JP3199786B2 - Image processing device - 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 for compressing an image by orthogonal transformation.

[0002]

2. Description of the Related Art The memory capacity required for storing a halftone image such as a photograph in a memory is (number of pixels) × (number of gradation bits), which is enormous for storing a high-quality color image. Memory capacity was required. For this reason, various information amount compression methods have been proposed, and the amount of information is compressed and then stored in a memory to reduce the memory capacity.

FIG. 9 shows a baseline system (basic system) proposed by JPEG (Joint Photogaraphic Experts Group) as an international standard for color still image coding.
(Yasuda "International Standardization of Color Still Image Coding" Journal of the Institute of Image Electronics Engineers of Japan, Vol. 18, No. 6, pp. 398-409, 1989
It is a block diagram which shows a structure of (year).

In FIG. 9, halftone image data input from an input terminal 1 is cut out into a block of 8 × 8 pixels (hereinafter, referred to as “pixel block”) in a blocking circuit 2 and discrete cosine transform (hereinafter, referred to as “DCT”). )), And the transform coefficients are supplied to a quantizer (hereinafter referred to as “Q”) 40. In Q40, linear quantization of the transform coefficients is performed according to the quantization step information supplied from the quantization table (hereinafter, referred to as “Q table”) 41. The DC coefficient among the quantized transform coefficients is calculated by a difference (prediction error) from the DC coefficient of the previous pixel block in a predictive coding circuit (hereinafter referred to as “DPCM”) 42, and the difference is calculated by one-dimensional Huffman coding. It is supplied to the circuit 43.

FIG. 10 is a block diagram showing a detailed configuration of the DPCM 42. The DC coefficient quantized by Q40 is supplied to delay circuit 53 and subtractor 54. The DC coefficient input to the delay circuit 53 is delayed by a time necessary for the DCT circuit 17 to perform an operation for one pixel block. Therefore, the delay circuit
From 53, the DC coefficient of the previous pixel block is supplied to the subtractor 54, and the subtracter 54 outputs the difference (prediction error) between the DC coefficient of the current pixel block and the DC coefficient of the previous pixel block. In the predictive coding described here, the DPCM 42 includes a delay circuit 53 because the value of the previous pixel block is used as the predicted value. The one-dimensional Huffman encoding circuit 43 supplies the multiplexing circuit 51 with a DC Huffman code obtained by performing variable length encoding on the prediction error signal supplied from the DPCM 42 according to the DC Huffman table 44.

On the other hand, the AC coefficient (D
The coefficients other than the C coefficient) are
As shown in FIG. 11, zigzag scanning is performed in order from the low-order coefficient, and supplied to the significant coefficient detection circuit 46. The significant coefficient detection circuit 46 determines whether the quantized AC coefficient is zero or a non-zero significant coefficient, and if it is zero, supplies a count-up signal to the run length counter 47 and increases the count value by +1. Let it.
If the AC coefficient is a significant coefficient other than zero, the 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. The run length counter 47 is a circuit for counting the run length of zero, and supplies the number of zeros (run length) NNNN existing between the significant coefficient and the next significant coefficient to the two-dimensional Huffman encoding circuit 49. The grouping circuit 48 divides the AC coefficient into a group number SSSS and additional bits shown in FIG. 12, supplies the group number SSSS to the Huffman encoding circuit 49, and supplies the additional bits to the multiplexing circuit 51.

The two-dimensional Huffman encoding circuit 49 supplies the multiplexing circuit 51 with an AC Huffman code obtained by variable-length encoding the supplied run-length NNNN of zero and the group number SSSS of the significant coefficient according to the AC Huffman code table 50. I do. The multiplexing circuit 51 includes a DC Huffman code for one pixel block,
The AC Huffman code and the additional bits are multiplexed, and the output terminal 52 outputs compressed image data. Therefore,
The compressed image data output from the output terminal 52 is stored in a memory, read out from the memory, and decompressed by a reverse operation, so that the image memory capacity can be reduced.

In general, a halftone image such as a photograph input by an image scanner or the like is orthogonally transformed by DCT or the like, and when quantized, significant coefficients tend to concentrate in a low band of a pixel block and a high band coefficient is high. Is often zero. The above-described variable-length coding operation is performed up to the last significant coefficient when the AC coefficient of the pixel block is zigzag scanned. Therefore, as the high-frequency AC coefficients become zero and the last significant coefficient becomes closer to the lower frequency, data compression by variable-length coding is more efficiently performed.

