JP3065732B2 - Spatial quantization noise reduction method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention uses a solid-state image sensor such as a CCD (charge-coupled device) image sensor in which a plurality of light-receiving portions such as photodiodes having a photo / electric conversion function are arranged in an array. BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image reading method for reading and generating an image signal, and more particularly to a spatial quantization noise reduction method for reducing spatial quantization noise generated due to the spaced arrangement of the light receiving sections of the solid-state image sensor.

[0002]

2. Description of the Related Art Conventionally, techniques in such a field include:
For example, there is one described in the following literature. Reference 1: Video information, 20 [9] (1988-5) Industrial Development Organization KK, P. 35-39 Literature 2: "Digital Signal Processing of Images", First Edition (Showa 56-5-25) by Nikkan Kogyo Shimbun, p. 112-1
13, p. Conventionally, as an image reading method in digital signal processing of an image, an image is read using a solid-state image sensor, for example, a CCD image sensor, an image signal is generated, and the image signal is digitally processed. FIG. 2 shows a configuration example of this image sensor.

FIG. 2 is a configuration diagram of a CCD image sensor described in the above-mentioned document 1. The CCD image sensor 1 includes a plurality of light receiving units, for example, photodiodes 2, arranged in an array, a plurality of vertical transfer units 3 including a CCD shift register, a horizontal transfer unit 4 including a CCD shift register, and an amplifier. And an output unit 5.

In order to read an original with this kind of CCD image sensor 1, an electric signal obtained by photoelectric conversion by a photodiode 2 is sent to each vertical transfer unit 3. The electric signal transferred to the vertical transfer unit 3 is transferred to the horizontal transfer unit 4, sequentially transferred within the horizontal transfer unit 4, and an image signal is output from the output unit 5. The image signal output from the output unit 5 is 2
The image is binarized by a binarization circuit to generate a binary image, and digital signal processing is performed on the binary image.

Conventionally, in this type of image reading method,
In order to improve image quality, as described in Document 2, when performing binarization using a binarization circuit, the threshold level of binarization is changed while referring to the value of the previous bit. Thus, a binary image is generated, or an image signal is enhanced using a one-dimensional filter or a two-dimensional filter to improve image quality.

[0006]

However, the conventional image reading method has the following problems. In the conventional method for improving the image quality, for example, the image quality is improved by focusing on the signal of the photodiode 2 in FIG.
The relationship between input information and output value per pixel is not considered.

That is, as described in Document 1, the photodiode 2 in the CCD image sensor 1 is used.
The width W _p of the small compared to the pitch P of one pixel is common. As a result, a white pattern WN and a black pattern BN as shown in FIG.
(The white part is white, and the outputs are “0” and “1” respectively.)
Is incident on the image sensor 1, a pixel which is entirely black or white over one pixel is directly converted to "0" or "1" as shown in FIG. , The noise between the original image and the output is zero. But,
Black 50% in one pixel, the pixel 50% is incident pattern of white, would enter into the black pattern BN when the width W _p of the photodiode 2 is in this example. Therefore, “0” is output even though the incident light amount is 50%, and the difference is an error. FIG. 5 schematically shows how this error appears.

FIG. 5 is a view for explaining the generation of a spatial quantization error. 2 _n to 2 _{n + 3} ... Are photodiodes, 10 is an incident pattern, 10 a is an incident pattern width of one pixel, and 11 is an incident pattern per pixel. The light amount, 12 is the width of one pixel, 13 is the actual incident light amount / pixel, and e1 and e2 are errors.

As shown in FIG. 5, the n-th photodiode 2 _n and the (n + 1) -th photodiode 2 _{n + 1} have no error, but the (n + 2) -th photodiode 2 _{n + 2} has an error e1 and the (n + 3) -th photodiode 2. _An error e2 occurs at _{n + 3} . FIG. 6 shows how the errors e1 and e2 appear in the pixel array direction.

