JPH0521385B2 - - Google Patents

Description

[Detailed description of the invention]

<Technical Field> The present invention relates to an image processing device such as a digital copying device or a facsimile device that handles images as electrical signals. <Prior art> With the development of digital devices, so-called digital copying devices, which generally sample images using a CCD sensor and output the digitized data from a digital printer such as a laser beam printer to reproduce the image, have changed from the conventional technology. It is becoming widely used as an alternative to analog copying machines. In order to reproduce halftone images, such digital copying apparatuses generally perform tone reproduction using a dither method or a density pattern method. However, this method has the following two major problems. (1) If the original is a halftone image, the copied image will have periodic striped patterns that are not present in the original. (2) If the manuscript contains line drawings, characters, etc.
Dither processing causes edges to become cut off, reducing image quality. The phenomenon (1) is called a moiré phenomenon, and its causes are considered to be: A. Moiré caused by the halftone original and input sampling B. Moiré caused by the halftone original and the dither threshold matrix. The phenomenon of A) is caused by the dot frequency 0 (=
1/P0) [PEL/mm] n times the high frequency n0 [PEL/mm] and the input sampling frequency s (1/Ps) [PEL/mm] found from the input sensor pitch Ps [mm], Δ= |s−n0|[PEL/mm] ...(1) A beat frequency occurs, which becomes moiré. The phenomenon (B) generally occurs when the dither threshold is fatting.
When the dots are arranged in a concentrated manner, such as in a mold, the output image also has a pseudo-dot structure, which causes a beat between the dots and the input dot document, resulting in a moiré phenomenon. If the repetition period pitch of the dither threshold value is P _D [mm] on the recording paper, the spatial frequency is _D = 1/P _D
[PEL/mm], and the most prominent beat frequency is Δ=|0− _D |[PEL/mm]...(2). Of the two moiré phenomena mentioned above, the one that occurs most strongly is (B)
It is the person who is. In the phenomenon of (A), this is generally n = 3 to 6, where n is the n times higher frequency of the halftone original, and the transfer function (MTF) of the optical system etc. that leads to the sensor is
The contrast of the moiré fringes is also low because it drops considerably at that frequency. Moiré phenomena caused by such causes significantly degrade the quality of output images. For this reason, various countermeasures and studies have been carried out for a long time. For example, the random dither method can remove moiré, but it produces grainy grain and degrades the image quality. J.Opt.
ARIES presented by Paul G.Roetling as shown in Soc.Am., Vol.66, No.10, October 1976 P985
compares the average value of the density before and after binarization and applies a feedback to the threshold value to make them equal, but this method requires complicated hardening and is not effective in removing moiré. . On the other hand, the re-halftone method seen in Takashima et al.'s Proceedings of the Institute of Image Electronics Engineers, 83-3, P13 "Half-dot conversion of mixed text/photo images" uses dithering to blur the half-tone image (or average the surrounding pixels). Moiré is removed by redotting the pattern, and grainy noise is also reduced. However, the resolution cannot be avoided due to blurring (or averaging with surrounding pixels). That is, if you try to remove moire, the resolution will decrease, and if you try to maintain the resolution, moire will not be removed. Therefore, it is essential to extract only the halftone image area in advance and apply the method only to that part. For this reason, a so-called image area separation technique is required. At the current level of image area separation technology, it is difficult to obtain a high-accuracy and high-speed method, especially a method suitable for implementation in hardware, and it is difficult to realize the above-mentioned method. Even if an image area separation technique were to be obtained, such a technique would not be fully satisfactory because even high frequency components within the image would be averaged and smoothed. On the other hand, regarding problem (2), the characters and line drawings on the document are fragmented by dithering, and the print quality is degraded because the digital parts in particular become cut off. This phenomenon is caused by the dither pattern described above.
