JP3337699B2 - Image processing apparatus and method - Google Patents

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 and method, and more particularly, to an image processing apparatus having an orthogonal transformation function and a method thereof.

[0002]

2. Description of the Related Art Halftone images such as photographs (hereinafter "images")
) Is required to be the number of pixels × the number of gradation bits, and an enormous storage capacity is required to store a high-quality color image in the memory. For this reason, various information amount compression schemes have been proposed, and the amount of information in an image is compressed and then stored in a memory to reduce the storage capacity.

FIG. 8 shows JPEG (Joint Photographic Experts Groove) as an international standardized method of color still image coding.
up), the coding method of the baseline system (basic method) proposed by Yasuda: "International Standardization of Color Still Image Coding", Journal of the Institute of Image Electronics Engineers of Japan, Vol. 18, No. 6, pp. 398-4
09, 1989).

In FIG. 1, pixel data of an image input from an input terminal 1 is divided into 8 ×
The data is cut out into blocks of eight pixels, subjected to discrete cosine transform in a discrete cosine transform (hereinafter referred to as “DCT”) circuit 17, and the transform coefficients are supplied to a quantizer (Q).
The quantizer 40 linearly quantizes the input quantization coefficient according to the quantization step information input from the quantization (Q) table 41. The DC coefficients of the quantized transform coefficients are subjected to a difference (prediction error) from the DC component of the previous block by a predictive coding circuit (hereinafter referred to as “DPCM”) 42, and the one-dimensional Huffman coding circuit 43 Supplied to FIG. 9 is a block diagram showing a detailed configuration of the DPCM. The DC coefficient input from Q40 is supplied to delay circuit 53 and subtractor 54. Delay circuit
53 delays the DC coefficient by the time necessary for the DCT 17 to calculate one block, that is, 8 × 8 pixels. That is, the delay circuit 53
, The DC coefficient of the previous block is supplied to the subtractor 54. Therefore, the difference (prediction error) of the DC coefficient between the current block and the previous block is output from the subtractor 54. In this prediction coding method, the prediction of the previous block is used for prediction.
Since the DC coefficient is used, the DPCM 42 has the delay circuit 53 as described above.
Composed of

The one-dimensional Huffman encoding circuit 43 converts the prediction error signal supplied from the DPCM 42 into a DC Huffman code table.
The obtained DC Huffman code is subjected to variable length encoding according to 44 and supplied to the multiplexing circuit 51.

On the other hand, the AC coefficient (D
The scan conversion circuit 45 performs zigzag scanning in order from a low-order coefficient in the scan conversion circuit 45 as shown in FIG. Significant coefficient detection circuit 46
Determines whether the input AC coefficient is “0”,
In the case of “0”, a count-up signal is supplied to the run length counter 47 to increase the count value by one. On the other hand, if it is other than “0”, a reset signal is supplied to the run length counter 47 to reset the count value, and the AC coefficient is supplied to the grouping circuit 48 so that the group number as shown in FIG. Split into SSSS and additional bits. The group number SSSS is supplied to a two-dimensional Huffman encoding circuit 49, and the additional bit is supplied to a multiplexing circuit 51. Run length counter 47
Is a circuit for counting the run length of "0", and supplies the number NNNN of "0" between significant coefficients other than "0" to the two-dimensional Huffman encoding circuit 49. The Huffman encoding circuit 49 supplies the run length NNNN and the group number SSSS supplied to the multiplexing circuit 51 with an AC Huffman code obtained by subjecting the run length NNNN and the group number SSSS to variable length encoding in accordance with the AC Huffman code table 50.

The multiplexing circuit 51 has one block (8 × 8 pixels)
The DC Huffman code, the AC Huffman code and the additional bits are multiplexed and output to the output terminal 52 as compressed image data. Therefore, by storing the compressed data output from the output terminal 52 in the memory and decompressing the compressed data read from the memory by the reverse operation, the storage capacity of the memory can be reduced.

[0008]

However, the above-mentioned conventional example is an information amount compression system to which a combination of DCT and scalar quantization is applied. However, the combination of DCT and scalar quantization has the following problems. .

That is, when the above conventional example is applied to a facsimile, a printer, a storage device, and the like, it is necessary to compress to a certain extent at a low bit rate. When trying to compress the above conventional example to a low bit rate, the Q table is switched to perform coarse quantization, or the scaling factor of the Q table is increased, that is, the coarse coefficient is increased by increasing the coefficient by which the quantization step is multiplied. Will be changed.

