JPS62112477A - Method and apparatus for binarizing image signal - Google Patents

Abstract

PURPOSE:To suppress notch at its minimum while displaying the effect of high frequency range emphasis sufficiently at matrix size 3X3 and to reproduce a thin line too by executing the gradation conversion of density data and then executing the binary-encoding of the converted data at the binarizing processing of the density data obtained by reading out an original. CONSTITUTION:When an original is read out by a reader having characteristics for outputting 255 as maximum density data, information 0-255 for respective picture elements are entered. After executing gradation conversion for picture elements thicker than the density data 150, high frequency range emphasis is executed by unsharpness masking method based upon 3X3 matrix size and a weight factor k=3.0 and the binarization of a floating threshold is executed by using the emphasized result. Namely, the 3X3 matrix consisting of adjacent picture elements is formed after the gradation conversion, the average E of density of nine picture elements in the matrix is found out and I' is obtained by emphasizing the high frequency of a noticed picture element on the basis of operation expressed by I=I+3.0(I-E). The I' is defined as the threshold of the noticed picture element at a current point of time and the noticed picture element is binary-coded to black if I>=I' and white if I<I'.

Description

Detailed Description of the Invention (1) Industrial Application Field Kokumei is a method for converting an image signal obtained by reading a document into a binary image signal, and an apparatus for implementing the method. Regarding.

(2) Background of the Invention As such a binarization method, there are the following methods.

(a) Fixed threshold method A method in which density data obtained by reading an original is binarized using a certain fixed threshold. That is, if the density data of a pixel of interest is larger than a fixed threshold value, the pixel of interest is determined to be black, and conversely, if the density data of the pixel of interest is smaller than the fixed threshold value, it is determined to be white.

(b) A method of increasing or decreasing the threshold value based on the binarized information of the matrix that has already been determined to be black or white when binarizing the density data obtained by reading the notchless processed original. When determining the white/black of the pixel of interest, refer to the white/black to the left of the pixel of interest.1. If it is a black pixel, the threshold value is lowered by a certain density so that the pixel of interest tends to become black, and conversely, if it is a white pixel, the threshold value is lowered (l) so that the pixel of interest tends to become white.
After iI is reduced to a certain concentration, it is binarized. This is called horizontal notch processing, and vertical notchless processing refers to a method that similarly refers to the black and white of pixels in the same column as the pixel of interest on one of the already binarized scanning lines in the row where the pixel of interest exists. The method of referring to both pixels is called water/vertical notchless processing.

(c) Floating amount ([f1 binarization method When binarizing 'aa data obtained by reading a manuscript, first a certain size consisting of pixels in the vicinity of the pixel of interest (a degree is I) Assuming a matrix of
The density average of the pixels in the matrix (note this, T'=1-)'k(+-E) (where l (weighting coefficient)) is used to emphasize the high frequency range at the pixel of interest. Obtain the value I'. This method of high-frequency emphasis is called the unsharpness mask method. The value l' obtained by high-frequency emphasis using this unsharpness mask method is set as the threshold at the fourth pixel, and the Ig I compare the pixel of interest with 2
How to value it.

(d) Combination method A method in which the method described in (C) is combined with the notchless treatment described in (b).

(3) Problems to be solved by the invention Regarding the methods (a) and (b) mentioned above, in order to binarize using a fixed threshold value, the image must have good contrast and no shading, and the change over time must be Otherwise, there is a problem that a slight shift in the threshold value will cause a large change in the line width of the reproduced image, and depending on the threshold value, thin lines may not be binarized correctly and may disappear. To improve this, it is necessary to find the optimal threshold value in real time, but the histogram method, the method that uses differential values, etc., which exist as methods to find a more optimal threshold value, are all non-real-time binarization methods. It is difficult to find the optimal threshold in real time.