This is particularly noticeable in dot-concentrated types such as Fatting type. <Object> The object of the present invention is to eliminate the two drawbacks described above and to provide an image processing device that can reproduce images with high quality and high definition. That is, in the present invention, the moiré phenomenon that occurs in halftone originals is removed, characters and line drawings are prevented from being cut into pieces, and high-frequency components in image areas are prevented from deteriorating. It happened. Furthermore, this method can be realized with a simple circuit configuration and can be provided at low cost. <Example> (Basic configuration) The basic configuration of the image processing apparatus of this example is shown in FIG. This image processing device is composed of an edge detector a, a smoother c, and a mixer d. The edge detector a detects the edges of characters, line drawings, and images as described later, and the halftone dots of the halftone image are given a spatial frequency characteristic that does not detect them as edges. The smoother c smoothes the image. The mixer d changes the mixing ratio of the input image and the smoothed image according to the signal from the edge detector a and outputs the mixed image. In this way, the halftone dots of the halftone dot image are determined to be non-edge areas, and smoothing is performed to average them and prevent moiré. Furthermore, since the edge area and the non-edge area are continuously connected, there is no texture change at the boundary. Next, the principle of this embodiment will be explained from the perspective of frequency characteristics. First, the number of screen lines for the halftone image of a document is usually 120 to 150 lines for black and white, and 133 to 175 lines for color. Moire is most likely to occur when the screen angle is 0 degrees or 45 degrees. Further, the halftone dot pitch in the main scanning direction during reading is maximum when it is 45 degrees and the spatial frequency is low, and is minimum when it is 0 degrees and the spatial frequency is high. Table 1 shows the spatial frequencies when the screen angle is 0 degrees and 45 degrees.

[Table] The frequency characteristics of such a halftone image have peaks at the fundamental frequency and its high frequencies, as shown in Figure 2a.
Further, the frequency characteristics of a character image and a continuous tone photographic image are as shown in FIGS. 2b and 2c, respectively. For such a mixed image of characters, photographs, and halftone dots, the spatial filter of the edge detector and smoother of this embodiment has a frequency characteristic that satisfies the following conditions. Condition 1: The peak frequency of the spatial filter of the edge detector is set to be lower than the first harmonic frequency of the halftone image. Condition 2: The frequency characteristics of the spatial filter of the smoother are sufficiently reduced at the first harmonic frequency of the halftone image. Also, the frequency is sufficiently lowered to correspond to the period of the output dither. There are various types of spatial filters for edge detection, but if the matrix size, which affects the scale of the hardware circuit, is held constant, the first-order differential filter has a peak at a lower frequency than the second-order differential filter. . However, the second-order differential filter does not have directionality, but the first-order differential filter has directionality, and the square root of the sum of the squares of the slopes in at least two directions, or as an approximate expression thereof, the absolute value of the slopes in at least two directions. It is necessary to take the sum of the values or the maximum value of the absolute values of the slopes in at least two directions. Furthermore, the first-order differential is more resistant to point noise than the second-order differential. As described above, it is better to use a first-order differential filter as the spatial filter for the edge detector a. In addition, as a spatial filter for edge enhancer b, there is no directionality and the peak is at a higher frequency.
A second-order differential filter is better than a first-order differential filter. For simplicity, the relationship between the frequency characteristics of the various spatial filters as described above is calculated using a one-dimensional fast Fourier transform (FET).The results are shown below. For example, the reading sampling interval of the input system is 1/16mm, and the output system is
Calculation is performed using a 4×4 dither matrix at 16 dots/mm. The period of the dither pattern is 41/mm in terms of spatial frequency. Furthermore, in reading with 1/16mm sampling, only frequencies up to 81/mm can be detected due to the sampling theorem. If the matrix size is 5 x 5, one dimension of the second-order differential filter (-1, 0, 2, 0, -1)
The FET is shown in Figure 3, and the first-order differential filter (-1,
0, 0, 0, 1) is shown in Figure 4, and another 1-dimensional differential filter (-1, -1, 0, 1,
Figure 5 shows the one-dimensional FET of 1). The peak positions are 4 1/mm and 2, respectively.
1/mm, 2.5 1/mm. Comparing this with the spatial frequency of the halftone image in Table 1, the first-order differential filter satisfies the condition of 1 for all lines in Table 1, but the second-order differential filter satisfies the condition of 1 for 120 lines,
Condition 1 cannot be satisfied at 45° of the 133 line, and the halftone dot is detected as an edge. Comparing two types of first-order differential filters, we found that the pulse width was increased (-1,
-1,0,1,1) is better. This is because the larger the pulse width, the smaller the intensity of the second peak, and the larger the pulse width, the wider the edge region can be detected. Next, apply a 5×5 smoothing filter (1,
1, 1, 1, 1) one-dimensional FET is shown in Figure 6.