FIG. 12 shows a reference Q table, FIG. 13 shows input information of a part of an image (3 × 4 blocks with 8 × 8 as one block), and FIG. 14 shows a DCT applied to the input information. FIG. 15 shows information obtained by performing scalar quantization on the transform coefficients with a step size obtained by multiplying the Q table shown in FIG. 12 by a scaling factor of 4.0. However, the DC component is independent of the scaling factor without using DPCM.
8-bit fixed length. FIG. 16 shows image information obtained by expanding (IDCT) the information shown in FIG.

As can be seen by comparing FIG. 13 with FIG.
Due to DCT and quantization for each block, the image information after expansion has distortion at the boundary between blocks. This is so-called block distortion. That is, compression at a low bit rate increases quantization errors, causes block distortion, and significantly lowers image quality.

The present invention has been made to solve the above-described problem, and it is possible to reduce the distortion at the block boundary by controlling the quantization of the orthogonal transform coefficient so as to cancel the quantization error, and to restore the data. An object is to reduce visual deterioration of an image.

[0013]

The present invention has the following configuration as one means for achieving the above object. An image processing apparatus according to the present invention includes: an input unit that inputs image information divided into blocks of a predetermined number of pixels; a conversion unit that orthogonally transforms image information in units of the blocks; and an AC component obtained by the orthogonal transformation. An image processing apparatus having quantization means for quantizing an orthogonal transform coefficient, wherein the quantization means generates, at least in the block, a quantization frequency generated when quantizing an orthogonal transform coefficient of a first frequency component. Detecting means for detecting an error, and a quantization error of the detected orthogonal transform coefficient of the first frequency component is generated such that a quantization error having a polarity opposite to that of the orthogonal transform coefficient of the first frequency component is generated. Also has control means for controlling the quantization of the orthogonal transform coefficient of the second frequency component present in the high frequency range,
The orthogonal transform coefficients of the first and second frequency components are odd function components in at least one of the horizontal and vertical directions in the block. The image processing method according to the present invention is characterized in that image information divided into blocks of a predetermined number of pixels is input, image information is orthogonally transformed in block units, and orthogonal transform coefficients of an AC component obtained by the orthogonal transformation are quantized. In the image processing method, at least the quantization, in the block, detects the polarity of the quantization error that occurs when quantizing the orthogonal transform coefficient of the first frequency component, the detected second detected The second frequency component orthogonal to the orthogonal transform coefficient of the first frequency component is generated in a higher frequency range than the orthogonal transform coefficient of the first frequency component so that a quantization error having a polarity opposite to that of the orthogonal transform coefficient of the one frequency component occurs. Controlling the quantization of the transform coefficients, the orthogonal transform coefficients of the first and second frequency components are odd function components in at least one of the horizontal and vertical directions in the block. That.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A preferred embodiment according to the present invention will be described below in detail with reference to the accompanying drawings. <First Embodiment> FIG. 1 is a main block diagram showing a first embodiment of the present invention. A description will be given based on FIG. In the figure, reference numeral 100 denotes an input terminal, and input image information which is blocked is sequentially input by raster scan. 1
Reference numeral 01 denotes a DCT circuit, which performs DCT conversion on information stored in a buffer (not shown) for one block. A zigzag scanning circuit 102 scans the DCT-transformed transform coefficients in a zigzag manner from a low-frequency component to a high-frequency component. From the zigzag scan circuit 102, the DCT
A conversion coefficient (referred to as F (i)) and an address (referred to as i) corresponding to zigzag scan data are output, and an address signal is input to the Q table 103. Reference numeral 103 denotes a Q table, which is constituted by, for example, an LUT (lookup table), and outputs information of a quantization step corresponding to the input address to the quantization circuit 104.

