JPH0434346B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an image processing device that outputs a halftone image by increasing resolution and gradation.

In general, scanning systems using rotating polygon mirrors or vibrating mirrors are widely used in laser-based facsimile machines, various display devices, recording devices, etc. because they allow a large scanning angle and have little chromatic dispersion. In particular, rotating polygon mirrors are widely used as high-speed scanning devices.

Conventionally, in such a scanning system, as a method for recording or displaying a halftone image, one pixel on the recording/display surface is constructed with a matrix structure consisting of n×n (n is a positive integer) fine pixels, and each pixel is There is a method that attempts to reproduce halftones by filling in micropixels made up of matrix elements, that is, by distributing white micropixels and black micropixels. FIGS. 1A to 1E show examples of pixels in a matrix configuration in such a recording method. Here, pixel 11 is 2×
It consists of 2 micropixels 13. With this pixel configuration,
When trying to reproduce halftones, by sequentially filling in minute pixels as shown in Figure 1 A to E, the density increases in sequence from 0, 1, 2, 3, and 45.
You can get gradations. In general, by configuring pixels as an n×n fine pixel matrix, n×n+1
A gradation image is obtained. The advantage of this halftone image method is that one fine pixel can be recorded in two colors: white or black.
The system is simple because it only needs to be recorded as a value, and the gamma of the output density with respect to the amount of light incident on the photoreceptor may be non-linear, so the type of photoreceptor does not matter. On the other hand, according to this method, one pixel is composed of multiple fine pixels, so each
The drawback is that the pixel size increases and the resolution of the image deteriorates. In order to eliminate such drawbacks, it is necessary to reduce the size of the pixels. Therefore, it is necessary to make the size of the micropixel extremely small. For example, when applied to a laser beam printer, the diameter of the beam spot on the photosensitive drum is required to be extremely small. It is possible to fully satisfy such requirements in an optical system. However, due to the characteristics of the photoreceptor, it may not be possible to obtain a sufficient response to fine pixels whose size has become smaller.

FIG. 2 shows an example of the characteristics of a photoreceptor, where the horizontal axis represents the spatial frequency and the vertical axis represents the MTF value. This characteristic applies when the photoreceptor is a photoreceptor drum for electrophotography. 21 is the latent image characteristic, 2
3 shows the development and transfer characteristics, respectively. As can be seen from these characteristics, for high spatial frequencies,
MTF value is low. Therefore, it is not possible to obtain an image output with sufficient gradation by the above-mentioned halftone image method, which requires an accurate response for each fine pixel. When the pixel size was reduced and actual recording was performed using this method, the filling of each fine pixel was not faithful. Therefore, an increase or decrease of one minute pixel did not appear as a change in the recorded output. Therefore, when the pixel size is reduced in order to increase the resolution, there is a drawback that the gradation is impaired. Furthermore, when the pixel size is increased in order to improve gradation, the resolution deteriorates.

An object of the present invention is to enable recording and displaying of stable halftone images with high resolution and high gradation using a simple configuration.

In order to achieve such an object, the present invention provides input means (input unit 61 in the embodiment) for inputting image signals.
(corresponding to), and the image signal inputted by the input means is used to generate a character image with a large change in the value of each pixel in a predetermined block, or a gradation with a small change in the value of each pixel in the block. identification means (also the CPU 81) that identifies whether the image is an image and generates an identification signal; storage means (also the frame memory 69) that stores the image signal inputted by the input means in a compressed form in predetermined block units; In order to express the gradation of the image signal according to the identification signal generated by the identification signal and the image signal read from the storage device, halftone processing means (also CPU 81, RAM 87, ROM 93), and when the identification means identifies the input image as a gradation image, the halftone processing means performs halftone processing on the input image signal so as to generate dots in a first period. , when the input image is identified as a character image, the input image signal is subjected to halftone processing so as to generate dots at a cycle smaller than the first cycle.

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 3 shows an output section in an embodiment of the present invention.
Here, the output beam emitted from the semiconductor laser 31 according to the output signal 101 is converted into a parallel beam 35 by the collimator lens 33. After this beam 35 is optically deflected by a rotating polygon mirror 37, θ
Pass through the lens 39. The beam 41 after passing through is irradiated onto the photoreceptor surface 45 of the photoreceptor drum 43 . Here, the main scanning (direction a) of the light beam is performed by rotating the rotary polygon mirror 37 in the direction b. Further, the photosensitive drum 43 is rotated in the c direction to perform sub-scanning.
Here, by turning the semiconductor laser 31 on and off by a modulation means to be described later, an image pattern including a desired halftone image can be formed on the photoreceptor surface 45.

