JP2543198B2 - Method for converting linear density of binary image signal - Google Patents

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention can be used for image communication such as a facsimile and various file systems for image data, and for smoothly reproducing an image received or read from a storage device. The present invention relates to a linear density conversion system for binary image signals.

Conventional technology In the example of G3 standard for facsimile communication, the resolution in the main scanning direction is fixed at 8.037 pixels / mm, and the resolution in the sub scanning direction is
Three types are defined: 3.85 pixels / mm, 7.7 pixels / mm, 15.4 pixels / mm, and can be selected according to the image content to be transmitted.

On the other hand, recently received data is recorded in the same way as laser printers.
Devices capable of high-density recording of 16 pixels / mm or more have come into wide use. On the transmitting side, low-density scanning is advantageous in terms of the amount of information to be transmitted and the communication cost / time, but on the recording side, there is a desire to obtain smoother reproduced images by utilizing the characteristics of high-density recording. There is a demand for an optimum linear density conversion method for converting a density image into a high density image. This is the same not only in facsimile communication but also in a general image data file system, and is a technique necessary for reproducing a smooth image from an image whose resolution has been reduced in order to reduce the file capacity.

By the way, SP is the conventional linear density conversion method for binary images.
Various methods such as C method, logical sum method, projection method, and 9 division method have been proposed. (Morimachi, Komachi, Yasuda: "High speed arbitrary pixel density conversion method for copying with editing function" 12th image Proceedings of Engineering Conference, (S56 December), Arai, Yasuda: "A Study on Facsimile Line Density Conversion", The Institute of Image Electronics Engineers of Japan, Vol. Method ”IPSJ 20th National Convention, P.
463) Among them, the projection method is known as a method with little deterioration in image quality. FIG. 13 is a diagram for explaining the projection method, and D
Represents an output pixel point, and P, Q, R, and S represent input pixel points adjacent to the point D. P ', Q', R ', S'are input pixel points P, Q, R, S
Of the output pixel point D, I _P , I _Q , I
_R, I _S input pixel point P, Q, R, when the data value of S {= 0 (white) or 1 (black)}, the weighted average value f a (D) for the output pixel points D by: calculate.

The following threshold processing is performed on f (D) to determine the data value I _D of the output pixel point D.

Similar to the projection method in any of the other methods, the value of the point D is determined based on the positional relationship between the values of the points P, Q, R, S of the input image and the point D of the output image.

However, a significant image quality improvement effect cannot be expected by the method of determining the data value of the output image only by the data values of the four points of the input image as in the above-mentioned conventional technique. The degree of image quality improvement when high-density conversion of scanning lines is performed can be evaluated based on whether the line segments are smoothly interpolated with an appropriate line width.

FIG. 14 shows an example in which the main scanning and the sub scanning are doubled. (A) shows image data obtained by binarizing an image of a black line indicated by diagonal lines, (b) shows an example of interpolating the image data by a conventional method, and (c) shows an example of ideal interpolation.

The problem with the conventional method is that since the data values to be supplemented are determined only by the positional relationship with the values of the four neighboring pixels, the smoothness of the line segment is effective only in the vicinity of the 45 ° direction. As in the formula (2), since the “=” in the binarization process is set to 1, the line segment becomes thicker (on the contrary, when “=” is set to 0, the line segment becomes thinner.

In view of the above problems, the present invention provides a method for converting the linear density of a binary image signal that can obtain a high quality image.

Means for Solving the Problems In order to achieve the above-mentioned object, the present invention firstly, when converting a low-density binary image of a facsimile or the like into a high-density binary image by pixel interpolation, a character excluding a halftone image is used. The correspondence of the pattern structure between the binary pattern A forming the partial area of the low density image such as a line drawing and the binary pattern B forming the same portion of the desired high density image is calculated in advance by the high density image for learning. When the low density image is not a halftone dot image, it is learned statistically, and an appropriate reference pixel selected from within the partial area of the low density image is used, and the low density binary pattern is obtained based on the learning result. A is converted into the high density binary pattern B, and where the low density image is a halftone dot image, it is converted into a high density image by repeating the same pixel of the low density image.