On the other hand, the DCT transform coefficient output from the zigzag scan circuit 102 passes through an adder 105 and is input to a quantization circuit 104. Reference numeral 104 denotes a quantization circuit. The quantization in the quantization circuit 104 is performed in the same manner as in the JPEG method described in the conventional example, C (i) = (F (i) + Q (i) / 2) / Q (i) (F (i) ≧ 0) ) C (i) = (F (i) −Q (i) / 2) / Q (i) (F (i) <0) i: address, F (i): DCT transform coefficient Q (i): quantum , C (i): quantization coefficient (1) The adder 105 will be described later. The quantization circuit 104 outputs a quantization coefficient C (i) and a value (Q (i) × C (i)) obtained by multiplying the quantization step by the quantization coefficient. Reference numeral 109 denotes an output terminal, and C (i) is transmitted to a block after quantization (not shown). Reference numeral 106 denotes a subtractor, which is a DCT transform coefficient that has not been quantized and a representative value above a threshold after quantization (quantization step × quantization coefficient).
Is a circuit for calculating the difference between the two. Reference numeral 107 denotes an addition value creation unit that inputs the difference value calculated by the subtractor 106,
Also, the address signal of the zigzag scan is input and LU
This is a ROM that outputs an added value corresponding to T (look-up table). Reference numeral 108 denotes a delay circuit, which is configured to be added to the DCT transform coefficient at a certain address after a predetermined delay. In this embodiment, in the circuit configuration described above, block distortion is reduced as described below.

FIG. 2 is a diagram for explaining zigzag scanning according to the first embodiment, and FIG. 3 is a diagram for explaining quantization according to the first embodiment.

FIG. 2A shows an address signal indicating the zigzag scan order in a block. Attention is now focused on the component at address 1. FIG. 2B shows a base image corresponding to the transform coefficient of the address 1 in the horizontal direction (hereinafter, referred to as the main scanning direction). For example, when the signal of address 1 can take a different value due to quantization, that is, when the AC power of address 1 increases and decreases, FIG. 2B shows amplitudes as shown in FIGS. 2C and 2D. Changes come out. This depends on whether the upper representative value is selected or the lower representative value is selected depending on whether the DCT transform coefficient is larger than the set threshold value, as shown in FIG. . The blockiness becomes visually poor in many cases where the image density is a single increase and a single decrease. That is, the odd function component is quantized as in the address 1, so that
2 (b) is amplitude-converted as shown in FIG. 2 (c) or (d), and occurs in a decompressed image because a tone jump occurs at a block boundary. Also in the address 2 of FIG. 2A, the vertical direction (hereinafter referred to as the sub-scanning direction)
The same can be said with. If any of the AC63 components in the block is an odd function in either the main scanning direction or the sub-scanning direction, it is caused by the occurrence of block distortion due to quantization, but the components of address 1 and address 2 which are most dominant in image degradation. Pay attention to. Originally, the DCT transforms the original function into an even function using a mirror image, and expresses only the cosine component as a superposition. The nature is different.

Now, suppose that the component at address 1 is quantized and the above representative value is selected. That is, the output of the subtractor 106 in FIG. 1 has a positive value. This is because the quantization coefficient × the quantization step is larger than the DCT transform coefficient.

Next, attention is paid to address 6, which is an odd function in the same main scanning direction.

FIG. 4 is a diagram showing a base image corresponding to the conversion coefficient in the main scanning direction at the address 6 according to the first embodiment.
Also in this case, for example, when the upper representative value is selected, the amplitude of FIG. 4A is selected as shown in FIG. 4B, and when the lower representative value is selected, as shown in FIG. The amplitude changes.

Now, when the upper representative value is selected at the address 1 and the state shown in FIG. 2 (c) is obtained, in addition, when the upper representative value is also selected at the address 6, that is, in FIG. In the state, the error at the block boundary is added, and the image is significantly degraded. That is, it is desired to offset the quantization error at the block boundary generated at the address 1 by the quantization error at the address 6 which is another odd function component. That is, if the upper representative value is selected at the address 1, it is sufficient to devise such that the lower representative value is selected at the address 6. This is the role of the adder 105 shown in FIG. If the output of the subtractor 106 for the address 1 is positive, a negative value corresponding to the input value is output by the LUT of the addition value creation unit 107, and after a delay of several addresses by the delay circuit 108, Since a negative value is added to the DCT coefficient, the lower representative value is easily selected. Then, the quantization error of both addresses 1 and 6 tends to cancel the error at the block boundary. Also, the subtractor 10
The same applies to the case where 6 outputs a negative value.
Outputs a positive value corresponding to the input value. By adding this positive value to the DCT coefficient of address 6, the above representative value is easily selected.

