JPH0575862A - Quantization error reduction method - Google Patents

Abstract

PURPOSE:To reduce quantization noise and to improve picture quality by adding a noise signal having an arbitrary number of bit within a prescribed area at random using an addition means in the multilevel width of a quantization signal quantized by an A/D converter. CONSTITUTION:A n bit quantization signal S20 which continuous signals are quantized by an A/D converter 20 is added to a noise signal S20 at random by an addition means 40. The average value of the level of a noise signal S30 added by the addition means 40 is added and subtracted by an addition and subtraction means 50. The output of the addition means 40 or the output of the addition and subtraction means 50 is converted into an analog signal by a D/A converter 60. The output of the addition means 40 or the output of the addition and subtraction means 50 is calculated by an arithmetic means 70 and the only n bit in the arithmetic result is outputted. Further, the reproduction of the quantization signal is performed by removing a low-order bit i of the n bit quantization signal quantized by the A/D converter 20 by a removing means 110 and by again adding the signal to a m bit noise signal S30 by the addition means 40.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a quantization error reducing method for reducing a quantization error generated during analog / digital conversion (hereinafter referred to as A / D conversion) in a signal processing method such as digital signal processing of an image. It is a thing.

[0002]

2. Description of the Related Art Conventionally, as a technique in such a field, there is one described in the following documents, for example. Reference 1; Norihiko Fukibuki, "Digital signal processing of images", first edition (Sho 56-5-25), Nikkan Kogyo Shimbun, P. 60-61 Reference 2; Tetsuji Hiwatari, "Image Engineering Handbook", first edition (1
986-5-20) Asakura Shoten, p. 52-53 Document 3; Mitsuo Kaji, "Printing and Image Engineering," first edition (Sho 63-6-
15) Printing Society Press, P. 298-303 Conventionally, for example, in the digital signal processing of an image, there is the one as shown in FIG. Figure 2
FIG. 9 is a block diagram showing a configuration example of a conventional image signal processing device. This image signal processing device has an image sensor 1 such as a CCD (charge coupled device) image sensor for reading an image, and an A / D converter 2 is connected to the output side thereof. The output side of the A / D converter 2 has an arithmetic circuit 3
, And a display device 5 such as a CRT is connected via a digital / analog converter (hereinafter referred to as a D / A converter) 4.

In this type of apparatus, the image sensor 1 reads information about a subject whose brightness is continuous. The read image signal S1 is quantized by the A / D converter 2, the quantized signal S2 is processed by the arithmetic circuit 3 for digital signal processing, and the D / A converter 4 converts it into an analog signal, which is then displayed. It is displayed on the device 5.

FIG. 3 is an explanatory diagram of the quantization error in FIG. As shown in this figure, the image signal S1 obtained by reading the continuous amount with the image sensor 1 is quantized by the A / D converter 2 into the gradation signal (quantized signal) S2 in the quantization step S. An error occurs, which causes quantization noise. When such quantization noise occurs, for example, when the luminance change is gentle, the luminance step causes a false contour on the screen of the display device 5, which causes a problem that the image quality is deteriorated.

In order to solve such a problem, in the signal processing method of the above-mentioned reference 2, in order to reduce the occurrence of false contours due to quantization noise, noise of 1/2 quantization level is previously added to the image signal S1. A dither method (halftone display method) in which quantization is performed by the A / D converter 2 after superposition is adopted. Further, in the signal processing method of Document 3, a threshold signal is created according to a certain rule, and the A / D
A dither method or the like in which the converter 2 binarizes is adopted.

[0006]

However, the conventional signal processing method has the following problems.

(A) The conventional method has a drawback that quantization noise is directly superimposed. Further, there are problems such as level accuracy when adding 1/2 quantization level in an analog manner, complexity when handling an analog signal such as noise superposition, and complexity of making multivalued threshold voltage.

(B) In binarizing an image by a dither matrix using the dither method, usually, after converting into a digital signal by the A / D converter 2, the values are digitally compared. Be seen. However, such a method has a problem that the quantization noise is directly superposed.

Quantization error occurs because continuous values are cut at a certain quantization step S, as shown in FIG. Assuming that the true value is So, the quantization noise is Se, and the quantized value (quantized signal) is Sb in the entire image, they have a relationship as in the following expression (1). So-Se = Sb (1) Then, the quantization noise Se can be expressed by the following equation (2).