Fundamental frequency of halftone image of 120 lines 45° or more, 3.341
The strength decreases above 1/mm. Also 4
×4 pitch of dither matrix, 4 1/
The strength is sufficiently small at mm and satisfies condition 3. In this embodiment, by using a spatial filter with frequency characteristics such as the conditions 1 and 2 described above for the edge detector and smoother, the flat part of the image and the halftone image are determined to be non-edge areas, and are averaged by smoothing. Text, line drawings, and edges of images are output as is. Further, the boundary between the edge region and the non-edge region is continuously connected by changing the mixing ratio in the mixer according to the signal from the edge detector. As described above, moiré in halftone images is prevented, halftone dot formation of characters and reduction in image sharpness are prevented, and discontinuous changes in texture between edge areas and non-edge areas do not occur.
Furthermore, since the matrix size of the spatial filter does not need to be large, the scale of the hardware circuit can be reduced, which is advantageous for LSI implementation. FIG. 7 is a block diagram showing an embodiment of the present invention.
S1 is the input image signal, 1 is 1 of the input image signal S1
The first differential value detector detects the absolute value of the second differential value.
Corresponds to figure a. S2 is a differential signal connected to the output of differential value detection section 1, 2 is a control signal generator that generates control signals S3 and S4 from differential signal S2, S3 is the output of control signal generator 2 and is a control signal, and S4 is also a control signal. The output of the signal generator 2 is a complementary control signal to the control signal S3, and 3 is a smoothing processing section for smoothing the input image signal S1, which corresponds to FIG. 1C. S6 is a smoothed image signal smoothed by the smoothing processing unit 3, 4 is a multiplier that takes the arithmetic product of the smoothed image signal S6 and the control signal S3, S7 is the output of multiplier 4, 8 input images A multiplier that takes the arithmetic product of the signal S1 and the control signal S4, S12 is the output of the multiplier 8, 9
is an adder that takes the arithmetic sum of output S7 and output S12,
S13 is the output of the adder 9 and is a processed image signal.
Here, multipliers 4, 8 and adder 9 constitute mixer d in FIG. The mixing unit 305 is a circuit unit that mixes the input image and the output of the smoothing process 3 at an appropriate ratio.
S3 and S4 are output in accordance with the output of the differential value detection unit 1 that is input to the control signal generation unit 2 of the front break. As described later, S3 and S4 are complementary control signals, but are not necessarily limited thereto. The characteristics of S3 and S4 can be freely selected and set using the control signal S5. Multiplier 8 performs multiplication determined by input image S4, multiplier 4 multiplies the output of the smoothing processing unit determined by S3, and outputs of multipliers 4 and 8 are applied to adder 9. Therefore, they are added together, and this becomes the image processing output. The block diagram of FIG. 7 can also be expressed by the following equation. First, the differential value detection section 1 and the control signal generator 2 perform calculations as shown in Equation 1 below. E=(|1 1 0 -1 -1 1 1 0 -1 -1 1 1 0 -1 -1 1 1 0 -1 -1 1 1 0 -1 -1 〓*1 ｜ ｜ ｜ ｜ ｜ ｜ 〓・I｜+｜110-1-1 110-1-1 110-1-1 110-1-1 110-1-1〓*2 | | | | | | | 〓・I|) …(Formula 1) Here I is input image data, E is control signal S4
It is. 5 is a normalization function that normalizes the control signal S4 to a maximum value of 1. Then, the following output is obtained from the smoothing processing section 3. H=11111 11111 11111 11111 11111·I (Formula 2) H is the output of the smoothing processing section 3. Therefore, the value O of the output S13 of the adder 9 can be expressed by the following equation 2. 0=E・I+(1-E) (11111 11111 11111 11111 11111〓＊３ ｜ ｜ ｜ ｜ ｜ 〓I)
...(Equation 3) In the above equation, the part in [ ] is a kernel that performs convolution with the image signal I. Kernel*1
~*3 can be modified in various ways, an example of which is shown in Table 2.
Shown below.