Secondly, according to the present invention, when converting a low-density binary image such as a facsimile into a high-density binary image by pixel interpolation, a partial area of the low-density image such as a character or line drawing excluding a halftone image is formed. Of the binary pattern A1 and the binary pattern B1 forming the same portion of the desired high-density image, and the desired binary pattern A2 forming the partial area of the low-density image of the halftone image. Of the binary pattern B2 forming the same portion of the high-density image is statistically learned in advance by the high-density learning image, and the low-density image is not the halftone image. Based on the learning result, the low-density binary pattern A1 is generated by using an appropriate reference pixel selected from the partial area of the low-density image.
Is converted into the high-density binary pattern B2, and where the low-density image is a halftone dot image, an appropriate reference pixel selected from within the partial area of the low-density image is used, and based on the learning result, The low density binary pattern A2 is converted into the high density binary pattern B2.

Operation The present invention uses the above method to learn in advance the properties (thickness and smoothness of line segments) of the target pixels corresponding to the linear density of the conversion output, create a dictionary, and input image data based on the dictionary. Since data interpolation is performed, a good converted output image is obtained, which is irrelevant to the characteristics of the target image as seen in the conventional method, and which is better than the one that determines the data interpolation value according to the defined method. Further, it is possible to perform an interpolation process suitable for each type of image having different properties such as a halftone image and a character / line drawing, and it is possible to obtain a higher quality output image.

First Embodiment FIG. 1 is a diagram showing a processing process of a binary image signal linear density conversion method according to a first embodiment of the present invention. Reference numeral 1 denotes learning image data, which has the same resolution as the scanning line density when the input image data is interpolated and recorded, and includes various input image contents such as characters and line drawings. In addition, the scanning line density of the learning image data is set to the main scanning m pixel / mm,
Sub-scanning n pixels / mm. Reference numeral 2 is a process of learning the interpolation data, which will be described in detail later, but the learning is performed separately for each type of interpolation magnification for the input image data. Now, learning for interpolating the main scanning by K times and the sub scanning by L times with respect to the input image data is referred to as K × L times interpolation learning. 3 is a process of creating an interpolation data dictionary, which creates a K × L times interpolation dictionary for a character / line drawing that determines 1/0 of an interpolation value based on the result of the K × L times interpolation learning. Reference numeral 4 is input image data, and the scanning line density is m / K pixels / mm for main scanning and n / L pixels / mm for sub-scanning.
Reference numeral 5 denotes a halftone dot image detection process, which determines whether or not the input image data is a halftone dot image. Reference numeral 6 denotes an input data interpolation process, which uses all the input image data, if the judgment is a halftone dot image, predetermined pixel data of the input image data, and if the judgment is not a halftone dot image, the character / line drawing. Main scan using K × L times interpolation dictionary for main scan
The sub-scanning is doubled to L times. 7 is interpolation image data,
Main scanning m pixels / m interpolated by the input data interpolation 6
Image data of m and sub-scanning n pixels / mm.

FIG. 2 is a diagram showing a processing process of a linear density conversion method for a binary image signal in the second embodiment of the present invention. First
Blocks with the same numbers as in the figure show processes of the same operation.
Therefore, only the process different from the embodiment of FIG. 1 will be described below. Reference numeral 8 is learning image data, which is shown in FIG.
This is a halftone dot image of the same format as the learning image data described in. Reference numeral 9 is a process of creating a K × L times interpolation data dictionary for halftone images having the same format as the interpolation data dictionary described in FIG. Reference numeral 10 denotes an input data interpolation process. If the determination in the halftone image detection process 5 is a halftone dot image, the K × L times interpolation dictionary for halftone dot images is used. The main scan is interpolated by K times and the sub-scan is interpolated by L times using the line drawing K × L times interpolation dictionary.