Although the methods (C) and (d) solve the above-mentioned problems, they have the problem that the image quality changes depending on the matrix size. As an example, the characteristics of the high-frequency emphasis filter when the reading density is 32 dat/mm are expressed as follows for each matrix size NXN.
Shown in Figure 2. Generally, 41ine pair/
It is said that it is better to emphasize around mm ~ 81ine pair/mm, and from this point on, the appropriate matrix size is about 5x5 ~ 9x9, but if the matrix size is 5x5 or more, the hardware will be large-scale. , and it takes too much time to process with software. On the other hand, the high frequency enhancement by matrix size 3X3 works most strongly at line widths around 141ine pair/mm, which are almost unreadable.
Pixels with low reliability are emphasized, and even when notchless processing is performed, notches are noticeable on thin lines or diagonal lines.
Roughness is noticeable near the high concentration area due to the characteristics of the leader. Particularly in semiconductor photodetectors (such as CCDs and MOS diode arrays), the operating range relative to the amount of light is narrow, so roughness becomes noticeable in high-density areas, that is, areas where the amount of light is low.

(4) Means for Solving the Problems The present invention has been made in view of the above points, and has two matrix sizes of 3 x 3, which minimizes the notch while sufficiently comparing the effects of high frequency emphasis. The purpose is to be able to reproduce fine lines, and in order to achieve this purpose, when performing 2lfi processing on the data obtained by scanning the original, the density data is Perform key conversion,
After that, it is configured to perform the 21st shift.

(5) The cause of the notch in the example is that the signal/noise ratio is small when reading a black image, and the thin line disappears due to the spatial frequency of the optical system and photodetector at the time of reading. This is because the characteristics deteriorate in the high frequency range.

Therefore, in the present invention, the following processing is performed.

The first stage of gradation conversion processing and the second stage of binarization processing will be explained, and then an apparatus for carrying out the method will be explained.

First, the first stage N-tone conversion process will be explained.

When converting the density data obtained by reading the 2nd image into binarization, the information contrast in areas with a small signal/noise ratio is lowered by gradation conversion processing. However, the gradation level is changed so that the density becomes higher. Specifically, gradation conversion is performed with linear characteristics up to a certain density value, and for data with a higher density than the above-mentioned density value, the density obtained by gradation conversion performed with the linear characteristics described above. Performs gradation conversion so that the density is higher than that of the original. An example of the density range and its degree in this gradation conversion is shown in FIG. In FIG. 1, Pl has a concentration of 0
．． 8 to 1.1 (PI indicates a certain concentration value above) P2 is concentration 1
．． P3 has a density of 0 to 1.4, the density is 2.0, 03 has a density of 1.5 to 1.7, and DS has a density of 0.1 to 0.6. However, this is because the source of the manuscript is photographic paper, etc., and the contrast is I:! If there is a problem with C, gaps between parts of the manuscript, blue smearing, etc.: It is necessary to match the values of P1 to DS in Section 4 with the IQ draft. For example, in the case of a document where the density of the black part of the document is one minute darker, but the background color is also slightly darker, such as in part 11■, move Pl towards higher density, and make sure that the background color is sufficiently white. In the case of a document in which the density of the portion corresponding to black is low, it is necessary to move [)1 by ζ in the direction of lower density. To do this, we need to prepare several tone conversion characteristics.
It may be possible to make the selection possible depending on the condition of the document, or it may be possible to read the document twice, take a rough histogram at the first time, and automatically determine the conversion characteristics. The histogram in this case is different from the one used when determining a threshold value, and high accuracy is not required, so it is sufficient to read it roughly or to obtain information only about the overall average density.

In this case, Pl, P2) P3.01, and DS are 2.
A table for various average densities may be prepared in advance, and the determination may be made by referring to this table from the obtained average densities.

Next, the binarization process as the second stage will be explained.