FIG. 4 shows details of the output section in the embodiment of the present invention. Here, the photoreceptor surface 4 of the photoreceptor drum 43
No. 5 uses a so-called NP method, in which an insulating layer is provided on a layer coated with a photoconductor.

In order to form an image, first, the surface of the photoreceptor 45 is uniformly positively charged by corona discharge using the primary charger 51 . Next, alternating current corona discharge is performed by the static eliminator 53 while irradiating the photoreceptor surface 45 with the laser beam 41 . Furthermore, by uniformly irradiating the photosensitive layer using the full-surface exposure lamp 55, image information can be accumulated on the photoreceptor surface 45 as a surface potential distribution, that is, as an electrostatic latent image. Next, the toner particles of opposite polarity in the developing device 57 are transferred to the photoreceptor surface 4.
5 and develop the electrostatic latent image. Thereafter, the developed image is transferred to a transfer material (paper) 59.

The output section explained in FIGS. 3 and 4 is a known technology, and further detailed explanation will be omitted.

FIG. 5 shows a signal processing system for image output.
Here, an input device 61 such as a television camera converts two-dimensional image information into a one-dimensional (time-series) luminance signal 63. The analog quantity of this luminance signal 63 is converted into a digital quantity by an analog-to-digital (hereinafter referred to as A/D) converter 65. The digital signal 67 is supplied to a frame memory 69 for one screen and stored therein. Now, as a recording of a halftone image, for example, as shown in FIGS.
If you select a pixel consisting of 4 micropixels, the density pattern will be 16 in black and white area ratio as shown in the same figure.
There are 17 possible recording tones (including full white). Therefore, when input signals are A/D converted, there are a maximum of 16 levels of quantization, that is, 4
A/D conversion of the bits is sufficient. If we consider the A/D conversion of the input signal as simply encoding, the difference between a pixel consisting of n×m fine pixels and k-bit quantization is 2 ^k ≦n×m ……(1) All you have to do is look at the relationship. If the inequality holds true in this relationship, the number of quantization levels of the input signal is smaller than the number of density patterns in each case. Therefore, gamma conversion or the like can be performed by appropriately diverting a part of the density pattern. At this time, the numerical value obtained by A/D conversion of the input signal can be viewed as simply a coded numerical value (code signal),
All you need to do is decide on a dot arrangement that corresponds to this. Therefore, at the time of output, a density pattern (dot pattern) corresponding to the code signal is output.

As shown in FIG. 7, the input information is stored in the frame memory 69 in the main scanning direction (H address) of the input device.
and the sub-scanning direction (V address), and the addresses are determined for each input pixel and stored in each address of the frame memory 69. Further, in FIG. 7, the numerical values in brackets [ ] represent the density of each input pixel using simple numerical values for explanation, and correspond to the density patterns A to P in FIG. 6. If the A/D converter 65 is a 4-bit one, the code signal representing the density of each pixel indicated by each address in the frame memory 69 is converted into a 4-bit image d ₀ to d ₂ as shown in FIG. 1 bit d ₄ that determines the information section and matrix size
(The matrix size will be described later). A central processing unit (hereinafter referred to as CPU) 81 determines the first bit _d4 out of 5 bits for pixel information for each pixel in the frame memory 69 to determine the matrix size. After that, the frame memory 69
Store in. This operation is repeated over the entire frame memory 69 (one screen).

Next, how to determine the matrix size will be explained. The pixel size used to record the output image is 1:2 as shown in FIGS. 9A and 9B.

As a rule for image processing, when the image information is changing rapidly, the small-sized matrix MXS shown in Figure 9A is output, and when the image information is not changing much, the large-sized matrix MXL shown in Figure 9B is output. I'll do what I do. (The output method will be described later.) In other words, in this embodiment, pixels with a small matrix size are used for thin parts of the image (high spatial frequency components), and coarse parts of the image (low spatial frequency components) are used. By outputting pixels with a large matrix size, both resolution and gradation are increased. This is because human visual characteristics require a lot of gradation for low-frequency components with low spatial frequencies (for example, the skin of a person's face), but for high-spatial frequency components (such as contours). is based on the fact that less gradation is required. As a result, if the matrix size has a large low-frequency component, sufficient gradation can be obtained. Note that in this case, the resolution does not need to be high for low frequency components. Furthermore, if the high frequency component is made to have a small matrix size, the resolution can be made sufficiently high.
Note that in that case, it is not necessary to increase the gradation.