FIG. 5 is a diagram showing an example of a real image according to the first embodiment. FIG. 6 is a diagram showing quantization and ID according to the first embodiment.
It is a figure explaining CT. FIG. 5A shows information of a real image which has been blocked, FIG. 5B shows a transform coefficient obtained by DCT conversion, and FIG. 5C shows (quantization coefficient × (D) of FIG. 5 shows information after the decompression. In this example, since the upper representative value is selected for both the address 1 and the address 6 in FIG. 2A, a quantization error at a block boundary is added, and a tone jump with a neighboring block is expected. You. In FIG. 6A, it is detected that the representative value at the address 1 is selected, that is, in FIG.
The subtractor 106 calculates a quantization error of “+8” and inputs the result to the added value creating unit 107, so that the lower representative value is selected at the address 6. FIG. 6B
Is information obtained by decompressing (a) in the same figure, and cancels out the quantization error at the block boundary. As described above, since the DCT transform coefficient is added or subtracted, the quantization error tends to be equal to or larger than that, and the S / N ratio tends to be worse. However, since the distortion is brought to the center of the block, the block distortion decreases. I do.

Although the first embodiment has described the control of the address 6 by the quantization error of the address 1 in FIG. 2A, other odd functions such as the control of the address 9 by the quantization error of the address 2 are shown. But the same is true.

The hardware configuration shown in FIG. 1 is an example for embodying the present invention, and it goes without saying that another hardware configuration may be used.

In the first embodiment, the D of address 6
Even if the value of Q / 2 for adding / subtracting the CT conversion coefficient is changed, the quantization threshold is changed, which is equivalent.

Even if quantization is performed by raster scan without using zigzag scan, the generation of odd functions has regularity and can be easily realized.

<Second Embodiment> FIG. 7 is a diagram for explaining a quantization error canceling method according to a second embodiment.

In this embodiment, the processing contents are more limited than in the first embodiment shown in FIG. (A) of FIG.
Each solid line in (b), (c) and (d) indicates the base image in the component of the address i. Now, the address i of the odd function
Are quantized, the AC power changes,
As a result, it is assumed that the amplitude changes as indicated by the dotted line. In this case, in the first embodiment, the error of the block boundary is canceled by the quantization error of another odd function component, but in this embodiment, the amplitude decreases in the processing of the address j which is another odd function. Shall be performed only for Therefore, the above-mentioned zigzag scan address shown in FIG.
6 will be described.

The occurrence of a quantization error at address 1 is one of the patterns shown in FIGS. 7A and 7B, that is, when the representative value is selected, at the address 1, the DCT transform coefficient F (1) is obtained from the product of the quantization step Q (1) and the quantization coefficient C (1). ) Is positive, that is, the quantization error is positive. The cancellation of the quantization error at the block boundary at address 6 is shown in FIG.
(E) and (f), the quantization error increases as the amplitude increases, the contrast increases, and the image deteriorates. Conversely, as the amplitude decreases, MT
Although the image has a blurred F, the degree of image deterioration is often better than the former. That is, even if the S / N ratio is the same, the image quality changes depending on the quantization error.
Therefore, in the cases of FIGS. 7A and 7B, the processing of FIG. 7F is performed. That is, F (6) is subtracted only when the polarity of F (6) is F (6) ≧ 0.
(6) If <0, the signal is passed through the quantization circuit as it is. Similarly, when the quantization error of the address 1 occurs in FIG. 7C and FIG. 7D, the processing of FIG.

As described above, in the present embodiment, even when the quantization error of another odd function increases, the processing is performed in the direction of decreasing the amplitude. Image degradation is less noticeable. In addition, since the absolute value of the quantization coefficient decreases,
For example, when a variable-length code compression method is adopted like the JPEG method, it is useful for reducing the code amount.

The present invention may be applied to a system constituted by a plurality of devices or to an apparatus constituted by a single device. The present invention can also be applied to a case where the present invention is achieved by supplying a program to a system or an apparatus. As described above, according to the present embodiment, block distortion, which is visually noticeable, is reduced with a very simple configuration. Also, by quantizing in the direction of decreasing the amplitude, the code amount can be reduced when variable length coding is applied.

[0032]

As described above, according to the present invention,
By controlling the quantization of the orthogonal transform coefficients to cancel the quantization error, the distortion at the block boundary is reduced,
Visual deterioration of the restored image can be reduced.