In the image processing process of FIGS. 1 and 2, the process of creating the interpolation data dictionary in the processing blocks 1 to 3, 8 and 9 may be a software process for performing a time-consuming operation using a computer. However, since the interpolation processing of the input data in the processing blocks 4 to 7 and 10 is usually required to be speeded up, hardware processing is appropriate.

Therefore, the embodiment of the present invention will also be described in detail with the above configuration.

It is assumed that the K × L multiplication dictionary used in the input data interpolations 6 and 10 of the processing block is provided as data stored in a memory such as a ROM or a RAM.

Details of the processing blocks 2, 3, 6, 9 and 10 are 2 × with K = 2 and L = 2.
Two examples of 2 × interpolation and 2 × 4 × interpolation with K = 2 and L = 4 will be described. The present embodiment is particularly applicable to G3 standard input image data in facsimile communication.

(1) Detailed Example of Interpolation Data Learning FIG. 3 shows a scanning window for 2 × 2 times interpolation learning.

r _{1 to} r ₁₆ are reference pixels, and h _{1 to} h ₃ are learning pixels.

Since there are 16 pixels to refer to, 2 ¹⁶ types of patterns are created by the reference pixels. Learning is to statistically examine whether h _{1 to} h ₃ are 1 or 0 in each pattern when the learning image data is scanned through the scanning window. The learning result improves as the number of reference pixels increases, but the interpolation dictionary grows in proportion to the square of the number of pixels. Therefore, it is limited in relation to the memory capacity used for the dictionary. When learning by software in the computer, the reference pixels r ₁ ~r ₁₆ as an address of the memory space, providing a three-up-down counter of h ₁ to h ₃ for each address, the learned pixel is 1 If it is 0, the operation is +1; if it is 0, the operation is -1. (The reason why the up-down counter is used is because it is premised that the interpolated value is 1 if the counter value after learning is positive and 0 if it is negative, and the interpolated value of 1 or 0 is intentionally set. If you want to perform many operations, you should prepare counters for 1 and 0, and determine the interpolated value by the ratio of the respective counter results.) The program configuration for the above software operation is general and easy. Since this is a technology, details are omitted. FIG. 4 shows a scanning window for 2 × 4 times interpolation learning.
The interference pixel spacing in the sub-scanning direction compared to the 2 × 2 × interpolation learning scan window of FIG. 3 is doubled, the 2 × 2 times, except that learning pixels are increased to seven h ₁ to h ₇ This is the same as the interpolation learning.

(2) Detailed Example of Creating Interpolation Dictionary FIG. 5 is a specific example of creating an interpolation dictionary. ROM or RAM
In this memory space, r _{1 to} r ₁₆ are used as data addresses, and each interpolation data h ′ ₁ , h ′ ₂ ,. h _'1, h' ₂ ...... 1 or 0
Is set as follows for each value of the learned counter h _i (i = 1, 2, ...).

h ′ ₁ = r ₆ r ₆ is a reference pixel value near the interpolation pixel position as shown in FIGS. 3 and 4.

(3) Detailed Embodiment of Input Data Interpolation FIG. 6 is a hardware configuration example of the 2 × 2 times interpolation in the first embodiment of FIG. 1, and FIGS. It is a timing chart of an output signal. Explaining the timing chart of FIG. 7 earlier, FIG. 7A shows a reset signal, which is a pulse generated in each scanning section (main scanning section) of the output image signal, and FIG. As shown in the figure, the data of one scanning line of the input image signal is input by the input image signal every other scanning period. FIG. 3C shows clock 1, which corresponds to the pixel clock of the output image signal {double the frequency of the input image signal pixel clock (not shown)}. FIG. 3D is a clock 2 and is a timing pulse for inserting interpolation data between input image data. The clock 2 has a frequency half that of the clock 1 and is generated every other scanning period as in the case of the input image signal.
FIG. 7E shows an output image signal. In the same figure (f), the write address pointer of the image memory 12 which will be described later is controlled by the write inhibit pulse.