[Table] FIG. 8 shows an example of the characteristics of the function of the control signal generator 2. In E=(x), E=0 for 0x<0.2 E=1.67x−0.33 for 0.2x<0.8 E=1 for 0.8x1. However, both input and output signals will be standardized to 0 to 1 for explanation. FIGS. 9 and 10 are diagrams for explaining the operation of the edge detecting section 1 (hereinafter referred to as differential value detecting section) based on first-order differentiation, and are shown in one-dimensional main scanning. As already explained, the differential value detection section 1 is a kind of band-pass filter, so the ninth
The input image signal S1, such as a halftone image with high frequency components shown in the figure, has a kernel (-1, -1, 0,
1,1) in the main scanning direction, the output signal S2 becomes a small value of 0.1-0.2. On the other hand, in the case of a relatively low-frequency input image signal (for example, vertical lines of characters), as shown in FIG.
Due to similar convolution, the output signal S2 takes on a large value. Here, the control signal generating section 2 that performs gamma conversion sets the control signal S3 to 1 and the control signal S4 to 0 when the differential signal S2 is smaller than 0.2, as shown in FIG. If the differential signal S2 is larger than 0.8, the control signal S3 is set to 0 and the control signal S4 is set to 1. Furthermore, when the differential signal is between 0.2 and 0.8, the control signal S3 and the control signal S
It changes according to the differential signal S2 so that the sum of 4 becomes 1. On the other hand, the input image signal S1 is connected to the input of the differential value detection section 1 and is also connected to the smoothing processing section 3 at the same time. FIG. 11 shows the operation of the smoothing processing section 3, and for the sake of explanation, shows a one-dimensional example in the main scanning direction. In this case the kernel is (1, 1, 1, 1,
In 1), the contents are all 1 and the low-pass filter is configured to output the average of 5 pixels, and the input image signal S1 is the smoothed image signal S6.
become that way. By the way, in the case where the output of the differential value detection section 1 is large, that is, at the edge portion, the control signal S3 is small and the control signal S4 is large. Conversely, when the differential signal S2 is small, S3 is large and S4 is small. Also, as mentioned in Figure 8, S3
and S4 are always gamma-converted so that their sum becomes 1. Therefore, the sum of the outputs of the multipliers 4 and 8 is controlled so that when the differential signal S2 is large, the input image S1 component is large, and when S2 is small, the smoothed signal S6 component is large. Figure 12 shows this situation, where the differential signal S
2 indicates that edges are detected in a portion of the input image signal S1 excluding vibrations with a short period (corresponding to the halftone dot period). The control signal S4 is obtained by gamma-converting the differential signal S2, and all but the four peaks of the signal S2 shown in FIG. 12 are set to zero. S3 is naturally 1-S4. Furthermore, FIG. 12 also shows the smoothed signal S6. In FIG. 12, the processed signal S13 is a signal obtained by adding S6 and S1 at the ratio of S3 and S4, smoothing the halftone dot part, and emphasizing only the edge part. (Description of each part) The input image signal S1 shown in FIG. 7 consists of data for five consecutive lines of image data, as shown in FIG. After being stored at , five lines are simultaneously outputted and outputted pixel by pixel in the main scanning direction of the image data in synchronization with an image transfer clock (not shown). Furthermore, as shown in Fig. 19, the attention area S in the image area is further enlarged, and the pixel data of interest is
Consider the surrounding image data when set to Smm. (Differential value detection unit 1) FIG. 13 is a detailed circuit diagram of the differential value detection unit 1. The output of the 5-line buffer 301 shown in FIG. 18 is input to the differential value detection section 1. In the figure, 306 is a first-order differentiator, and its output 306-a is divided into a data section 306-c and a sign section 306-b indicating positive/negative. The absolute value of the data 308-a is determined by selecting either the input data and +/− inverted data by the inverter 307 or the data 306-c.
is obtained. Similarly, the absolute value of the output of the first-order differentiator 312 is output from the selector 311, 308-a and 311-a are added by the adder 309, and the sum of the first-order differential values in two directions is output from the adder 309. Ru. Next, a block diagram showing details of the first-order differentiators 306 and 312 shown in FIG. 13 is shown in FIG. 14. First, in order to explain the basic operation of this first-order differential circuit, block X in FIG. 14 will be explained. First, all shift registers in FIG. 14 are shifted in synchronization with an image transfer clock not shown in the figure. To simplify the explanation, the multipliers 243~
Assume that all 247 multiplication coefficients are 1. According to timing chart Figure 15, the output of the shift register 230 at t-3 is Sn,m-1+Sn,m-
2, and the shift register 231 at t-2
The output of the shift register 232 at t-1 is Sn,m+Sn,m-1+Sn,m-2, and the output of the shift register 232 at t-1 is Sn,m+1+Sn,m+Sn,m-1+Sn,m-
2, the output of adder 260 at t0 is Sn,m
+2+Sn, m+1+Sn, m+Sn, m-1+Sn,
It is m-2. In this way, the sum of five pixels in the main scanning direction is calculated in block X. Here, the multiplication coefficients of the multipliers 243-247 are a, b, c,
By setting d and e, the output of the adder 260 becomes e・Sn, m+2+d・Sn, m+1+c・
Sn, m+b・Sn, m−1+a・Sn, m−2. It can also be seen that the circuits after the shift register 223 and the circuits after the shift register 233 operate in the same manner. By the way, the first-order differential to be sought is *1, *2 in formula 1.