[Brief description of the drawings]

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

FIG. 2 is a diagram illustrating a zigzag scan according to the first embodiment.

FIG. 3 is a diagram illustrating quantization according to the first embodiment.

FIG. 4 is a diagram showing a base image for a conversion coefficient in the main scanning direction at an address 6 according to the first embodiment.

FIG. 5 is a diagram showing an example of a real image according to the first embodiment.

FIG. 6 is a diagram illustrating quantization and IDCT according to the first embodiment.

FIG. 7 is a diagram illustrating a quantization error canceling method according to a second embodiment.

FIG. 8 is a block diagram showing a configuration of a conventional image compression apparatus.

FIG. 9 is a block diagram showing a configuration of the DPCM of FIG. 8;

FIG. 10 is a diagram illustrating a scan order of DCT coefficients according to a conventional example.

FIG. 11 shows an AC coefficient and a group number SSS according to a conventional example.
FIG. 4 is a diagram for explaining a relationship with S.

FIG. 12 is a diagram illustrating an example of a standard Q table.

FIG. 13 is a diagram showing an example of a 3 × 4 block representing input of a real image.

FIG. 14 is a diagram showing an example of a 3 × 4 block representing DCT of a real image.

FIG. 15 is a diagram showing an example of a 3 × 4 block representing quantization of a real image.

FIG. 16 is a diagram showing an example of a 3 × 4 block representing an IDCT of a real image.

[Explanation of symbols]

 1,100 Input terminal 2 Blocking circuit 17,101 DCT circuit 40 Q 41,103 Q table 42 DPCM 43 One-dimensional Huffman coding circuit 44 DC Huffman code table 45 Scan conversion circuit 46 Significant coefficient detection circuit 47 Run length counter 48 Group Circuit 49-dimensional Huffman coding circuit 50 AC Huffman table 51 multiplexing circuit 52, 109 output terminal 53 delay circuit 54, 105 adder 102 zigzag scan circuit 104 quantization circuit 106 subtractor 107 addition value creation unit 108 delay circuit 109 output Terminal

Claims (4)

(57) [Claims] An input unit for inputting image information divided into blocks each having a predetermined number of pixels; a conversion unit for orthogonally transforming the image information in block units; and an orthogonal transform coefficient of an AC component obtained by the orthogonal transform. An image processing apparatus having a quantization unit for performing quantization, wherein the quantization unit detects a quantization error generated when quantizing an orthogonal transform coefficient of a first frequency component in at least the block. Detecting means, such that a quantization error having a polarity opposite to the detected quantization error of the orthogonal transform coefficient of the first frequency component is generated, so that the quantization error is higher than the orthogonal transform coefficient of the first frequency component. Control means for controlling quantization of the orthogonal transform coefficient of the second frequency component that is present, wherein the orthogonal transform coefficient of the first and second frequency components is An image processing device characterized by an odd function component in at least one of the vertical direction and the vertical direction. 2. The orthogonal transform coefficient of the first frequency component is such that one of a horizontal or vertical component in the block is a fundamental wave component and the other is a DC component, and the orthogonal transform coefficient of the second frequency component is the horizontal or the <br/> one is odd function component other than the fundamental wave component of the vertical component, the image processing apparatus according to claim 1, wherein the other is characterized in that it is a DC component. 3. The method according to claim 1, wherein the control unit adds a value based on a quantization error of an orthogonal transform coefficient of the first frequency component to an orthogonal transform coefficient of the second frequency component. The image processing device according to claim 2. 4. An image processing method for inputting image information divided into blocks of a predetermined number of pixels, orthogonally transforming the image information in block units, and quantizing orthogonal transform coefficients of AC components obtained by the orthogonal transformation. In the quantization, at least in the block, a polarity of a quantization error generated when quantizing an orthogonal transform coefficient of a first frequency component is detected, and the detected first frequency component is detected. In order to generate a quantization error of the opposite polarity to the quantization error of the orthogonal transform coefficient of the second frequency component, the quantization of the orthogonal transform coefficient of the second frequency component present in a higher frequency range than the orthogonal transform coefficient of the first frequency component occurs. Image processing method, wherein the orthogonal transform coefficients of the first and second frequency components are odd function components in at least one of horizontal and vertical directions in the block. .