FIG. 6 is a diagram showing a spatial quantization error. A value S _{rms obtained} by taking the square root of the error e over the square of one pixel is:

[0011]

(Equation 1)

This corresponds to the RMS value (effective value) of noise between the original image and the binarized binary image. That is, the conventional image reading method has a problem that spatial quantization noise occurs. This kind of spatial quantization noise is generated particularly in the contour portion, and has a bad effect of deteriorating the image quality, and it has been difficult to solve it.

The present invention does not consider the relationship between input information and output value per pixel at all as a problem of the prior art, and spatial quantization noise is generated. It is an object of the present invention to provide a spatial quantization noise reduction method that solves the problem of image quality degradation at a contour portion.

[0014]

To solve the previous SL problems SUMMARY OF THE INVENTION The first aspect of the present invention, an image receiving unit that outputs a pixel signal by converting light / electric is in units of pixels, a plurality The following means are taken in an image reading method for generating an image signal by reading the image using an image sensor arranged in an array.

That is, the n-th light output from the adjacent first and second light receiving units in the image sensor, respectively.
Pixel signals V _n and n + (where n is any positive integer)
The first pixel signal V _{n + 1} or the (n−1) th pixel signal V
the difference ΔV _n of the _n-1 Ru asked Me. The first parameter arbitrarily selected is A (where A = 1−a, and a is the n-th parameter).
1 smaller emphasis coefficient of the pixel signal V _n emphasize), the second parameter is applied selected by random or pseudo-random number arbitrarily B (where A B = b / M, b is n + 1
Pixel signal V _{n + 1} or n−1 pixel signal V
Influence coefficient indicating the degree of influence of _n-1 , M is a random number or pseudo
As random number of), so as to perform the calculation processing for correcting (performs A · B · ΔV _n) multiplication, 該画 Motoshingo V _n by adding the multiplication result before outs Motoshingo V _n I have.

According to a second aspect, the emphasis coefficient a of the first aspect is
The values are between 0.15 and 0.4. According to a third aspect, in the spatial quantization noise reduction method according to the first or second aspect, when the pixel signals V _n−1 , V _n , and V _{n + 1} are two-dimensionally arranged, The arithmetic processing is executed in the horizontal or vertical direction, or in the horizontal and vertical directions with respect to the array.

[0017]

In accordance with the first invention, since it is configured spatial quantization noise reduction method as described above, by correcting the images signals V _n by the operation processing, to some extent 該画 image signal V _n Random correction can be performed while maintaining the same. This allows
The density average of the image at the outline portion approaches the original image, spatial quantization noise is reduced, and the image quality at the outline portion of the image can be improved.

According to the second aspect of the invention, by setting the value of the enhancement coefficient a in the range of 0.15 to 0.4, it is possible to accurately reduce the spatial quantization noise. According to the third aspect, substantially the same noise reduction can be achieved regardless of whether the arithmetic processing of the first aspect is performed in the horizontal direction or the vertical direction. Further, by performing the arithmetic processing of the first invention in both the horizontal direction and the vertical direction, accurate noise reduction corresponding to various images can be performed. Therefore, the above problem can be solved.

[0019]

FIG. 1 is a functional block diagram of a main part of an image reading processing apparatus for implementing a spatial quantization noise reduction method according to an embodiment of the present invention. This image reading processing device includes an image sensor 20 such as a CCD image sensor 1 as shown in FIG. This image sensor 2
Numeral 0 denotes a light receiving portion for outputting a pixel signal by performing a photoelectric conversion on an image in pixel units, for example, a plurality of photodiodes 21 are arranged in a two-dimensional array. The n-th pixel signal V _n (n
= 1, 2, 3,..., K) is the (n + 1) th pixel signal V _{n + 1}
Is subtracted by the subtractor 31 from the subtractor 31 and the subtracted value ΔV _n (= V
_{n + 1−} V _n ) is multiplied by the first parameter A (= 1−a) in the multiplier 32. “a” is an enhancement coefficient for enhancing the n-th pixel signal V _n .