In a case like this, if the pixel of interest is on the nth line,
The kernel elements of the n-2 line and the n-1 line are equal, and the kernel elements of the n+1 line and the n+2 line are also equal. Therefore, by adding the pixel data at the n-2nd line and the n-1st line using the adder 231, and then performing first-order differential processing as shown in equations 1*1 and *2, the circuit size can be reduced to 1/2. It can be reduced to Also, n+
The same applies to the first line and the (n+2)th line. In this way, the adder 400 can obtain addition values corresponding to the kernels for five lines. In addition, when the multipliers shown in 238 to 252 have a simple multiplication value such as 1or-1or0, the first
As shown in Figure 7, it can be easily configured with an inverter 291 and a selector 292, and SL in the figure
It can be switched 1 or -1 and set to 0 using CL. In addition, to obtain the first derivative of the kernel *1 in Equation 1, the multiplication coefficients of the multipliers 238 to 242 are set to 1,
The multiplication coefficients of multipliers 243 to 247 are set to 0, and the multiplication coefficients of multipliers 248 to 252 are set to -1. To find the first derivative of the kernel *2 in Equation 1,
Multipliers 242, 238, 239, 240, 241
Let the multiplication coefficients of
1, and the multiplication coefficients of the multipliers 248 to 252 are 1,
1, 0, -1, -1. In addition, the circuit configuration for calculating the first-order differentials such as *1 and *2 in Table 2 can be realized using FIG.
Since the operating principle is the same as in FIG. 4, the explanation of how to give coefficients to the multiplier will be omitted. (Smoothing Processing Unit 3) Next, details of the block of the smoothing processing unit 3 shown in FIG. 7 are shown in FIG. The image signal S1 consists of five lines of data that are continuous in the sub-scanning direction of the image, and the adder 271
Addition of 5 pixels in the sub-scanning direction is performed. This data is transferred to a shift register 272 to be delayed by 1 bit.
is input. The output data of shift register 272 is input to adders 277-280. The adder 272 outputs the output of the shift register 272 and its 1
The data before the pixel is added. The result of this addition is 2
278 after being latched into 74 shift registers
The adder adds the next pixel. Similarly, at time T2, the adder 280 outputs S _N , m+2+S _N , m+1+S _N , m+S _N , as shown in FIG.
m-1+S _N , m-2 is output (S _N , j=S _N
−2,j+S _N −1,j+S _N ,j+S _N +1,j+
S _N +2,j). In this way, when the pixel of interest is Sn, m, the pixel sum represented by *3 in equation 2 is output from the adder 280 and divided by the total number of pixels by the divider 281 to obtain smoothed data. FIG. 23 shows a circuit that performs weighted smoothing as shown in FIG. 22. Although the operation timing etc. are the same as in FIG. 20, in FIG.
Smoothing as shown in FIG. 22 is performed by weighting by 360. This smoothing processing section adds all the added values in the image sub-scanning direction and then adds them in the image main scanning direction, so the circuit size can be reduced. (Other Examples) In this example, the kernel of the differential value detection unit and the smoothing processing unit is 5×5, but depending on the number of lines aimed at removing moiré, it may be 3×3 or 5× 5
More than this may be necessary in some cases. Furthermore, depending on the purpose, it is not necessary to use kernels of the same size for the differential value detection section and the smoothing processing section. Furthermore, the kernel does not have to be square. Further, in this embodiment, a set of 5-line buffers is provided and edge detection, smoothing, and edge enhancement are performed in parallel processing, but it is not necessarily necessary to perform them in parallel. Furthermore, in this embodiment, the smoothed signal S6, which is the output of the smoothing processing section 3, and the input image signal S1 are added at a ratio according to the output of the gamma conversion section 2, but S
1, a signal obtained by edge-emphasizing the input image signal S1 may be used. In addition, in the embodiment of the present invention, the characteristics of the control signal generator 2 that performs gamma conversion have been explained as shown in FIG.