The CPU 81 takes the following steps to determine whether to output the small-sized matrix MXS or the large-sized matrix MXL.

First, as rule 1, take out every 4 pixels,
If all of these four pieces of information are the same, a large matrix MXL is created. That is, pixel addresses (V, H) in FIG. 7 = (2, 0), (3,
Taking four pixels 0), (2,1), and (3,1) as an example, they all have the same density of 4. In such a case, draw with a large pixel size as shown in FIG. 10A.

Next, as rule 2, every four pixels are taken out, and if the difference between these four pieces of information is within 1, a large matrix MXL is created. That is, the seventh
Pixel address in the figure (V, H) = (4, 0),
Taking four pixels (5,0), (4,1), and (5,1) as an example, the difference is 1 between densities 4 and 5. In such a case, a matrix MXL with a large pixel size is used as shown in FIG. 10B. The concentration at this time is
It is assumed that the average value of the four pixels takes a value closer to the quantization level.

Further, as rule 3, if the difference between the four pieces of information is larger than 1 (2 or more) by extracting every 4 pixels, a small matrix MXS is used for each pixel. That is, the pixel address (V, H) in FIG. 7 = (0, 0), (1, 0),
Taking four pixels (0, 1) and (1, 1) as an example, their densities are 2, 3, 4, so the difference is greater than 1. Therefore, in this case, Figure 10
A small matrix MXS is used to draw density patterns corresponding to each pixel such as C ₁ , C ₂ , C ₃ , and C ₄ . Although the difference in density is assumed to be 1 here for the sake of explanation, in reality, an appropriate difference may be determined.

Based on these three rules, it is determined whether all pixels are drawn using the large size matrix MXL or the small size matrix MXS. In this case, the combination of four pixel units is the upper left side of the frame memory 69 ((V,H)=(0,0)
), the units are sequentially determined as the units enclosed by the dotted lines in FIG. 7. Therefore frame memory 6
The number of addresses of 9 is an integral multiple of 2 in both the main scanning direction and the sub-scanning direction.

By the way, another effect of Rule 2 is that the false contour is relaxed. In other words, as mentioned above, in order to faithfully produce gradation, it is better to draw with a large matrix, and if you draw a part where the image changes slowly (for example, human skin) with a small matrix, it will look like a false image. The outline appears. Therefore, if a small difference in image information is drawn using a large matrix as in Rule 2, the gradation will be more natural and smooth.

If low-frequency components in which the image changes slowly in this way are drawn with a small-sized matrix, the false contours will be emphasized, so Rule 2 is provided to eliminate this drawback. However, a weakness of Rule 2 is that thin information with a difference value of 1 is easily ignored. However, this weakness can be avoided by performing image processing for edge enhancement as a preprocessing as shown in FIG. 11 before inputting the density data as a code signal to the frame memory 69.

That is, in FIG. 11, the contour emphasis circuit 71
By providing the circuit before the A/D converter 65 or by providing the contour enhancement circuit 73 after the
The weaknesses of can be avoided. Note that the contour emphasis circuits 71, 7
3 is a known technique, and detailed explanation thereof will be omitted.
Components having the same functions as those in FIG. 5 are given the same numbers.

By the way, when determining the pixel size performed by the CPU 81, the first bit _d4 of the 5 bits provided for each pixel shown in FIG. "0"
The first step is to determine the bit information as follows. Such a judgment operation is performed by the CPU 81 via the bus lines 83 and 85, using the information stored in the frame memory 69 as a random access memory (hereinafter referred to as RAM) 8 as a working area.
Import into 7. After the calculation is performed in this RAM 87, it is returned to the frame memory 69 again. FIG. 12 is a flowchart showing such a pixel size determination operation performed by the CPU 81. Further, FIG. 13 is a diagram showing each address in the frame memory 69. In FIG. 13, M represents the number of pixels for one line of the frame memory 69, and L represents the number of lines of the frame memory 69. Further, Ai indicates the address of line N, and Bi indicates the address of line N+1.

The flowchart in FIG. 12 will be explained.