[0010]

[Equation 1]

For example, in 4-bit A / D conversion,
According to the equation (2), the quantization error Se of about 1.8% is superimposed on the peak value. If the signals quantized with the same number of gradations are calculated,

[0012]

[Equation 2]

The noise level becomes.

The present invention has the problems that the prior art has, for example, deterioration of image quality due to superposition of quantization noise, and complication of processing due to analog handling for preventing deterioration of image quality. The present invention provides a quantization error reduction method that solves points and the like.

[0015]

[Means for Solving the Problems] FIGS.
It is a functional block diagram for explaining the composition of the 1st-7th invention. In the quantization error reducing method in the first invention of FIG. 1A, the continuous signal S10 is converted into an A / D converter 20.
In the signal processing method for performing signal processing using the n-bit quantized signal S20 quantized in step S1, the addition unit 40 is used to add an arbitrary bit number m in the gradation width of the quantized signal S20.
The noise signal S30 is added randomly within a predetermined area.

In the quantization error reducing method of the second invention shown in FIG. 1B, the average value of the levels of the noise signal S30 added to the quantized signal S20 by the adding means 40 of the first invention is added / subtracted. At 50, addition and subtraction are arbitrarily performed.

In the quantization error reducing method of the third invention shown in FIG. 1 (c), the output of the adding means 40 of the first invention is
The (n + m) -bit D / A converter 60 converts the analog signal. In the quantization error reducing method in the fourth invention of FIG. 1 (d), the output of the addition / subtraction means 60 of the second invention is converted into an analog signal by the (n + m) -bit D / A converter 60. There is.

In the quantization error reducing method according to the fifth aspect of the invention shown in FIG. 1 (e), the output of the adding means 40 of the first aspect of the invention is operated by the operating means 70 and only n bits in the operation result are output. I am trying. In the quantization error reducing method according to the sixth invention of FIG. 1 (f), the output of the addition / subtraction means 50 of the second invention is operated by the operation means 70, and only n bits in the operation result are output. There is. In the quantization error reducing method according to the seventh invention of FIG. 1 (g), signal processing for performing signal processing using an n-bit quantized signal S20 obtained by quantizing a continuous signal S10 by an A / D converter 20. In the method, the following measures are taken. That is, A / D
After removing the lower bit i of the n-bit quantized signal S20 by the converter 20 by the removing unit, an arbitrary bit is added to the (n−i) -bit quantized signal using the adding unit 40 in the gradation width. A number of noise signals S30 are randomly added within a predetermined area to reproduce the quantized signal.

[0019]

According to the first aspect of the invention, since the quantization error reducing method is configured as described above, the n-bit quantized signal S20 obtained by quantizing a continuous signal by the A / D converter 20 is added. The means 40 adds the noise signal S30 at random. As a result, the quantization error can be reduced.

In the second invention, the average value of the levels of the noise signal S30 added by the adding means 40 is added and subtracted by the adding means 50. As a result, a signal with an arbitrary number of bits can be output.

In the third and fourth inventions, the output of the adding means 40 or the output of the adding / subtracting means 50 is converted into an analog signal by the D / A converter 60. Accordingly, if an output device such as a display device or a printer is connected to the output side of the D / A converter 60, image display without quantization noise can be performed.

In the fifth and sixth aspects of the invention, the output of the adding means 40 or the output of the adding / subtracting means 50 is operated by the operating means 70, respectively, and only n bits in the operation result are output. This makes it possible to reduce quantization noise even when a large number of quantized signals are calculated.

According to the seventh aspect of the invention, the lower bit i of the n-bit quantized signal quantized by the A / D converter 20 is removed by the removing means 110, and the addition means 40 again converts it into the m-bit noise signal S30. The addition is performed and the quantized signal is reproduced. As a result, the increase in the quantization error caused by the truncation of the lower bits of the quantized bit by the removing unit 110 is compensated for by the superposition of the noise signal S30 by the adding unit 40, and the data compression and the reproduction thereof with less quantization error are performed. You can do it. Therefore, the above problem can be solved.