[0009]

In the case of compressing image data at a low bit rate, the orthogonal transformation is performed by simply changing the quantization step width while the representative quantization value is set at the middle of the quantization step width. A method of coarsely quantizing at least a part of the coefficient range is used. When coarse quantization is performed, so-called mosquito noise is generated near the edges of the expanded image in areas such as edges where the AC power is large, but the AC power is small, and the flat parts of the original image do not cause image quality deterioration. Ringing occurs, and the image quality is significantly degraded.

The present invention has been made to solve the above-described problem, and has as its object to suppress the deterioration of image quality due to quantization.

[0011]

The present invention has the following configuration as one means for achieving the above object.

An image processing apparatus according to the present invention provides orthogonal transform means (for example, an orthogonal transform means for orthogonally transforming an image to obtain orthogonal transform coefficients).
A detecting means (corresponding to the DCT circuit 102 shown in FIG. 1) for detecting the magnitude of a high-frequency component of the orthogonal transform coefficient (for example, FIG.
, And quantization means for quantizing at least a high-frequency component of the orthogonal transform coefficient using a predetermined quantization step (for example, FIG. 1).
And the quantization means sets a quantization threshold as an intermediate point of the quantization representative value when the magnitude of the high-frequency component is equal to or smaller than a predetermined value (for example, If the magnitude of the high-frequency component exceeds the predetermined value, quantization is performed to set the quantization threshold to a value other than the intermediate point of the quantization representative value (for example, FIG. 3).
(Equivalent to the quantization shown in Fig. 4).

Further, orthogonal transform means (corresponding to, for example, the DCT circuit 102 shown in FIG. 5) for orthogonally transforming an image to obtain orthogonal transform coefficients, and detecting means for detecting the magnitude of a high-frequency component of the orthogonal transform coefficients (For example, the Q table selection circuit 11 shown in FIG. 5
0) and a quantization means (for example, corresponding to the Q table groups 111 and Q106 shown in FIG. 5) for quantizing the entire frequency space of the orthogonal transform coefficients using a predetermined quantization step group. If the high-frequency component does not exceed a predetermined value, the quantization means may use a quantization step group in which the quantization threshold is an intermediate point of the quantization representative value (for example, as shown in FIG. 3). (Corresponding to quantization), and in some cases, quantization is performed by a group of quantization steps whose quantization threshold is other than the intermediate point of the quantization representative value (for example, corresponds to the quantization shown in FIG. 8). And

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment according to the present invention will be described below in detail with reference to the drawings.

[0015]

[First Embodiment] FIG. 1 is a block diagram showing a configuration example of a first embodiment. The purpose of this embodiment is to reduce mosquito noise generated when quantizing orthogonal transform coefficients, and the method of encoding orthogonal transform coefficients is the same as the above method. Therefore, FIG. 1 shows details of a configuration necessary for quantization, and omits details of a configuration necessary for encoding.

In FIG. 1, 100 is an input terminal, 101 is a block circuit, 102 is a DCT circuit, 103 is a zigzag scan circuit, 104 is an AC power value detection circuit, 105 is a Q table, and 106 is
Q and 107 are encoding sections, and 108 is an output terminal. The halftone image data input from the input terminal 100 is cut out into a pixel block of 8 × 8 pixels by the blocking circuit 101, orthogonally transformed by the DCT circuit 102, and the transform coefficient is supplied to the zigzag scan circuit 103. .

In the zigzag scan circuit 103, FIG.
The orthogonal transform coefficient is zigzag-scanned from the low frequency band to the high frequency band, and the orthogonal transform coefficient F (i) and the address i representing the position of the orthogonal transform coefficient F (i) in the block are output. . The orthogonal transform coefficient F (i) is determined by the AC power value detection circuit 1
04 and Q106, the address i is the AC power value detection circuit 1
04.

The AC power value detection circuit 104 compares the power value (or simply the absolute value) of the orthogonal transform coefficient F (i) with a predetermined threshold value. However, the AC power value detection circuit 104 determines from the input address i whether the input orthogonal transform coefficient F (i) is a low-frequency component or a high-frequency component, and determines whether the address i has a predetermined value. In the following case (low frequency side), a result is output that is smaller than the threshold value regardless of the power value of the orthogonal transform coefficient F (i). The comparison result in the AC power value detection circuit 104 and the address i are supplied to the Q table 105.