In stepl, line N of frame memory 69
Set to 1. In step 2, two lines of data (N, N+1) are read from the frame memory 69 and written to the RAM 87. In step 3, the address of each line is set to 1. Ai, Ai ₊₁ , Bi, written in RAM87 in step 4,
Load Bi ₊₁ data into CPU81. In step 5, the CPU 81 compares the differences between each data of Ai, Ai ₊₁ , Bi, and Bi ₊₁ . When the difference is "0" or "1", the 5 pixels provided for each pixel shown in FIG.
Insert "1" into the first bit _d4 . Also, if the data difference is greater than 1, d ₄ is “0”.
Put in. In step 6, each data of Ai, Ai ₊₁ , Bi, and Bi ₊₁ containing bit information in the first bit _d4 is returned to the frame memory 69. step 4 in step 7
It is determined whether the operation in step 6 has been performed on all data of the specified two lines. Further, if the pixel size has not been determined for all data, the operations from step 4 to step 6 are repeated. In step 8, it is determined whether the operations in steps 2 to 7 have been performed for all lines in the frame memory 69, and when the line reaches L,
The CPU 81 ends the pixel size determination operation.
If this has not been done for all lines yet, repeat steps 2 to 7.

Next, the image information in the frame memory 69 given the matrix size is output as an image in the following procedure. Here, the output section 91 is assumed to be the laser beam printer described in FIGS. 3 and 4.

First, the CPU 81 takes in information in the frame memory 69 in units of four pixels (in units of dotted lines in FIG. 7). It is determined whether bit _d4 , which indicates the matrix size of the captured image information, is "1" or "0". Therefore, when bit _d4 is "0", a small size matrix MXS is commanded, and when it is "1", a large size matrix MXL is commanded to be output. Please note that read-only memory (hereinafter referred to as ROM)
A dot array pattern for each value of image information is stored in the dot array 93. Therefore, bit
Pattern data corresponding to the code information _d0 to _d2 is read from the ROM 93 via the bus line 95. Next, according to the previous size command data, the pattern data is supplied to the line memory 99 via the bus line 97, and the pattern data is written for each pixel according to the specified size. Thereafter, the output device 91 receives an output signal 101 based on the data written in the line memory 99, and sequentially records the signal by the operation described in FIGS. 3 and 4.

As shown in Fig. 14, the line memory 99 stores 2 pixels in the vertical direction with a small matrix size (1 pixel with a large matrix size), and the frame memory in the horizontal direction with a small matrix size (pixels). Determine the size of one line in the H direction of 69. That is, the line memory 99 has a size that can store two lines of pixels in the frame memory 69 in a dot pattern. However, one pixel with a small matrix size has 16 bits (=4 x 4), and each bit is assigned either black (“1”) or white (“0”). sell.
The way to put it is as shown in Figure 14.
When the output unit 91 outputs and records an image, it scans in the horizontal direction and reads them sequentially (main scanning direction), and then sends them in the vertical direction (sub-scanning direction). The information is set to “0” by the output signal 101 from the line memory 99.
If it is "1", the laser 31 is turned off, and if it is "1", the laser 31 is turned on and recorded as a white part or a black part, respectively, on the transfer material 59.

15A to 15P show dot patterns stored in the ROM 93. Here, the concentrations indicated by A to P in the figure are [1], [2]...
…Compatible with [16]. Therefore, if the value taken out from the frame memory 69 is [2] and bit _d4 representing the matrix size is "0", the data representing the dot array pattern of FIG. 15B is stored in the line memory 99. be transferred. At the same time, dot pattern data of corresponding densities for the other three pixels are transferred to the line memory 99.

By the way, the bits that represent the matrix size
When d ₄ is "1", this means that the data for four pixels are all equal (rule 1 described above) or when the difference is 1 (rule 2). In that case, 4
Add the two data, take the average, round it to an integer, and calculate the dot pattern for each level value.
5 Choose from Figures A to P. Incidentally, since the density data is actually represented by 4 bits, it is sufficient to obtain four density data and take the average thereof. Thereafter, the matrix size is doubled, that is, the large matrix MXL is stored in the line memory 99.

As a result, the frame memory 69 shown in FIG.
The information in the line memory 99 is converted into a dot array as shown in FIG. If this is outputted to an output section 91 consisting of a laser beam printer, the recording pattern shown in FIG. 10 will be obtained. In this way, density patterns having two different matrix sizes are output and recorded.