81-ZI-Meko/gradation conversion 1-Then, the data obtained by
1'2 Sharpness mask method ``C attempts to emphasize high frequency range,
, hit, 4th picture-((yodoq"-・riCldo-・oro)
Form a 3×3 matrix with b as the center and its neighboring pixels. Find the average density of the (] pixels in the matrix, and perform the calculation l゛=l+k (+-E).
, two; market coefficient). With this operation, I′
is set as the threshold value of the 4-time pixel, and compared with the density -γ-·g I of each 4-time image before high frequency enhancement, floating threshold binarization is performed. When performing this floating vJ threshold 1 straight 2 (1a): After completing the notchless processing, the noise of thin lines and diagonal lines 1- and the roughness in high density areas disappear and the image becomes beautiful. I
V: If applied in the vertical direction, the image reproducibility will be even better. This notchless processing in the horizontal and vertical FF directions; well, when trying to determine white and black, 1.・If the pixel to the left of the n white pixel is a black pixel, the pixel of interest becomes black;] To improve the accuracy, move the threshold by a constant value ↑ ・On the other hand, if the pixel to the left is a white pixel If the white ζ of the pixel of interest is 2, the threshold (1
Move α by a constant value Δt, and then perform the same process by referring to the white and black pixels of the binarized scanning line A and the pixel of interest in the binarized scanning line one line before the pixel of interest. means to do.

When floating threshold binarization is performed using the above-described method, in the upper part where the same level of density continues for a long time, the density data of the first ten before high frequency enhancement will match, and noise will occur in that part. Therefore, the black pixel determining threshold B1-1 and the white pixel determining threshold WL are used, and the floating threshold binarization described above is applied only to pixels having a density corresponding to between the two thresholds. At this time, it would be even more effective to prepare several thresholds for black pixel determination BL and white pixel determination threshold WL, and to make it possible to select an appropriate value depending on the background color, density, and contrast of the original. Increase. Also, as mentioned above, please refer to the table based on the density of the entire screen. Similarly to determining the area and degree of gradation conversion based on the density of the background of the original, a rough histogram is taken, and the threshold BL for determining black pixels and the threshold value BL for determining white pixels are determined. A method of automatically determining the threshold value WL may also be considered.

I was told to perform binarization using the method described below, and the 7-shape mask was removed (・Matrix size in 2 or 3 X: 3-
Even with C, it is possible to obtain a beautiful image in which notches are kept to a minimum and even a single line is reproduced. The reasons are summarized as follows.

The above-mentioned Notsuchi, the roughness of the image of the pond is white%, "ji
This is caused by the small signal/noise ratio of the concentration information near the W portion. When the density data of pixels with high density data adjacent to those pixels is converted to a higher density, the difference in density data between adjacent pixels near the black-white boundary will become larger. When high-frequency emphasis is performed using the 3×3 unsharpness mask method, the effect of high-frequency enhancement becomes more noticeable in areas where the signal/noise ratio is sufficiently large; This is because the effect of high-frequency enhancement between small pixels is small, and notches that are erroneously reproduced by a floating threshold where the same density lasts for a long time are removed by notchless processing. Dimensional: This expression.

The results are shown in Figures 2 and 3. FIG. 2 shows the case where gradation conversion processing is performed, and FIG. 3 shows the case where gradation conversion processing is not performed for comparison. Note that the arrow indicates the direction in which the threshold value is moved by a constant value Δt. Note that as a device, I and I' are compared by setting I-1' = k (, I-E) because ■ "= I + k (I-E) and r are compared, but the idea is that All you need to do is to binarize the high-frequency emphasized image.Also, as explained using a matrix size of 3 x 3,
If you don't mind increasing the cost, use matrix size 5X5-9.
X9 (4 l ine pair/mm in spatial frequency
~81 ine pair/mm). However, in terms of image quality, the matrix size is 3X3.
According to this method, binarization with sufficiently high image quality can be performed using a 3×3 matrix.