Next, the hardware configuration of FIG. 6 will be described with reference to each signal of this timing chart. 11 is a reset signal input terminal, 12
Is an input terminal of an input image signal, 13 is an input terminal of a clock 1, 14 is an input terminal of a clock 2, 15 is an output terminal of an output image signal, 16 is an input terminal of a write inhibit pulse, 17 is a dot area detection signal. It is an input terminal. 18 is an 8-bit FIFO (First In Fir) marketed as a dedicated image memory.
line memory having a st out structure (for example, μPDN 41101C manufactured by NEC Corporation; NEC, IC memory data book 19
87 years pp.293〜). D _{IN0 to} D _IN5 are data input terminals, D _{OUT0 to} D _OUT5 are data output terminals, and ▲ ▼ are reset input terminals for initializing the internal read address pointer by reset read input (data address = 0), ▲ ▼ is a reset input terminal for reset write input to initialize the internal write address pointer (data address = 0),
PCK is a clock input terminal for incrementing the read address pointer for each clock input with a read clock input, WCK is a clock input terminal for incrementing the ride address pointer for each clock input with a write clock input, and ▲ ▼ is a read When the enable input is at "H" level, the read address pointer is RC
Read operation control input terminal to stop at the current position without incrementing even if a clock is input to K, ▲
▼ is a write operation control input terminal for stopping the write address pointer at the present position without incrementing even when a clock is input to WCK when the write enable input is at “H” level. ▲ ▼ and ▲
▼ indicates input terminal 11, RCK and WCK indicate input terminal 13, ▲
▼ is connected to the input terminal 16, and ▲ ▼ is grounded (“L”). 19 has seven 7-bit shift registers (20 to 26)
Individual data sets are also possible with the registers that it has. As shown in the figure, each register of r _{1 to} r ₁₆ and h _{1 to} h ₃ is a third register.
It is arranged corresponding to the figure. FCK the registers common shift clock pulse input terminal, SCK is a clock input terminal for setting the data of h ₁ to h _3, input terminals 1
3, connected to the input terminal 14. The final bit output of each 7-bit shift register 20-25 is the image memory 18 respectively.
Connected to each input terminal D _{IN0 to} D _IN5 of the first bit input of 21 to 26 is each output terminal of the image memory 18 D _{OUT0 to} D _OUT5
The first 20 bit inputs are connected to the input terminal 12. The 24 register bits r ₆ are connected to the output terminal 16 as output image signals. 27 is an interpolation data memory, which connects each bit r _{1 to} r ₁₆ of the register group 19 to the address line of the memory and outputs data h ′ _{1 to} h ′ ₃ to the data line of the memory. The contents of the interpolation data memory 27 have the structure shown in FIG. Reference numeral 28 denotes a data selector which outputs three pieces of data identical to the register bit r _{5 of} 24 when the halftone dot area detection signal of the input terminal 17 is ON (= 1), and the data h ′ ₁ when OFF (= 0). ~ H ' ₃ is output and each register group 19
Connect to the data set input pin of each bit h _{1 to} h ₃ of.

The operation will be described with the above configuration. The read address pointer and the eye address pointer of the image memory 18 are set to 0 at the beginning of one scanning section by the reset signal of the input terminal 11, and then the write address pointer is incremented by the clock 1 of the input terminal 13. The read address pointer counts 7 by the write disable pulse of input terminal 16.
It increments with a delay. This is a group of registers
Each of the 19 input signals corresponds to the delay of being shifted to the output end. The input image signal of the input terminal 12 is stored from D _{IN0 of} the image memory via the shift register 20 by the clock 1 of the input terminal 13 in the first scanning section, and is output from D _{OUT0 in} the second scanning section. The data is stored from D _{IN1 of} the image memory via the shift register 21, and then flows from D _OUT1 → register 22 → D _IN2 → D _OUT2 → register 23 → D _IN3 → ... The input image signal is subjected to data interpolation of h ₂ and h _{3 in} the register 23 and h _{1 in} the register 24 by the clock 2 of the input terminal 14.