The multiplication result of the multiplier 32 is multiplied by the second parameter B (= b / M) and the multiplier 33, the multiplication result is added by the adder 34 and the n-th pixel signal V _n, calculating A later corrected new n-th pixel signal V _Nn is output.

In the second parameter B (= b / M), b is an influence coefficient indicating the degree of influence of the (n + 1) th pixel signal, and M is the number of pseudorandom numbers such as a dither Bayer array. This pseudo random number is generated by a random number generator 35. Next, a description will be given of a spatial quantization noise reduction method according to the present embodiment using such an image reading apparatus.

First, an image is read by the image sensor 20, and among the pixel signals V ₁ ,..., V _n , V _{n + 1} ,.
_n is provided to the subtractor 31 and the adder 34, and n
The (+1) th pixel signal V _{n + 1} is supplied to the subtractor 31.
The subtracter 31, according to the following equation (1) and is represented by the sum of (2) (3) of the arithmetic expression, obtaining the subtraction value [Delta] V _n.

[0023]

(Equation 2)

In equation (3), the n-th pixel signal V _n and n
If the (+1) th pixel signal V _{n + 1} is equal, the corrected new n-th pixel signal V _Nn becomes the n-th pixel signal V _n
Is to be equal to Equation (3): emphasis coefficient a
Is arbitrarily selected under the condition of a <1. The subtraction value ΔV _n obtained by the subtractor 31 is the first parameter A (= 1
−a) is multiplied by a multiplier 32, and the multiplication result is multiplied by a second parameter B (= b / M) by a multiplier 33.
Then, with respect to n th pixel signal V _n, the multiplication result of the multiplier 33 in the adder 34 is added, it corrected new n-th pixel signal V _Nn is output.

Here, the second parameter B (= b / M)
Is set to an arbitrary value from 1 to M. M is the number of pseudo-random numbers generated by the random number generator 35. FIG. 7 is a diagram showing an example of a pseudo random number (that is, an example of a coefficient map of the influence coefficient b), and FIG. 8 is a diagram showing an example of a fetched data map.

As shown in FIG. 7, for example, in the case of a 4 × 4 Bayer array pseudo random number, M = 16. Coefficient map of influence coefficient b shown in FIG. 7, for example, will be described with reference to the data map as shown in FIG. 8, i-th row and j the th data x _n coefficient maps b i-th row and j Assign a column. In this case, b = 1. In the case of the (i + 1) th row and the (j + 2) th row, the data x1 _{n + 2} is assigned so that b = 15 is assigned. In the example of the coefficient map in FIG. 7B, a 4 × 4 map is repeatedly arranged in both the horizontal direction and the vertical direction, and is assigned to each data.

FIGS. 9A to 9D are diagrams showing calculation examples when the pseudo random numbers shown in FIG. 7 are applied. FIG. 9 (a)
Indicates the ratio of the density of the original image. “1” is white and “0” is black. A density of 0.5 arranged on the right diagonal line indicates that half of the black and white pattern is input. 4x4
The average density x in the matrix is 8/16.

FIG. 9B shows an example of binarization when the original image of FIG. 9A is read using the conventional CCD image sensor 1 of FIG. The average density x at this time is 6/16
Thus, the original image becomes darker, and the noise execution value S _{rms at} that time is 25%. FIG. 9C shows the result of calculation using the value of FIG. 9B and the arithmetic expression of (3) with a = 0.5. The calculation at this time is performed in the horizontal direction. The average density x in FIG. 9C is 7 × 22/32 ×
1/16, and the noise from the original image is reduced to 5%. Display on a display device such as a CRT capable of halftone,
Alternatively, if printing is performed by a printer, a density per unit area close to the original image can be obtained.