The above operation will be explained using the flowchart of FIG. 16.

In step1, line N of frame memory 69
Set to 1.

In step 2, two lines (N, N+1) of the frame memory 69 are designated, and the address of each line is set to 1. In step 3, the four addresses Ai, Ai+ ₁ , Bi, Bi of each line specified in step 2 are
₊₁ data is taken into the CPU. In step 4, it is determined whether d ₄ is O. If bit _d4 is O, indicating a small matrix size, Ai from ROM93,
The dot patterns of the small size matrix corresponding to each data of Ai+ ₁ , Bi, and Bi+ ₁ are taken out and transferred to the line memory 99. If hit d ₄ is 1, that is, large matrix size, CPU81
Adds each data of Ai, Ai+ ₁ , Bi, Bi+ ₁ , takes the average, and creates a dot pattern corresponding to that value.
Take it out from ROM93. Thereafter, the matrix size is doubled and transferred to the line memory 99 as a large size matrix dot pattern.
In step 5, it is determined whether the operations in steps 3 and 4 have been performed on all data of the designated two lines. If the dot patterns for all data for two lines have not yet been transferred to the line memory 99, the operations of steps 3 and 4 are repeated. In step 6, the data (dot pattern) written in the line memory 99 is output. In step 7, it is determined whether the operations from step 2 to step 6 have been performed for all lines in the frame memory 69, and when the line reaches L,
CPU81 stops working, otherwise
Repeat steps 2 to 6.

Here, a case will be described in which the matrix size described in step 4 is converted from a 4×4 dot pattern to an 8×8 dot pattern.

FIG. 17 is a flowchart for doubling the matrix size. In addition, here 4×4
Each component of ROM data is ak, l (o≦k, l≦3)
shall be. Also, each component of the 8×8 dot pattern is
b _2k , _2l , b _2k+1 , _2l , b _2k , _2l+1 , b _2k+1 , _2l+1 (0≦k
，
l≦3).

In step 1, set l=o. In step 2, set k=o. In step3, b _2k , _2l = b _2k+1 , _2l
= b _2k , _2l+1 = b _2k+1 , _2l+1 = ak, l as 4×4
Each component of the 8×8 dot pattern is determined for each component ak, l of the ROM data. In step 4, it is determined whether k=4, and if not, the operation in step 3 is repeated. In step 5, l =
4, and if not,
Repeat steps 2 to 4.

In this embodiment, the size of the matrix is doubled, but a density pattern of twice the matrix size may be stored in the ROM from the beginning.
Alternatively, the density pattern shown in FIG. 6 may be formed without being stored in the ROM.

By the way, the actual size of a pixel is a square with a side of 0.4mm for a large one, and a square with a side of 0.2mm for a small one.
mm square. Therefore, each fine pixel is a square with a side of 100 μm and a square with a side of 50 μm. These pixels come together to create a picture. In the density pattern method shown in FIG. 6, an output similar to a halftone photograph is obtained.

Furthermore, although the shape of the matrix in this embodiment is a square, it is also possible to create an image by combining a matrix with a part of the square protruding, a matrix with a part of the square concave, or other deformed matrices. The present invention is acceptable if it can be formed.

In the above processing, it takes a certain amount of time to convert the dot pattern from the frame memory 69 into the line memory 99 and store it.

However, the output section 91 continues to output continuously. Therefore, in order to eliminate the time lag, the line memory 99 is provided with two similar line memories as shown in FIG. Each of the A-line memory 991 and B-line memory 992 has a configuration as shown in FIG.

While the output unit 91 is outputting the information stored in the A-line memory 991, the CPU 81 continues to send the next information to the B-line memory 992. By sequentially switching between reading and storing in this manner, high-speed recording without time lag is possible.

Note that the signal processing system of the present invention is not limited to the embodiment shown in FIG. 5, and various modifications can be considered. For example, input signal 6
After A/D converting 3, the density data may be immediately input into the RAM 87 and the matrix size may be designated. In this case, the RAM 87 must store at least two lines of image data. In this way, other modified signal processing systems may be used as long as the basic means of signal processing is not destroyed.

Next, a memory compression technique in this embodiment will be described. This is a further feature of this embodiment, and is a useful technique for storing a large amount of information with a small amount of memory.