Next, a more specific example of the above-mentioned binarization method will be described. As shown in FIG. 4, when a document is read by a reader having the characteristic of outputting 255 as the highest density data, information from 0 to 255 is read for each pixel. Next, for pixels darker than density data 150, the first
After performing the gradation conversion shown in the figure, the matrix size is 3×
3. Emphasize high frequencies using the unsharpness mask method with a weighting factor of 3.0, and use the results to perform floating threshold binarization.
Black pixel determination threshold BL = 55 and white pixel determination threshold WL
=175 and Δt=10, notchless processing is performed in both horizontal and vertical directions. That is, after performing gradation conversion, a 3×3 matrix of ° is formed from neighboring pixels, the density average E of the 9 pixels in the matrix is determined, and I=I+3.0 (
The high frequency region of the pixel of interest is emphasized by the calculation I-E).

get. This ■′ is taken as the threshold value of the pixel of interest at the current moment, and the pixel of interest is binarized as black if I≧I゛, and white if I〈T“.However, at that time, the pixel to the left of the pixel of interest is determined to be black, the aforementioned threshold ia T' is reduced by △t=10 so that the pixel of interest also tends to become black, and conversely, if the pixel to the left is determined to be white, the pixel of interest also becomes white. In order to make it easier for
The white color of the pixel in the same column as the pixel of interest on the scanning line being converted into a value.
The black color is also taken into account and the threshold value is increased or decreased in the same way, and then binarized as described above. At that time, in order to remove noise that occurs in areas where the same level of density continues for a long time, if ■ is less than the nose pixel determination threshold BL (~55), the black and white pixel determination threshold W
If it is equal to or larger than L (~175), white processing is performed. The above operations are performed on all pixels, and the binarized and output results are shown in FIG.

Next, an embodiment of a pixel signal binarization device according to the present invention will be described.

FIG. 6 shows a block diagram of the configuration of the device.

In FIG. 6, an analog signal from a photodetector (not shown) is sampled and held and supplied to an amplifier 14 (signal S). The amplifier 14 performs voltage level shifting and amplitude amplification for A/D conversion. The output of the amplifier 14 is A/D converted by an A/D converter I and output as an 8-bit digital signal. The image average value is determined using the signal b' of the upper 4 bits of the signal. The average value calculator 2 performs this operation. The A/D converted signal is inputted to the gradation converter 3 for gradation conversion. The gradation converter 3 is composed of a RAM, and the stored contents are arbitrarily changed by the main control section 4. It should be noted that, instead of this RAM, it may be configured with several types of ROM and the selection may be changed over. The gradation-converted information f is computed by the mean value computing unit 68 to become the in-matrix mean value (signal j). On the other hand, information r is synchronized with signal J by passing through delay device 7 (signal l), is combined with information V by threshold value calculator 9, and is output as a threshold value. The binarization calculator 11 uses the white level and black level signals n and p and the signals J and k to
A value signal q is generated. Signals n and p are threshold setter 12
The contents of the threshold value setter 12 are set by the main controller 4.
It is set arbitrarily by The delay device 13 is 1 b for one line.
i t, the delay circuit sends the binarized information immediately before the pixel of interest (signal l) and the binarized information (in the same column in the row immediately before the pixel of interest) to the threshold variation width calculator 10 in the delay circuit. output signal m). The image quality setting device 5 is for pixel setting, and depending on the document, the operator can set the document density to "dark" or "normal".
Set to either "Light" or "Auto". The main control section 4 controls the apparatus, ie, controls the lighting of the lights, the speed of operation, etc. using the signal group t, and also controls and sets each section according to the output of the image quality setting device 5. That is, when "auto" is instructed, the document is scanned at a faster speed than usual, and the average density value of the document is calculated by the average value calculator 2.
From that value, refer to a predetermined table and set the predetermined gradation to the gradation converter 3, Δt to the threshold variation width calculator 1o, and the black pixel determination threshold BL and white pixel determination threshold to the threshold value setter 12. WL
are set in the same way, and normal scanning is performed again this time to obtain binarized information r. Also, when either "Dark", "Normal", or "Light" is selected, the tone converter 3, threshold variation width calculator 10, and FJ III Set the setting device 12 to perform main scanning, and
Get value information.

FIG. 7 shows the configuration of the binarization calculator 11.

Comparators 15, 16, and 17 are digital comparators that receive 1-bit signals U, ■, and W, respectively.
Output. This output is: If j>k, then 11-White J≦k, then U-Black j>n (WL), If l = White j<n (WI-), then (f u-Black (1 sushi W L) > F3 L ) If j > p (BL), then 1" = white If J≦p (BL), then U - black. The binarization calculator 18 is (i'3 1"),
The final output q is output from the results of (1) and W, and the logic is as follows.

