the number of sheets to be sorted, it is desirable to compress the compressed data having a fixed length of 3/8 to a smaller capacity. Therefore, the "edge region" and the "non-edge region" are expressed by 8 bits by a vector of a low frequency component (LL) and a high frequency component (H) by a wavelet transform, and the low frequency component (LL) of each block.
Only gray coding “0100” (FIG. 12), “010
000 "(FIG. 13), when performing entropy coding,
The compression efficiency can be further increased as compared with the case where the gray coding is not performed.

As described above, according to the second embodiment, it is possible to faithfully reproduce a sharp edge portion by extracting a block having a large gradation difference from the input image data. Also, the remaining blocks can be further divided into several types of regions according to the result of the frequency decomposition as compared with the first embodiment, and more efficient quantization can be performed for each region.

Further, "0" which is a portion where the gradation difference is large,
Since data close to "255" is stored with almost no deterioration, it is possible to prevent moiré due to quantization of a halftone dot image.

Further, efficient quantization can be performed by vectorizing high-frequency components in a region other than the "strong edge region".

Further, in regions other than the "strong edge region", the low-frequency component (LL) and the high-frequency component (H) are represented by 8 bits by a vector of wavelet transform, and only the low-frequency component (LL) of each block is gray. By performing coding “0100” and “010000”, the compression efficiency can be further improved when entropy coding is performed, as compared with the case where gray coding is not performed.

[0031]

As described above, according to the first aspect of the present invention, since an image is divided into blocks of 2 × 2 pixels and subband conversion is performed, only two lines of working memory are required. .

Since an image area is separated by directly detecting an edge based on input data, efficient quantization suitable for each area can be easily performed.

According to the second aspect of the present invention, it is possible to faithfully reproduce a sharp edge portion by detecting a block having a large gradation difference by looking at input data.

The remaining blocks can be further divided into several types of regions based on the result of the frequency decomposition, and efficient quantization suitable for each region can be performed.

According to the third aspect of the present invention, for the first area, of the four pixels of the block, data that is close to the maximum or minimum possible pixel value is converted to the maximum or minimum value, respectively. Since these data are stored almost without deterioration, it is possible to prevent moiré due to quantization of the halftone dot image.

The remaining pixels are averaged and the second
Since the high-frequency component after the sub-band conversion is vector-quantized for the region of, efficient quantization can be performed.

According to the fourth aspect of the present invention, it is possible to improve the compression ratio at the time of enttopic coding.

[Brief description of the drawings]

FIG. 1 is a functional block diagram illustrating a signal processing process of a compressor of a digital image forming apparatus according to a first embodiment of the present invention.

FIG. 2 is an explanatory diagram illustrating an example of image data of an edge area separated by an image area separation unit in FIG. 1;

FIG. 3 is an explanatory diagram illustrating an example of image data of a non-edge area separated by an image area separation unit in FIG. 1;

FIG. 4 is an explanatory diagram of quantization in an edge region of the quantization unit in FIG. 1;

FIG. 5 is an explanatory diagram showing a conversion state of subband conversion (Wavelet conversion).

FIG. 6 is an explanatory diagram of quantization of a non-edge area in the quantization unit of FIG. 1;

7 is a diagram illustrating a vector code table used when vector quantization is performed on a high-frequency component after subband conversion by the quantization unit in FIG. 1;

FIG. 8 is a functional block diagram illustrating a signal processing process of a compressor of the digital image forming apparatus according to the second embodiment.

9 is an explanatory diagram illustrating an example of image data of an edge area separated by a second image area separation unit in FIG. 8;

FIG. 10 is an explanatory diagram illustrating an example of image data of a non-edge area separated by a second image area separation unit in FIG. 8;

FIG. 11 is an explanatory diagram of quantization of a non-edge area in the quantization unit of FIG. 8;

FIG. 12 is an explanatory diagram showing a gray coding process of an edge region in the second image area separation unit in FIG. 8;

FIG. 13 is an explanatory diagram showing a gray coding process of a non-edge area in the second image area separation unit in FIG. 8;

[Explanation of symbols]

 101,801 Frame memory 102 Image area separation unit 103,805 Quantization unit 104,806 Entropy coding unit 105,807 Storage unit

Claims (4)

[Claims] A block dividing unit that divides image data into 2 × 2 pixel blocks; a region separating unit that divides the image data into a plurality of types of regions by a gradation difference in the block; Quantizing means for quantizing the number of bits to a fixed length, wherein the image area separating means comprises: a first area having a large gradation difference from a pixel value of input image data; An image processing apparatus characterized in that the image processing apparatus is separated into images. 2. The apparatus according to claim 1, wherein the image area separating unit separates the first area first, and then further separates the second area using a high-frequency coefficient after subband conversion. 2. The image processing device according to 1. 3. The quantization means converts data close to the maximum value or minimum value of a pixel value among the four pixels of the block into the maximum value or minimum value for the first area. The image processing apparatus according to claim 1, wherein the image is stored, the remaining pixels are averaged, and the high frequency component after the subband conversion is vector-quantized in the second region. 4. The block according to claim 2, wherein the quantization means applies a gray code to the low-frequency component after the sub-band conversion for the block to be subjected to the sub-band conversion. Image processing device.