In FIGS. 9A to 9C, it is considered that the ratio of black and white in one pixel of the original image randomly occupies a value between 0 and 1, and the average of the entire image is 0.5. Therefore, in this example, the ratio of black and white was described as 0.5.

FIG. 9 (d) is a drawing not shown in FIG. 9 (c).
This shows an example of binarization by a binarization circuit. The binarization threshold level V _{T is set} to 1 when V _T > 13/32.
FIG. Average density x is 8/1
6, which is equal to the original image. This is made possible by randomly giving the influence coefficient b.

FIG. 10 shows the emphasis coefficient a and the effective value S _{rms of the} noise when the pseudo random numbers having the Bayer arrangement shown in FIG. 7 are used.
FIG. 14 is a diagram showing a calculation result of the relationship of FIG. That is, FIG.
_Shows the state of the noise execution value S _rms when the emphasis coefficient a is changed. The best line segment is a line in which noise in a column, a row, or a right and left oblique line in FIG. 7 becomes the smallest. The worst line segment indicates a case where the contour portion is on the largest line. The average is the average of those 16 line segment noises. As is apparent from FIG.
The vicinity of 0.3 is small as the first noise.

FIG. 11 is a diagram showing a pseudo random number map b _ij arranged as a pseudo random number to reduce the variation.
FIG. 12 shows a calculation result of the relationship between the enhancement coefficient a and the noise execution value S _rms when this map b _ij is applied. In addition,
In the above description, the case where the output of the image sensor 20 of FIG. 1 is a binary value of “1” and “0” has been described. examples is corrected by the contrast difference n th pixel signal V _n, it can be applied above method of reducing.

Next, the width of the photodiode 21 constituting the image sensor 20 of FIG. 1 is finite, and the ratio of the photodiode width to the pixel pitch will be considered. FIG. 13 shows a black-and-white pattern using the image sensor 20.
FIG. 4 is an explanatory diagram of a width of a photodiode 21 and a spatial quantization error when a value is converted.

In FIG. 13, Pa is the pixel pitch, and W is the width of the photodiode 21. If V _w is a white output level of the photodiode 21 and V _TA is a threshold level for determining whether the photodiode output value is “1” or “0”, the effective photodiode width W1 is W1
= V _TA / V _w · W The range of ± 1/2 · (P _a −W1) is a part where a spatial quantization error occurs. For example,
If V _TA = 0.6, W1 = 0.6W for black, and the effective photodiode width for white is W2 = (1−1).
0.6) W = 0.4W. In other words, the way in which the error occurs for white and black changes. Here, for simplicity of description, the case where W _TA = 0.5 and W 1 = W 2 will be described.

FIG. 14 is an explanatory diagram of the effective photodiode width W1. It is considered that the value within the effective photodiode width W1 is a true value, and that an error occurs in other places. The following equation (4) is a correction equation for the spatial quantization error when the photodiode width W is finite.

[0036]

(Equation 3)

In the equation (4), the emphasis coefficient a in the equation (3)
There only replace a1 = a + (1-a ) · W1 / P a, the same correction expression.

FIG. 15 is an explanatory diagram of the photodiode width and the spatial quantization error when the photodiode 21 outputs a gradation. In FIG. 15, the gray scale number at the time gradation output at photodiode 21 is shown an example when the 2 ^K. Regarding the error at the time of gradation output, the effective diode width only becomes W1. Since the emphasis coefficient a1 can be selected arbitrarily, the expression (4) can be applied regardless of the photodiode width.

The same noise as that shown in FIG. 10 can be obtained when the influence coefficient b shown in FIG. 7 is used. That is, the pixel signal V
The noise when the spatial quantization noise is corrected using _n and V _{n + 1} has a certain constant value. What differs depending on the value of the photodiode width is the spatial quantization noise.

[0040]

(Equation 4)

The pattern is also shown in FIG.
When the photodiode width is large, the space quantization noise is small.