[0024]

First Embodiment FIG. 4 is a functional block diagram of an image signal processing apparatus for carrying out a quantization error reducing method according to a first embodiment of the present invention. This image signal processing device has an image sensor 10 such as a CCD image sensor that outputs a continuous signal (image signal) S10 by photoelectrically / electrically converting a continuous image such as a picture and outputs the image. A / D converter 20 on the side
Are connected. The A / D converter 20 is a circuit that quantizes a continuous signal S10 and outputs an n-bit quantized signal S20, and n signal lines 21 are connected to the output side thereof. In addition, this signal processing device includes an m-bit noise signal S including an operator for generating a pseudo signal.
A noise source 30 for generating 30 is provided, and m signal lines 31 are connected to the output side thereof.

The n signal lines 21 and the m signal lines 31 are connected to the D / A converter 60 via the adding means 40. The adding means 40 divides the 1-gradation width (1/2 ^{n of the} total amplitude) of the quantized signal S20 into 2 ^m under the least significant bit of the n signal lines 21 so that the m signal lines are divided. 31
Is added configuration. The D / A converter 60 is a circuit that converts the (n + m) -bit signal from the adding means 40 into an analog signal, and the output side of the D / A converter 60 has a display device through a voltage conversion circuit 86 for driving the display device. 90 is connected. The display device 90 is composed of a CRT or the like capable of analog luminance modulation.

5A and 5B show the noise source 3 of FIG.
0A is a diagram showing a configuration example of 0, FIG. 10A is an explanatory diagram when the noise source 30 is configured by an operator using a Bayer array, and FIG. It is a partial view which shows the example of arrangement | sequence of a child.

In FIG. 5, for simplification of description, the number of quantization bits of the continuous signal S10 is n = 4 (that is, the number of gradations is 16), and the noise source 30 is 4 × 4 with m = 4. The dither signal of the Bayer array of is capable of being superposed.

The function of the operator shown in FIG.
As shown in (b), for example, for the signal at the address (A, a), one gradation width of 1/16 is added to the original signal, and 8 /
The value obtained by subtracting one gradation width of 16 = 1/2 acts as a new value of the address. Similarly, at the address (D, d), one gradation width of 6/16 is added, and 1/2 gradation width is subtracted so that the new value of the address is obtained.
As shown in FIG. 5B, in the 4 × 4 Bayer array, the repeated arrangement of this set is made to act on the original signal. That is, the address (A, a), (E, a),
(A, e), ... Have the same value.

Next, referring to FIGS. 6A to 6C,
A quantization error reducing method in the signal processing device of FIG. 4 will be described. 6A to 6C are diagrams showing numerical examples of each pixel after calculation in the signal processing device of FIG. 4, and FIG. 6A shows quantization with 4 bits (16 gradations). An example of the quantized signal S20 that has been generated, and FIG.
FIG. 6C is a diagram showing an example after 8-bit display (256 gradation display) using the display device 90.

For example, an image consisting of a continuous picture is read by the image sensor 10, and its continuous signal (image signal) S10 is quantized by the A / D converter 20 to obtain an n-bit quantized signal S20. Are output to the n signal lines 21.
In the example of the quantized signal S20 shown in FIG. 6A, it is shown that the eighth gradation and the seventh gradation, which are almost intermediate values of 16 gradations, are output to the n signal lines 21. There is.

When the operator of FIG. 5 which constitutes the noise source 30 is operated, the noise signal S30 of m bits is output from the noise source 30 to the m signal lines 31. The n signal lines 21 and the m signal lines 31 are added via the (n + m) signal lines 41 forming the adding means 40. The (n + m) -bit signal after the addition of the noise signal S30 is
It is shown in FIG.

The (n + m) -bit signal to which the noise signal is added is converted into an analog signal by the D / A converter 60, converted into a voltage by the voltage conversion circuit 80, and then displayed on the display device 90 in an analog manner. To be done. 8 with this display device 90
An example of bit display is shown in FIG.