The Q table 105 is composed of an LUT (look-up table) or the like, and has the input address i and A
C: Add value A corresponding to the comparison result of power value detection circuit 104
(I) and the quantization step Q (i) are sent to Q106. Q
Reference numeral 106 also includes an LUT and outputs a quantization coefficient C (i) by the following calculation. However, in the following equation, i is an address, F (i) is an orthogonal transform coefficient, Q (i) is a quantization step, and A (i) is an addition value.

C (i) = {F (i) + A (i)} / Q (i) (F
(I) ≧ 0) C (i) = {F (i) −A (i)} / Q (i) (F
(I) <0) The quantization coefficient C (i) output from Q106 is encoded by the encoding unit 107, and the compressed image data is output to the output terminal 10.
8 is output. Next, a method of realizing low-noise quantization will be described.

FIG. 2 shows a base image corresponding to a certain high-frequency orthogonal transform coefficient F (i) (in FIG. 2, for simplicity, it is described in a one-dimensional direction). When a quantization error is added to the base image of FIG. 2 (a) by quantization, the amplitude indicated by the broken line in FIG. 2 (b) increases, or the amplitude indicated by the broken line in FIG. 2 (c) decreases. become. Whether the amplitude increases or decreases due to the error depends on whether the absolute value of the orthogonal transform coefficient F (i) is larger than the set quantization threshold or not. It depends on which is output as the quantization coefficient C (i).

At the time of quantizing the high-frequency orthogonal transform coefficient F (i), if the “upper representative value” is selected, the orthogonal transform coefficient F (i) is selected.
The amplitude of the image component relating to (i) increases (the contrast of the image is enhanced), and image quality degradation called mosquito noise occurs in which a set of white and black points starts to appear near the image edge. Conversely, when the "lower representative value" is selected, the amplitude of the image component is reduced and the contrast of the image is reduced, but no mosquito noise is generated, and generally, the deterioration is not noticeable and the user does not feel discomfort.

Therefore, in the present embodiment, low-noise quantization is realized by devising the high-frequency component in the DCT space to be quantized to a "lower representative value". Specifically, by changing the added value A (i) output from the Q table 105 according to the power value of the high-frequency orthogonal transform coefficient F (i), the orthogonal transform coefficient F (i) is changed to “ It is quantized to the following representative value.

Hereinafter, the specific setting of the addition value A (i) will be described. For example, assuming that the added value A (i) is 1/2 of the quantization step Q (i) (A (i) = Q (i) / 2), the quantization threshold value is an intermediate point between the representative values, and the orthogonal transform is performed. The result obtained by performing an operation equivalent to rounding on the coefficient F (i) is output as a quantization coefficient C (i). For example, when A (i) = 0, the orthogonal transform coefficient F (i)
The result of the operation equivalent to the truncation is the quantization coefficient C
Output as (i). That is, by changing the added value A (i) between 0 and Q (i) / 2, the degree of easiness of quantization to the “lower representative value” can be adjusted.

FIG. 3 is a diagram showing an example of the linear quantization used in the above-described JPEG system, and FIG. 4 is a diagram showing an example of the nonlinear quantization of the present embodiment. 3 and 4, the vertical axis represents the quantization coefficient C (i) output from Q106, the horizontal axis represents the orthogonal transform coefficient F (i) input to Q106, and the scale marked on the horizontal axis represents the quantization step. Represents Q respectively. FIG. 3 shows linear quantization in which the quantization threshold is set to the midpoint of the representative value with the addition value A (i) = Q (i) / 2, and quantization equivalent to rounding is always performed.

FIG. 4 shows an example of quantization for a certain high-frequency component in the DCT space. When the absolute value of the orthogonal transform coefficient F (i) is less than a certain value, that is, | F (i) | ≦
In the case of 2Q, the addition value A (i) = Q (i) / 2, and quantization corresponding to rounding is performed. Orthogonal transform coefficient F (i)
Is larger than a certain value, that is, | F
(I) When |> 2Q, the addition value A (i) is set to 0, and quantization corresponding to truncation is performed.