An example of a modification of the characteristics of is shown below. In FIG. 24, only the characteristics of the control signal S4 are shown, but the control signal S3 is expressed as S3=1-S4. FIG. 24-a has the following characteristics: S4=0, but 0<S2<0.5, S4=1.0, but 0.5<S2<1.0, and is particularly characterized in that the circuit of the gamma conversion section can be constructed easily. FIG. 24-b has the characteristic S4=-arctan(k·S2+k), and is particularly characterized in that the connection between the smooth signal and the edge emphasis signal is smooth. Figure 24-c has the following characteristics: S4=0 but 0 <S2<0.25 S4=0.33 but 0.25<S2<0.5 S4=0.67 but 0.5<S2<0.75 S4=1.0 but 0.75<S2<1.0; Compared to the embodiment shown in the figure, the circuit is relatively simple, and the connection between the smoothed signal and the edge emphasis signal is smoother than when the characteristics of the gamma conversion section shown in FIG. 24-a are used. There is a characteristic that More specifically, for example, the well-known Prewitt edge detection method or Sobel edge detection method may be used as a differential value detection section. Furthermore, the Prewitt edge detection method, Sobel edge detection method, or Laplacian is usually 3x3
Spatial filter processing is performed using the kernel of
Even if the size of the nucleus is expanded to a size other than 3x3, the essence of the present invention will not be affected. <Effects> As explained above, according to the present invention, it is possible to separate the flat halftone dot portion from the image, so that the halftone dot portion is smoothed, so that moiré caused by the halftone dot and dither pattern can be suppressed. Furthermore, since characters and thin lines are not smoothed, the original image can be faithfully reproduced. Furthermore, it does not affect non-halftone images such as photographic images.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing the basic concept of an embodiment of the present invention;
Figure 2 is a frequency characteristic diagram of various images, Figure 3,
4 and 5 are frequency characteristic diagrams of various differential filters, FIG. 6 is a frequency characteristic diagram of a smoothing filter, FIG. 7 is an image processing block diagram of this embodiment, and FIG. 8 is a control signal generation unit 2. Gamma conversion characteristic diagram, Fig. 9,
FIG. 10 is a diagram showing an operation example of the differential value detection section 1 that performs first-order differentiation, FIG. 11 is a diagram showing an operation example of the smoothing processing section 3, and FIG. 12 is a signal waveform diagram of each section of FIG.
FIG. 13 is a detailed circuit diagram of the edge detection section 1, and FIG.
The figure shows a detailed block diagram of the first-order differentiators 306 and 312, FIG. 15 is a timing chart showing the operation of the first-order differentiator, FIG. 16 is an absolute value circuit diagram, and FIG.
The figure is a detailed block diagram of another example of a first-order differentiator.
8 is a buffer circuit diagram, FIG. 19 is a diagram showing the image area, FIG. 20 is a detailed block diagram of the smoothing processing section 3, FIG. 21 is a diagram showing the operation of the smoothing processing section, and FIG. 22 is a diagram showing the other parts. Fig. 23 is a block diagram for performing the smoothing process of Fig. 22, Fig. 24a, Fig. 24b, and Fig. 24c.
The figure is a diagram showing another gamma conversion characteristic of the control signal generating section 2. In the figure, a is an edge detector, c is a smoother,
d is a mixer, 1 is a differential value detection section, 2 is a control signal generation section, 3 is a smoothing processing section, and 4, 6, and 8 are multipliers, respectively.

Claims (1)

[Scope of Claims] 1. Edge detection means for detecting edge portions of an image signal, smoothing means for smoothing the image signal, and mixing means for mixing the smoothing means and the image signal, An image processing apparatus characterized in that the mixing ratio of the mixing means is varied depending on the output of the edge detection means. 2. An image processing apparatus according to claim 1, wherein the edge detection means is constituted by a first-order differentiator. 3. The image processing apparatus according to claim 2, wherein the edge detection means obtains the sum of first-order differentials in two or more directions.