White Black White White 1 White of Black Black White Black (However, the x mark can be either black or white) FIG. 8 shows the configuration of the threshold value variation range calculator 1o.

Four types of Δt are selected according to the states of signals l and m using two's complement. Threshold fluctuation range register 19.20.21,
and 22 are control signal group g supplied from the main control unit 4
(go, gl, g2)g3), and output signals vO1v1, v2) and v3, respectively, according to Δt. The fluctuation range selector 23 selects vO~ by the signals l and m.
This is a selector that selects v3, and operates as follows.

For example, vO to ■3 are set as follows, and are set to predetermined values by the main control section 4 according to the setting state of the image quality setting device 5.

vo=-20 vl=-IQ v2=-10 v3 =+20 FIG. 9 shows the configuration of the threshold 11α calculator 9, which is composed of a digital two's complement adder 24. That is, k is output as addition information of the output V of the threshold variation width calculation unit 1o and the signal 1 obtained by delaying the image information f.

The main scanning average calculation unit 6 is a calculation unit that calculates an average value in the main scanning direction, and has the configuration shown in FIG. 10. The delay device 25 and the delay device 26 are delay devices for one pixel in the main scanning direction. The adder 27 adds these outputs fl, f2) and f and outputs the result as a signal. Through this operation, the sum of two pixels in the horizontal direction is obtained.

FIG. 11 shows the configuration of the sub-scanning average calculator 8 that calculates the average value in the sub-scanning direction. The delay device 28 and the delay device 29 output hl and hl as delay devices in the - row. The adder 3o adds h, bl, and h3 and outputs the result as an output J'. The average value output ROM 31 is a table that divides the calculated sum of the nine pixels by 1/9, and is composed of a ROM. Through the operations of the main scanning average calculator 6 and the sub-scan average calculator 8, it is possible to calculate the average value of the pixel of interest and nine surrounding pixels at the same speed as the main scanning speed.

Although the present invention has been described above with reference to embodiments, the following modifications are possible according to the technical idea of the present invention.

For example, information after gradation conversion can be spatially filtered,
Any configuration is possible as long as it can be compared with the information before filtering.

In this example, 2M=I+k (r-E) is used, but the formula is not limited to this, and the formula I'=I+k(1-E) is used.
It goes without saying that it may be modified in any way as long as it is equivalent in principle. In addition, in the explanation of the device, Equation 1
Although the coefficient k of '=r+k(1-E) is not mentioned, in the case of the device according to the invention, an increase or decrease in the coefficient is equivalent to an increase or decrease in Δt.

In this embodiment, image information after A/D conversion is directly used, but shading correction, which is commonly performed, may be performed.

In this embodiment, tone conversion is performed digitally, but equivalent performance can be obtained even if this is performed using an analog amplifier in front of the amplifier 14.

(6) Effects of the Invention As explained above, the present invention performs gradation conversion on the density data when performing binarization processing on the density data obtained by reading the original, and then Since the binarization is performed in a matrix size of 3×3, it is possible to sufficiently compare the effect of high-frequency emphasis, minimize notches, and reproduce thin lines.

[Brief explanation of drawings]