That is, in an image recording apparatus such as a printer, a limit value of an orthogonal transform coefficient F (i) at which mosquito noise due to quantization is generated is first experimentally obtained, and the obtained limit value and the orthogonal transform coefficient F (i) are obtained. (I) is compared by the AC power detection circuit 104. From the above comparison, the orthogonal transform coefficient F
When (i) is equal to or less than the limit value, from the Q table 105,
The quantization step Q (i) and the added value A (i) are supplied to Q106 so that quantization corresponding to rounding is performed.
At 106, quantization is performed to minimize the quantization error. On the other hand, when the orthogonal transform coefficient F (i) exceeds the limit value as a result of the above comparison, the quantization step Q (i) and the added value A (i) are obtained from the Q table 105 so as to perform quantization corresponding to truncation. The quantization is supplied to Q106, and the mosquito noise is suppressed in Q106. However, when the orthogonal transform coefficient F (i) is a low-frequency component, regardless of the orthogonal transform coefficient F (i), the AC power value detection circuit 1
04 outputs a result that is less than or equal to the limit value.
Is performed to minimize the quantization error.

In FIG. 4, when | F (i) | ≦ 2Q, the added value A (i) = Q (i) / 2, and | F (i) |> 2
An example in which the addition value A (i) = 0 at the time of Q has been described. However, the present embodiment is not limited to this, and 0 ≦ A (i) ≦
Needless to say, the addition value A (i) satisfying Q (i) / 2 can be set. In FIG. 1, the AC power value detection circuit 104 is connected to a Q table 105 or
OM conversion is also considered as an application of this embodiment.

Also, in the above description and FIG. 1, an example is described in which orthogonal transform by DCT of 8 × 8 pixels is used. However, the present embodiment is not limited to this, and an arbitrary pixel block size and an arbitrary pixel block size are used. An orthogonal transformation by a method can be applied. As described above, according to the first embodiment, even when coarse quantization is performed, ringing called mosquito noise does not occur near the edge of an expanded image, and an image processing apparatus with little image quality degradation is provided. be able to.

[0030]

Second Embodiment Hereinafter, a second embodiment according to the present invention will be described. In the second embodiment, the same components as those in the first embodiment are denoted by the same reference numerals, and the detailed description thereof will be omitted. No.
In one embodiment, the quantization step Q (i) and the sum A are supplied to Q106 in accordance with the AC power value of each orthogonal transform coefficient F (i).
Although (i) was adjusted to an appropriate value, in the second embodiment, an appropriate Q table was selected from a plurality of Q tables and quantized to Q106 according to the state of the high-frequency orthogonal transform coefficient F (i). Q (i)
Is to supply.

FIG. 5 is a block diagram showing a configuration example of the second embodiment. In FIG. 5, 110 is a Q table selection circuit, and 111 is a Q table group. Operation from the input terminal 100 to the zigzag scan circuit 103, and Q10
The operation from 6 to the output terminal 108 is the same as in the first embodiment, and a description thereof will be omitted. 6 and 7 are diagrams illustrating an example of the quantization table.

FIG. 6 shows an example of a standard quantization table.
The quantization corresponding to the rounding described in the first embodiment is performed. FIG. 7 shows an example of a quantization table suitable for quantizing high-frequency components, in which quantization corresponding to truncation described in the first embodiment is performed. The Q table selection circuit 110
When there is no orthogonal transform coefficient F (i) equal to or larger than a predetermined threshold value in the high frequency band in the CT space, for example, a quantization table as shown in FIG.
Performs quantization corresponding to rounding over the entire area of the DCT space as shown in FIG. 3, for example.

On the other hand, when there are orthogonal transform coefficients F (i) that are equal to or larger than a predetermined threshold value in the high frequency band in the DCT space, for example, the quantization table shown in FIG.
Q106 is, for example, DC as shown in FIG.
Quantization equivalent to truncation is performed on the entire region of the T space to suppress generation of mosquito noise. Generally, in image edges and line drawing characters, not only the high-frequency but also the low-frequency AC power in the DCT space increases, so as described above, performing the quantization corresponding to the truncation over the entire DCT space requires the code This reduces the pre-encoding quantization coefficient and contributes to a reduction in the amount of information after encoding.

As described above, according to the second embodiment, as in the first embodiment, even when coarse quantization is performed, ringing called mosquito noise is generated near the edge of the expanded image. Therefore, it is possible to provide an image processing apparatus with little image quality deterioration. Further, there is also an effect of reducing the amount of encoded information such as image edges, line drawings, and characters. The present invention may be applied to a system including a plurality of devices or to an apparatus including a single device. Needless to say, the present invention can be applied to a case where the present invention is achieved by supplying a program to a system or an apparatus.