As shown in FIGS. 6A to 6C, a 4-bit (16 gradations) quantized signal S20 is output by the display device 90 as 8 bits.
Since bits (256 gradations) are displayed, this means that the 16 gradation signals have been converted into 256 gradation signals. This 16
The average value Z of the densities of the gradation signal and the 256 gradation signal is both 118, and the same density can be seen in the square of the 4 × 4 Bayer array. However, since the noise signal S30 is added to the quantized signal S20 by the adding means 40 as in the present embodiment, the level 8 to the level 7 are added in FIG. 6A.
While the way to change to 1/16 changes,
In FIG. 6B, the change is 1/256.
Therefore, there is an advantage that the unevenness of the island-shaped false contour, which is a conventional defect, disappears because the change in the amount of change is small, and the unevenness in shade in the contour portion of the image displayed on the display device 90 becomes inconspicuous. This is for the following reason.

For example, consider a case where a picture whose true value is So is read and displayed in consideration of the entire image. This picture is A /
When the value quantized by the D converter 20 is Sb and the quantization noise is Se, the above equation (1) is established. An example of this quantized value Sb is shown in FIG. That is, when only the quantized value Sb is displayed on the display device 90 as it is, it is displayed as an image including the quantization error Se. Since the quantization error Se occurs randomly in the entire image,
It can be said that the quantization error X = (Σx) shown in the equation (2) is random. Therefore, in the present embodiment, when the image is displayed, the noise signal S30 is superimposed on the quantized signal S20 and displayed as a signal Sb1 = Sb + Se1 ...
Is being reduced. Here, Sb1 in equation (3)
Is random noise generated arbitrarily, and is noise selected to be substantially equal to the quantization noise Se.

When the noise source 30 is formed by using the operator shown in FIG. 5, the quantization noise Se is about 1.7%. The quantization noise Se when a continuous picture is quantized is
According to the equation (2), it is about 1.8%, and almost equal noise values are obtained. However, the noise signal S30 is ideally randomly added, not in the Bayer array as described above.
And the two values are equal.

A continuous picture to be quantized and FIG.
The noise e between the display image shown in (c) and

[0037]

[Equation 3]

[0038] Therefore, a continuous picture is simply A / D
In contrast to the noise of 1.8% when quantized by the converter 20 and displayed on the display device 90, the noise can be reduced to about 1/3 in this embodiment.

Further, in FIG. 4, (n
Since the + m) -bit signal is converted into an analog signal by the D / A converter 60 and then displayed on the display device 90, the second term (−1 / n · 1/2) shown in FIG. ) Adjusts the reference signal of the display device 90 by that amount without calculation, so that an image with good image quality can be obtained.

Second Embodiment FIG. 7 is a functional block diagram of an image signal processing apparatus showing a second embodiment of the present invention. Elements common to those in FIG. Has been done. In this signal processing device, a memory 100 for temporarily storing data is provided between the output side of the A / D converter 20 and the input side of the adding means 40. Further, the addition / subtraction means 5 is provided on the output side of the addition means 40.
The calculation means 70 is connected via 0. The adder / subtractor 50 inputs a (n + m) -bit signal in which the m-bit noise signal S30 is superimposed on the n-bit quantized signal S20, and arbitrarily adds and subtracts the average value of the levels of the m-bit noise signal S30. , And has a function of giving the addition / subtraction result to the arithmetic means 70. The calculation means 70 has a function of calculating the output of the addition / subtraction means 50, cutting m bits in the calculation result, and outputting only n bits.

Consider that the signal processing apparatus of FIG. 7 uses some quantized signals S20 to perform calculations and the like. For example, three quantized signals S with the same number of bits S
If you use 20, the noise

[0042]

[Equation 4]

It becomes Each quantized signal S20 is 4
In the case of bits, the noise is about 3%. On the other hand, in the present embodiment, the noise signal S30 is added to the quantized signal S20 quantized by the A / D converter 20 by the addition means 40.
And the (n + m) -bit signal is added and subtracted by
After addition and subtraction by 0, the arithmetic means 70 performs arithmetic operation to output only the necessary n-bit signal. Thus, in this embodiment, the noise signal S30 is added to the quantized signal S20.
Is superposed, so that when the desired calculation or the like is performed using the addition / subtraction means 50 and the arithmetic means 70, it is treated as a statistic amount of about 1
% Noise and one quantized signal S20
Noise is smaller than 1.8%.