Using the method described above, if the frame memory 69 has a capacity of 512 x 512 pixels, the density of each pixel and matrix size are represented by 5 bits, so 512 x 512 x 5/8 ≒ 164 kbytes (2) Required. It was hot. In this case, if it is assumed that each pixel is stored in the memory 69 as a 5-bit code, it can be said that the code is compressed compared to when each pixel is stored in a 16-bit dot pattern. However, it can be further compressed by using the following method.

FIG. 18 shows the principle, in which differential quantization is performed for every four pixels. That is, for pixels 1 to 4 shown in FIG. 19, the values of pixels 2 to 4 are given as difference values with respect to the original value of pixel 1. By doing so, it is possible to reduce the number of bits where a difference is given based on the correlation depending on the position of the image. FIG. 20 shows the case where pixel 1 is given with 4 bits, other pixels are given with 3 bits, matrix size is given with 1 bit, and overflow is given with 1 bit. By the way, the matrix size can be given in units of four pixels, so four pixels and 15 bits are sufficient. Note that in the method described above, it is 20 bits. However, since the values of pixels 2 to 4 cannot always be represented by 3 bits, a bit is provided to represent overflow, and in that case, the difference is again given to the corresponding pixel using the next 15 bits. However, since the probability of such appearance in a general image is extremely low, the overall image is considerably compressed. However, when giving a difference, + and - are generally required. Therefore, the first bit of the above-mentioned 3-bit image information is a code bit.

In this way, resolution and gradation can be improved by outputting pixels with different matrix sizes in accordance with the spatial frequency characteristics of the image.

In the above explanation, one pixel is formed by 4×4 micropixels, and the size of one pixel on the recording surface is explained using only two types of 1:2. It is not limited to this. A large number of pixel sizes may be selected.

Further, the density pattern shown in FIG. 6 is not limited to this, and the number of types may be increased, or the way in which fine pixels are filled may be changed. Further, although the electrophotographic method has been described, it goes without saying that the present invention can also be applied to other general recording methods that have a low MTF value for high spatial frequencies. The input section 61 may be a device for reading original images or a device for receiving image information, and the output section 91 may be another recording device, a display device such as a CRT, or a dot device.

As described in detail above, according to the present invention, it is possible to realize an image processing device with a simple configuration and excellent resolution and gradation.

[Brief explanation of drawings]

1A to 1E are diagrams showing examples of pixels with a matrix configuration, FIG. 2 is a characteristic diagram of a photoreceptor, and FIGS. 3 and 4 are configurations showing an output section in an embodiment of an image processing apparatus according to the present invention. 5 is a block diagram of an embodiment of the present invention, FIGS. 6A to 6P are pattern diagrams of an example of halftone display by one pixel, and FIGS. 7 and 18.
The figure shows the frame memory 69 showing the arrangement of pixels to be recorded, FIG. 8 shows the code of the encoded signal, FIGS. 9A and B show the matrix forming one pixel, and FIG. 10 is a diagram showing a recorded image, the first
Figure 1 is a block diagram when using the contour enhancement circuit.
FIG. 12 is a flowchart for determining pixel size, FIG. 13 is a diagram showing each address of the frame memory 69, FIG. 14 is a diagram of image information stored in the memory, and FIG. 15 A to P are bit state diagrams of various density patterns, FIG. 16 is a flowchart for outputting image information, FIG. 17 is a flowchart for converting matrix size, and FIGS. 19 and 20 are explanations of pixel configurations. It is a diagram. 31... Semiconductor laser, 37... Rotating polygon mirror,
45...Photoreceptor surface, 59...Transfer material, 65...
Analog-digital converter, 69...Frame memory, 81...Central processing unit, 91...Output section, 99...Line memory.

Claims (1)

[Scope of Claims] 1. An input means for inputting an image signal, and using the image signal inputted by the input means, a character image with a large change in the value of each pixel within a predetermined block or a character image within the block. identification means for identifying whether the gradation image has a small change in the value of each pixel and generating an identification signal; storage means for storing the image signal inputted by the input means in a compressed form in units of predetermined blocks; halftone processing means capable of generating dots at mutually different cycles in order to express the gradation of the image signal according to the identification signal generated by the identification signal and the image signal read from the storage means; , when the input image is identified as a gradation image by the identification means, the halftone processing means performs halftone processing on the input image signal so as to generate dots in a first cycle, and identifies the input image as a character image. The image processing apparatus is characterized in that the input image signal is subjected to halftone processing so as to generate dots at a cycle smaller than the first cycle.