FIG. 1 is a conversion characteristic diagram showing an embodiment of the high-frequency emphasis filter of the image information binarization method according to the present invention, and FIGS.
The figure is a characteristic diagram explaining the effect of gradation conversion of the image information vii binarization method according to the present invention, Figure 4 is a characteristic diagram showing the characteristics of the reader used in the embodiment of the present invention, and Figure 5 is a characteristic diagram according to the present invention. An output characteristic diagram showing the effect of gradation conversion of the image information binarization method, FIG. 6 is a block diagram showing an embodiment of the image information binarization device according to the present invention, and FIG. FIG. 8 is a block diagram showing the configuration of the value calculation unit 11, FIG. 8 is a block diagram showing the configuration of the threshold variation width calculation unit 1O shown in FIG. 6, and FIG. 10 is a block diagram showing the configuration of the main scanning average calculator 6 shown in FIG. 6, FIG. 11 is a block diagram showing the configuration of the sub-scanning average calculator 8 shown in FIG. The figure is a characteristic diagram showing the characteristics of a high-frequency emphasis filter. 3... Gradation converter 9... Threshold value calculator 10... Threshold variation width calculator 11... Binarization calculator 12... Threshold value setter 18... Binarization calculator 19 .20...Threshold fluctuation range register 21.22...
・Threshold value variation range register 23...Variation range selector 68...Average value calculator Patent applicant: Roku Konishi Photo Industry Co., Ltd. Figure 1 Figure 3 Figure 2 Figure 4 Density Figure 5 (al Hei 2 History study A history appetizer
(b) Figure 8 Figure 9

Claims (18)

[Claims] (1) 2 for the density data obtained by scanning the original
An image signal binarization method characterized in that, when performing the digitization process, tone conversion is performed on the density data, and then binarization is performed. (2) An image signal binarization method as set forth in claim 1, characterized in that tone conversion is performed on density data having a higher density than a predetermined density value so that the density becomes higher. . (3) In claims 1 and 2, a pixel of interest and its neighboring pixels form a matrix, the density average of the pixels in the matrix is determined, and the average value and data of the pixel of interest are calculated. An image signal binarization method characterized in that a value obtained by performing an operation between the two is used as a threshold, and the threshold is compared with data before the operation to perform binarization. (4) An image signal binarization method as set forth in claims 1 to 3, characterized in that processing is performed to remove unnecessary irregularities appearing at black-white boundaries during binarization. . (5) Claims 1 to 3 are characterized in that processing for removing unnecessary irregularities that appear at black-white boundaries during binarization is performed in both horizontal and vertical directions. Image signal binarization method. (6) An image signal binary as set forth in claims 1 to 5, characterized in that a threshold for determining a black pixel and a threshold for determining a white pixel are provided as thresholds for binarization. method. (7) The image signal binarization method as set forth in claim 6, characterized in that the threshold value for determining black pixels and the threshold value for determining white pixels can be selected depending on the background color and contrast of the document. (8) An image signal binarization method according to claims 3 to 7, characterized in that processing is performed with a matrix size of 3×3. (9) An image signal binarization method according to any one of claims 1 to 8, characterized in that the density data is obtained by a semiconductor photodetector. (10) An image signal characterized in that when performing binarization processing on density data obtained by reading a document, gradation conversion is performed on the density data and then binarization is performed. Binarization device. (11) In the description of claim 10, the image signal 2 is characterized in that density data having a density higher than a predetermined density value is subjected to gradation conversion so as to have a higher density.
Value device. (12) In claims 10 to 11, a pixel of interest and its neighboring pixels form a matrix, the density average of the pixels in the matrix is determined, and the average value and data of the pixel of interest are calculated. An image signal binarization device characterized in that a value obtained by performing an arithmetic operation between the two is used as a threshold value, and binarization is performed by comparing this threshold value with data before the arithmetic operation. (13) An image signal binarization device as set forth in Claims 10 to 12, characterized in that the image signal binarization device performs processing to remove unnecessary unevenness that appears at the black-and-white boundary during binarization. . (14) Claims 10 to 12 are characterized in that processing for removing unnecessary irregularities that appear at black-white boundaries during binarization is performed in both horizontal and vertical directions. Image signal binarization device. (15) An image signal binary as set forth in claims 10 to 14, characterized in that a threshold for determining a black pixel and a threshold for determining a white pixel are provided as thresholds for binarization. conversion device. (16) The image signal binarization device as set forth in claim 15, wherein the threshold for determining black pixels and the threshold for determining white pixels can be selected depending on the background color and contrast of the original. (17) An image signal binarization device according to claims 12 to 16, characterized in that processing is performed with a matrix size of 3×3. (18) An image signal binarization device according to any one of claims 10 to 17, characterized in that the density data is obtained by a semiconductor photodetector.