In the case of 2 × 4 times interpolation, the configuration of the register group 19 corresponds to that shown in FIG. 4, and accordingly, the capacity of the image memory 18 is 6
The number of lines is increased from 12 to 12 and the input image signal and the signal of the clock 2 shown in FIG. 7 are every 3 scanning intervals, which is different from the 2 × 2 interpolation, and the details will be omitted. Further, as another configuration example that achieves a similar interpolation purpose other than the configuration of FIG. 6, the scanning window of the register group 19 is set to reference pixels r _{1 to} r _16.
It is easily conceivable to set only (thus, 4 × 4 scanning window) and insert the interpolation data in a separate system, but the description is omitted.

FIG. 8 is a hardware configuration example of 2 × 2 interpolation in the second embodiment of FIG. The difference from the configuration of FIG. 6 is that an interpolated data memory 29 for halftone dot images is prepared, and when the data selector 30 turns on the halftone dot area detection signal of the input terminal 17, the interpolated data for halftone dots in the interpolated data memory 29. Value h ″ _{1 to} h ″ ₃
When it is OFF, the character / line drawing interpolation data h ′ _{1 to} h ′ ₃ in the interpolation data memory 27 are output and used as the data set values of the bits h _{1 to} h ₃ of the register group 19, respectively.

(4) Detailed embodiment of halftone dot image detection Learning method A dictionary memory for halftone dot image detection is created with a halftone dot image for learning having the same scanning line density as the input image. FIG. 9 shows an embodiment in which the image data area 31 is scanned by a 4 × 4 scanning window.
The dictionary memory in the case of the present embodiment can be created by setting the data value of the address to 1 as the data value {x ₁ , x ₂ , ... X ₁₆ } of the scanning window 32.
(However, all the initial values of the memory are cleared to 0.) To detect halftone dots of the input image data, refer to the data value of the dictionary memory with the data value in the scanning window for scanning the input image data as an address, and If so, it is determined to be a halftone dot area.

The judgment result of the dot area detection is statistically 1 in the dot area.
However, the output is discrete such that 1 is small (there are many 0s) in the character / line drawing area, and the area separation determination result is incomplete. (If the scanning window size is increased and the number of reference pixels is increased, the determination accuracy can be improved and the discreteness can be reduced, but the hardware configuration is correspondingly heavy. Therefore, the determination result is two-dimensional. It is preferable that the final halftone dot area separation determination result be obtained by performing smoothing processing.

FIG. 10 shows an example of the smoothing process. Reference numeral 33 is a first scanning window for scanning an input image signal, 34 is the halftone image detecting dictionary memory, and a halftone dot area detection signal based on the learning result using the data value in the first scanning window 33 as an address. Is output. Reference numeral 35 is a second scanning window for scanning the halftone dot area detection signal, and 36 is an area separation dictionary memory for outputting the final halftone dot area detection signal using the data in the second scan window 35 as an address. The area separation dictionary memory 36 stores 1 of the address data value.
Is checked, and if (the number of 1s of the address data value) ≧ N, the data value of the address in the dictionary memory 36 is set to 1
And the others can be created by setting it to 0. N is a constant, and when it is set to a small value, the judgment result is shifted to the halftone dot image area, and when it is set to a large value, the judgment result is shifted to the character / line drawing area.

Pattern structure determination method of data in the scanning window Without having a learning dictionary of halftone image data like the learning method described above, the pattern structure itself is examined in the scanning window,
A method of sequentially determining whether or not the halftone image has the characteristics will be described.

(A) Main scanning direction and sub scanning direction 1 in the scanning window 32 of FIG.
If the number of alternations of 0 is larger than a certain value, it is determined as a halftone image. First
FIG. 1 shows a configuration example of the present embodiment, and 38 is a 1/0 alternation number determination dictionary. It is obvious that the dictionary 38 can be created in advance from the data structure of the scanning window 37 like the area separation dictionary 36 of FIG.