Therefore, in the present embodiment, even when a color signal handling three primary colors is quantized by the A / D converter 20 and the quantized signal S20 is processed, an image can be reproduced with low noise. The noises for the three primary colors may have different shapes.

Next, it is considered that after calculating the three primary colors, only the upper 4 bits are taken out in order to obtain the original number of bits of the signal. This is equivalent to re-quantization. The output of the adder / subtractor 50 is calculated by the calculator 70, and only the upper 4 bits of the calculation result are extracted. Then,
Since the noise between the signal calculated by the calculation means 70 and the original picture to be quantized is about 1%, and the quantization noise Se is 1.8%, requantization using the calculation means 70 is performed. The noise Ssai in

[0046]

[Equation 5]

Therefore, it is possible to obtain a signal reduced to about 2/3 of the conventional noise.

Further, when a 4-bit signal containing, for example, about 2% noise is displayed on the display device 90 as shown in FIG. 4, another adding means 40 may be provided and noise may be superimposed again for display. .. In this case, the noise Sdis

[0049]

[Equation 6]

It can be reduced to a certain degree. This is because the lower 4 bits of the result calculated using the three quantized signals S20 are considered to be new random noise, and when this is cut and displayed on the display device 90, new random noise is added. Even so, they are statistically equal.

Third Embodiment FIG. 8 is a functional block diagram of an image signal processing apparatus showing a third embodiment of the present invention, in which elements common to those in FIGS. 4 and 7 are common. The reference numeral is attached. In this signal processing device, the n number of signal lines 21 connected to the output side of the A / D converter 20 are added to the adding means 40 via the removing means 110.
The noise source 30 is connected to the adding means 40 via m signal lines 31 and the output of the adding means 40 is output to the display device 90. Removal means 1
Reference numeral 10 has a function of removing the lower bit i of the quantized signal S20 of n bits and giving it to the adding means 40. The removing means 110 is, for example, (n−i) signal lines 111.
It is composed of. In the signal processing device of FIG. 8, for example, the least significant 1 bit of the 6-bit quantized signal S20 is removed by the removing means 110, which is sent to the adding means 40 via the five signal lines 111, and further the noise source 30. Consider a case in which a noise signal S30 of 5 bits is generated from and the noise signal S30 of 5 bits is given to the adding means 40 via the five signal lines 31.

In the first or second embodiment, the quantization error is reduced to about 1/3. This means that the number of quantization bits is equivalent to about 1.6 bits more. Therefore, for example, even if the image reproduction of the n-bit quantized signal S20 is performed using the (n-1) -bit signal line 111, the image can be reproduced as a 0.6-bit image with less noise.

Therefore, in this embodiment, the least significant 1 bit of the 6-bit quantized signal S20 is removed by the removing means 110, and the 4-bit signal is sent to the adding means 40 via the signal line 111. In the noise source 30, the gradation width is 1 of the full amplitude.
/ 2 ^n-1 , that is, a 4-bit noise signal S30 is given to the adding means 40 via the signal line 31. The adding means 40 adds the signal of the 5-bit signal line 111 and the noise signal S30 of the 4-bit signal line 31 and sends the 9-bit reproduction signal to the display device 90 for display on the display device 90. ..

In this embodiment, the lower bits of the original quantized bits are discarded by the removing means 110, and the increase in the quantization error caused thereby is compensated by superimposing the noise signal S30 by the adding means 40. Quantization error can be reduced and a high quality image can be obtained.

The present invention is not limited to the above embodiment, and instead of the display device 90 shown in FIG. 4, for example, another output device such as a printer capable of gradationally or analogically modulating the print density is provided. Alternatively, various modifications are possible such as applying the above-described embodiment to a signal processing device other than an image.

[0056]

As described in detail above, according to the first invention, a random noise signal is added to a quantized signal obtained by quantizing a continuous signal by an A / D converter by using addition means. Since it is digitally superimposed within one quantization gradation width, the quantization noise can be accurately reduced. Moreover, since it is possible to configure only a digital circuit that can be handled easily without handling analog signals, it is possible to reduce the number of bits of the A / D converter and simplify the circuit configuration.