As described above in detail, the spatial quantization noise reduction method of the present embodiment has the following advantages. (A) The method of reducing the spatial quantization error of the present method will be described with reference to FIG. Selecting the value of a1 to 0.5 or less, for the effective photodiode width W1 and the ratio of the pixel pitch P _a W1 / P _a is 0.8 or less of the image sensor 20, the worst value of the corrected quantization error Even this is smaller than the conventional spatial quantization noise, and is reduced to about 65% as compared with the conventional spatial quantization noise. a
If the value of 1 is selected between 0.15 and 0.4, the average of the corrected quantization errors will be 50% or less of the conventional value. This value is W
This corresponds to 1 / P = 0.9.

That is, the method of the present embodiment is equivalent to randomly modulating the photodiode width at the contour of an arbitrary image, and the correction amount is proportional to the contrast of the contour. Therefore, it is possible to accurately reduce the spatial quantization noise that occurs randomly in the contour portion. (B) Since it acts on the outline portion of the image, it works more effectively for fine images. (C) As described in (a) above, this corresponds to a change in the photodiode width, so that the apparent photodiode arrangement pitch also changes randomly, and moire can be reduced.

The present invention is not limited to the above embodiment,
Various modifications are possible. For example, there are the following modifications. (I) In the equation (3), the value of the influence coefficient b is 1 to M
If the average density becomes approximately 0.5, any value may be randomly assigned. In the above embodiment, a pseudo random number composed of a Bayer array of dither is used. However, a pseudo random number other than the Bayer array may be used, or the random number itself may be used instead of the pseudo random number. (Ii) In the above embodiment, a monochrome pattern has been described, but color can be handled similarly. In particular, in a color image sensor (CCD) with an on-chip filter,
This is more effective because the color pixels are further apart. Further, in the case of a color image or the like, if the method of the above-described embodiment is applied to at least one of the signals of each color such as blue (B), green (G), and red (R), noise is reduced as compared with the related art. it can. (iii) In the above embodiment, the calculation in the horizontal direction is performed, but the calculation in the vertical direction may be performed. Further, depending on the content of the image, the calculation may be performed in the horizontal direction and then in the vertical direction, or vice versa. By sequentially calculating both directions in this way, spatial quantization noise can be further reduced. (Iv) In place of the subtraction value ΔV _n in the equation (3), n−
Focusing on the _first pixel signal V _{n− 1} ,
be used subtracted value _{ΔV n = V n -V n-} 1 which is calculated with _n,
Functions and effects similar to those of the above embodiment can be obtained. (V) The arithmetic processing unit in the image reading processing apparatus of FIG. 1 may be configured as an individual circuit using an integrated circuit or the like, or may be executed under program control using a microcomputer or a digital signal processor (DSP). You may do it. Further, the image sensor 20 uses another one-dimensional or two-dimensional image sensor such as a MOS image sensor, or the light-receiving unit 21 included in the image sensor 20 is configured by another light / electric conversion unit other than the photodiode 21. Is also good.

[0045]

[Effect of the Invention] As described above in detail, according to the first invention, the first and second parameters A, performs a calculation using the B, since to correct the image Motoshingo V _n , the
While maintaining some the field Motoshingo V _n, it can be corrected at random. Therefore, spatial quantization noise can be accurately reduced, and the image quality at the outline of the image can be improved. Further, it works more effectively on the output of an image sensor or the like in which a color image such as a color image sensor is spaced apart, and can reduce false color contours. In addition, the degree of occurrence of moire can be reduced for fine images and the like. Therefore, spatial quantization noise can be effectively reduced in a still image or the like, and the invention can be applied to various image signal processing circuits for contour correction.

In the second invention, the value of the enhancement coefficient a is set to 0.1
Since the setting is made in the range of 5 to 0.4, the spatial quantization noise can be reduced accurately.