(B) If there is an isolated point of 1 or 0 in the data in the scanning window 32 of FIG. 9, it is judged as a halftone image. The isolated point can be determined by the logical expression of the data within the scanning window, but it is also possible to use the 1/0 isolated point determination dictionary 39 as shown in FIG.

Using the features of the above two methods (A) and (B), (Dictionary
It is also possible to judge the halftone image with (38) + (dictionary 39). Also in the present pattern structure determination method, it is desirable to perform the two-dimensional smoothing process on the determination result as in the above-described learning method, and then set the final halftone dot area separation determination result.

EFFECTS OF THE INVENTION As described above, according to the present invention, a dictionary in which the features of the image corresponding to the linear density of the converted output are stored is prepared in advance,
Since the input data is interpolated according to the dictionary, the amount of learned information is large compared to other interpolation methods without learning, so line segments are interpolated more smoothly and with an appropriate line width, and high-quality image output is obtained. That effect is great.

[Brief description of drawings]

FIG. 1 is a diagram showing a processing process of a linear density conversion method for a binary image signal in the first embodiment of the present invention, and FIG. 2 is a linear density conversion for a binary image signal in the second embodiment of the present invention. FIG. 3 is a diagram showing a processing process of the method, FIG. 3 is a conceptual diagram of a scanning window for 2 × 2 interpolation learning which is a main part of the embodiment, and FIG.
5 is a conceptual diagram of the scanning unit for 4 × interpolation learning, FIG. 5 is a conceptual diagram of the same interpolation dictionary, and FIG. 6 is 2 × in the first embodiment of FIG.
FIG. 7 shows a block diagram of a main block for realizing double interpolation.
FIG. 8 is a timing chart of main parts, FIG. 8 is a block connection diagram of main parts for realizing the 2 × 2 times interpolation in the second embodiment of FIG. 2, and FIG. 9 is a 4 × 4 scanning window of the same image data area. Fig. 10 is a conceptual diagram of scanning, Fig. 10 is a block diagram of a main part that realizes the smoothing process, and Fig. 11 is a block diagram of a main part that realizes a halftone dot image determination by an alternating number of 1/0. Fig. 12 is a block diagram of a main part for realizing the halftone dot image determination by the same isolated point detection, Fig. 13 is a conceptual diagram showing a conventional projection method, and Fig. 14 is a double interpolation for both main scanning and sub scanning. It is a conceptual diagram. 1 ... Learning image data (characters / line drawings, etc.), 2 ... Interpolation data learning, 3 ... Interpolation data dictionary (characters / line drawings, etc.),
4 ... Input image data, 5 ... Halftone dot image detection, 6, 10 ...
Input data interpolation, 7 ... Interpolation image data, 8 ... Learning image data (dot image), 9 ... Interpolation data dictionary (waist image), 34 ... Halftone image detection dictionary memory, 36 ... Area separation dictionary memory, 38 …… 1/0 Alternation number judgment dictionary, 39 …… 1/0
Isolated point judgment dictionary.

Front page continuation (72) Inventor Toshiharu Kurosawa 3-10-1, Higashisanda, Tama-ku, Kawasaki, Kanagawa Prefecture Matsushita Giken Co., Ltd. (72) Hidehiko Kawakami 3--10-1, Higashisanda, Tama-ku, Kawasaki, Kanagawa Issue Matsushita Giken Co., Ltd. (72) Inventor Hiroaki Kodera 3-10-1 Higashisanda, Tama-ku, Kawasaki City, Kanagawa Prefecture Matsushita Giken Co., Ltd. (72) Inventor Hideaki Ohira 2-3 3-Shimomeguro, Meguro-ku, Tokyo No. 8 in Matsushita Electric Transport Co., Ltd. (72) Inventor Mutsuo Hashizume 2-3-8 Shimomeguro, Meguro-ku, Tokyo Matsushita Electric Power Co., Ltd. (56) Reference JP-A-2-268073 (JP, A)