[0035]

As described above, according to the present invention,
If the magnitude of the high-frequency component of the orthogonal transform coefficient is equal to or less than a predetermined value, the quantization threshold is set to the midpoint of the quantization representative value.If the magnitude of the high-frequency component exceeds the predetermined value, the quantization threshold is set to the quantization representative. By performing quantization other than the intermediate point of the value, or when there is no higher-frequency component of the orthogonal transform coefficient exceeding a predetermined value, a quantization step group in which the quantization threshold is the intermediate point of the representative quantization value Accordingly, in some cases, the quantization is performed by a group of quantization steps in which the group of quantization thresholds is other than the intermediate point of the representative quantization value, so that the deterioration of the image quality in quantization can be suppressed. For example, by performing quantization corresponding to rounding or truncation according to the magnitude of the high frequency component of the orthogonal transform coefficient, it becomes possible to suppress the occurrence of mosquito noise and the like.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration example of an embodiment according to the present invention.

FIG. 2 is a diagram illustrating an example of a quantization error of a base image according to the present embodiment.

FIG. 3 is a diagram illustrating a general quantization example.

FIG. 4 is a diagram illustrating an example of quantization in the present embodiment.

FIG. 5 is a block diagram showing a configuration example of a second embodiment according to the present invention.

FIG. 6 is a diagram illustrating an example of a general quantization table.

FIG. 7 is a diagram illustrating an example of a quantization table according to the present embodiment.

FIG. 8 is a diagram illustrating an example of quantization in the present embodiment.

FIG. 9 is a block diagram showing a conventional configuration.

FIG. 10 is a diagram illustrating a configuration of a conventional predictive encoding circuit.

FIG. 11 is a diagram showing a state of a conventional forward scan of DCT coefficients.

FIG. 12 is a diagram illustrating a conventional relationship between an AC coefficient and a group number.

[Explanation of symbols]

 Reference Signs List 101 Blocking circuit 102 DCT circuit 103 Zigzag scan circuit 104 AC power value detection circuit 105 Quantization table (Q table) 106 Quantizer (Q) 107 Encoding unit 110 Q table selection circuit 111 Q table group

Continuation of the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) H04N ^7/ 24-7/68 H04N 1/41-1/419 H03M 7/30 JICST file (JOIS)

Claims (6)

(57) [Claims] 1. An orthogonal transformation means for orthogonally transforming an image to obtain an orthogonal transformation coefficient, a detection means for detecting a magnitude of a high-frequency component of the orthogonal transformation coefficient, and at least a high-frequency component of the orthogonal transformation coefficient, Quantization means for performing quantization using a predetermined quantization step, wherein the quantization means sets a quantization threshold to an intermediate point of the quantization representative value when the magnitude of the high frequency component is equal to or smaller than a predetermined value. An image processing apparatus, wherein when the magnitude of the high frequency component exceeds the predetermined value, quantization is performed to set the quantization threshold to a value other than the intermediate point of the quantization representative value. 2. The method according to claim 1, wherein the quantizing unit quantizes the orthogonal transform coefficients by adding or subtracting a variable value to or from the orthogonal transform coefficients and dividing by a predetermined quantization step. Item 1. The image processing device according to item 1. 3. The image processing apparatus according to claim 2, wherein the variable value satisfies one-half or less of the predetermined quantization step. 4. The image processing apparatus according to claim 1, wherein the quantization unit performs quantization corresponding to truncation when the size of the high-frequency component exceeds the predetermined value. 5. An orthogonal transformation means for orthogonally transforming an image to obtain an orthogonal transformation coefficient, a detection means for detecting a magnitude of a high-frequency component of the orthogonal transformation coefficient, and an entire frequency space of the orthogonal transformation coefficient, Quantization means for performing quantization using a predetermined quantization step group, wherein the quantization means sets the quantization threshold to an intermediate value between the quantization representative values when none of the high frequency components exceeds a predetermined value. An image processing apparatus, wherein quantization is performed by a group of quantization steps, which are points, and in some cases, by a group of quantization steps whose quantization threshold is other than an intermediate point of the quantization representative value. 6. The image processing apparatus according to claim 5, wherein said quantizing means performs quantization corresponding to truncation when said high frequency component exceeds a predetermined value.