According to the third aspect, by performing the arithmetic processing of the first aspect in the horizontal or vertical direction, spatial quantization noise when a two-dimensional image sensor is used can be accurately reduced. If the arithmetic processing is performed in both the horizontal direction and the vertical direction, the spatial quantization noise can be reduced more accurately.

[Brief description of the drawings]

FIG. 1 is a schematic functional block diagram of an image reading processing apparatus for implementing a spatial quantization noise reduction method according to an embodiment of the present invention.

FIG. 2 is a configuration diagram of a conventional CCD image sensor.

FIG. 3 is a diagram illustrating an example of an incident pattern with respect to the CCD image sensor of FIG. 2;

FIG. 4 is a diagram showing a binarized output of the CCD image sensor of FIG. 2;

FIG. 5 is an explanatory diagram of generation of a conventional spatial quantization error.

FIG. 6 is an explanatory diagram of a conventional spatial quantization error.

FIG. 7 is a diagram illustrating an example of a pseudo random number according to the embodiment;

FIG. 8 is a diagram illustrating an example of a data map in which the example of the pseudo-random numbers in FIG. 7 is imported;

FIG. 9 is a diagram illustrating a calculation example according to the present embodiment.

FIG. 10 is a diagram illustrating a calculation result of a relationship between an enhancement coefficient a and a noise S _{rms according to} the present embodiment.

FIG. 11 is a diagram illustrating an example of a pseudo random number map b _{ij according to} the present embodiment.

FIG. 12 is a diagram illustrating a calculation result of a relationship between an enhancement coefficient a and a noise S _{rms according to} the present embodiment.

FIG. 13 is an explanatory diagram of a photodiode width and a spatial quantization error when the photodiode output of this embodiment is binary.

FIG. 14 is an explanatory diagram of an effective photodiode width of the present embodiment.

FIG. 15 is an explanatory diagram of a photodiode width and a spatial quantization error when the photodiode output of this embodiment is a grayscale output.

[Explanation of symbols]

20 image sensor 21 photodiode _{V n-1, V n,} V n + 1 pixel signal 31 subtractor 32 multiplier 34 adder 35 random number generator A, B first, second parameters a emphasis coefficient b influence coefficient M Number of pseudorandom numbers

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int. Cl. ⁷ , DB name) H04N 1/40-1/409 H04N 1/46 H04N 1/60

Claims (3)

(57) [Claims] 1. An image reading apparatus comprising: a light receiving unit that outputs a pixel signal by performing optical / electric conversion on an image in pixel units and uses an image sensor arranged in a plurality of arrays to read the image and generate an image signal In the method, an n-th (where n is an arbitrary) output from the adjacent first and second light receiving units in the image sensor, respectively.
Pixel signals V _n and n + 1 th pixel signal V positive integer)
A difference ΔV _n from the ( _{n + 1} ) th or ( n−1) th pixel signal V _n−1 is obtained, and the first parameter arbitrarily selected is represented by A (where
A = 1−a, where a emphasizes the n-th pixel signal V _n
A second parameter arbitrarily selected and to which a random number or a pseudo-random number is applied is represented by B (where,
B = b / M, where b is the (n + 1) th pixel signal V _{n + 1}
Or the degree of influence of the (n-1) th pixel signal V _n-1
(A is the number of random numbers or pseudo-random numbers)
· B · ΔV _n) performs multiplication of the multiplication result is added before outs Motoshingo V _n to perform calculation processing for correcting the 該画 Motoshingo V _n, the spatial quantization noise reduction, characterized in that Method. 2. The spatial quantization noise reduction method according to claim 1, wherein the value of the emphasis coefficient a is 0.15-0. 4. The spatial quantization noise reduction method that is 4 . 3. The spatial quantization noise reduction method according to claim 1, wherein the pixel signals V _n−1 , V _n , and V _{n + 1} are arranged two-dimensionally. A spatial quantization noise reduction method, wherein the arithmetic processing is performed in a horizontal direction or a vertical direction, or in a horizontal direction and a vertical direction.