Claims (16)

(57) [Claims] 1. When converting a low-density binary image such as a facsimile into a high-density binary image by pixel interpolation, a partial area of the low-density image such as a character or line drawing excluding a halftone image is formed. The correspondence relationship of the pattern structure between the value pattern A and the binary pattern B forming the same portion of the desired high density image is statistically learned in advance by the high density image for learning, and the low density image is a halftone dot. Where the image is not an image, the low density binary pattern A is converted into the high density binary pattern B based on the learning result by using an appropriate reference pixel selected from within the partial area of the low density image, and the low density binary pattern B is converted into the low density binary pattern B. Where the density image is a halftone dot image, the same pixel of the low density image is repeated to convert it into a high density image. 2. A linear density conversion method for a binary image signal, which comprises manually instructing whether or not the low-density image according to claim 1 is a halftone dot image. 3. The determination as to whether or not the low density image according to claim 1 is a halftone dot image is performed based on the result of statistical learning by a low density image for learning in advance.
Value image signal linear density conversion method. 4. A linear density conversion method for a binary image signal, wherein the determination result according to claim 3 is smoothed on an image plane and then used as a determination result as to whether the image is a halftone dot image. 5. A binary image characterized in that the determination as to whether or not the low density image according to claim 1 is a halftone image is made by a pattern structure determination within a scanning window for scanning the low density image. Signal linear density conversion method. 6. A linear density conversion method for a binary image signal, wherein the pattern structure determination according to claim 5 is performed by an alternating number of 1 and 0 within a scanning window. 7. A method for converting linear density of a binary image signal, wherein the pattern structure determination according to claim 5 determines that the image is a halftone image when there are 1 or 0 isolated points in a scanning window. 8. A linear density conversion method for a binary image signal, which comprises smoothing the determination result according to claim 5 on an image plane and then determining whether the image is a halftone image. 9. When converting a low-density binary image of a facsimile or the like into a high-density binary image by pixel interpolation, a partial area of the low-density image such as a character or line drawing excluding a halftone image is formed. The correspondence of the pattern structure between the value pattern A1 and the binary pattern B1 forming the same portion of the desired high-density image, and the binary pattern A2 forming the partial area of the low-density image of the halftone image and the desired high-density image. The correspondence of the pattern structure with the binary pattern B2 forming the same portion of the density image is statistically learned in advance by the learning high-density image, and the low-density image is a place where the low-density image is not a halftone image. Using an appropriate reference pixel selected from within the subregion of
The low-density binary pattern A1 is converted into the high-density binary pattern B2 based on the learning result, and the low-density image is a halftone image which is selected from a partial area of the low-density image. The low density binary pattern A2 is converted to the high density binary pattern based on the learning result using a reference pixel.
A method of converting a linear density of a binary image signal, characterized by converting to B2. 10. A linear density conversion method for a binary image signal, which comprises manually instructing whether or not the low density image according to claim 9 is a halftone dot image. 11. The determination as to whether or not the low density image according to claim 9 is a halftone dot image is performed based on a result of statistical learning previously performed by a low density image for learning. Value image signal linear density conversion method. 12. A linear density conversion method for a binary image signal, which comprises smoothing the determination result according to claim 11 on an image plane and then determining the result as a halftone image. 13. A binary image, characterized in that the determination as to whether or not the low density image according to claim 9 is a halftone dot image is made by a pattern structure determination within a scanning window for scanning the low density image. Signal linear density conversion method. 14. A linear density conversion method for a binary image signal, wherein the pattern structure determination according to claim 13 is performed by an alternating number of 1 and 0 within a scanning window. 15. The method for linear density conversion of a binary image signal according to claim 13, wherein when the scan window has an isolated point of 1 or 0, it is determined as a halftone image. 16. A method for converting linear density of a binary image signal, wherein the determination result according to claim 13 is smoothed on an image plane and then used as a determination result as to whether the image is a halftone dot image.