According to the second aspect of the invention, since the addition / subtraction means arbitrarily adds / subtracts the average value of the levels of the superimposed noise signals, the addition / subtraction means can output a signal having an arbitrary number of bits. The range of application of the device connected to the output side of the adder / subtractor becomes wider.

According to the third and fourth aspects of the invention, the output of the adding means or the output of the adding / subtracting means is changed to D of (n + m) bits.
Since I tried to convert it to an analog signal with the / A converter,
For example, if a display device, a printer, or the like is connected to the output side of the D / A converter, the quantization noise between the original image to be quantized and the displayed image can be reduced, and high-quality image display can be achieved. It will be possible. Moreover, it is possible to prevent the image quality from being deteriorated only by a digital circuit which is easy to handle without handling an analog signal. Therefore, the number of bits of the A / D converter can be reduced, and high-quality image display that is easy to manufacture and has good reproducibility can be performed.

According to the fifth and sixth inventions, the output of the adding means or the output of the adding / subtracting means is operated by the operating means, and only n bits in the operation result are output.
For example, the quantization noise can be accurately reduced even when a large number of quantized signals are calculated. At this time, the quantization error can be further reduced by treating only the number of bits of the original signal as a signal and then again superimposing a random noise signal as in the first invention. Further, even when the B, G, and R primary colors are calculated and stored using a color image sensor or the like, the quantization error can be efficiently reduced, so that the number of quantization bits can be reduced and the number of low bits can be reduced. High image quality can be obtained with the A / D converter.

According to the seventh aspect of the invention, the lower bits of the original quantized bits are truncated by the removing means, and the increment of the quantization error caused by the truncation is random noise of an arbitrary number of bits by using the adding means. The signal is superimposed to compensate for it. Therefore, it is possible to compress the quantized bit number by removing the original quantized bit number by the removing means, and transmit it to reproduce the quantized bit number by the adding means. Then, a quantized image signal can be sent, and if the sent signal is displayed on a display device, a printer or the like, a high quality image can be obtained.

[Brief description of drawings]

FIG. 1 is a diagram showing a functional block diagram of the present invention, and (a) to (g) in FIG. 1 are functional block diagrams showing the first to seventh inventions.

FIG. 2 is a functional block diagram showing a conventional image signal processing device.

FIG. 3 is an explanatory diagram of a quantization error in FIG.

FIG. 4 is a functional block diagram of an image signal processing apparatus showing the first embodiment of the present invention.

5 is a diagram showing a configuration example of a noise source of FIG.

FIG. 6 is a diagram showing a numerical example of each pixel after the calculation in FIG.

FIG. 7 is a functional block diagram of an image signal processing apparatus showing a second embodiment of the present invention.

FIG. 8 is a functional block diagram of an image signal processing apparatus showing a third embodiment of the present invention.

[Explanation of symbols]

 20 A / D converter 30 noise source 40 adding means 50 adding / subtracting means 60 D / A converter 70 computing means 110 removing means S10 continuous signal (image signal) S20 quantized signal S30 noise signal

Claims (5)

[Claims] 1. A signal processing method for performing signal processing using an n-bit quantized signal obtained by quantizing a continuous signal by an analog / digital converter, wherein an adding means is included in the gradation width of the quantized signal. Is used to randomly add a noise signal having an arbitrary number m of bits within a predetermined area. 2. The quantization error reduction method according to claim 1, wherein an average value of the levels of the noise signal added to the quantized signal is arbitrarily added or subtracted by an addition / subtraction means. Method. 3. The quantization error reducing method according to claim 1, wherein the output of the adding means or the output of the adding / subtracting means is (n +
m) A quantization error reducing method characterized by converting into an analog signal with a bit / digital converter. 4. The quantization error reducing method according to claim 1, wherein the output of the adding means or the output of the adding / subtracting means is operated by the operating means, and only n bits in the operation result are output. And a quantization error reduction method. 5. A signal processing method for performing signal processing using an n-bit quantized signal obtained by quantizing a continuous signal with an analog / digital converter, wherein a lower bit i of the n-bit quantized signal is removed. After removal by the means, a noise signal of an arbitrary number of bits is randomly added within a predetermined area by using an adding means in the gradation width of the (n−i) -bit quantized signal, and the quantized signal is added. A method for reducing a quantization error, characterized in that