JPH10187959A - Image processor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image processor which can eliminate the impulse noises without impairing the form features of an image. SOLUTION: An impulse noise detector 10 learns via a learning image to make its output approximate to the 1st and 2nd value when the processing object pixel is identical and not identical to an impulse noise respectively based on the data calculated from the object pixel and its peripheral pixels. Then the detector 10 outputs an impulse noise detection signal based on its learning result. An impulse noise eliminator 20 defines plural peripheral pixels of the pixel under processing as the reference points when an impulse noise exists. Then the eliminator 20 calculates the input signal value that undergoes the fuzzy control based on the signal value of the reference points and the prescribed signal value included in a filter window and then calculates the value of pixel data serving as an output based on the value obtained by applying the fuzzy control to be learnt to the input signal value based on the learning image.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to an image processing apparatus, and more particularly to an image processing apparatus for restoring an original image by removing noise from two-dimensional image data on which impulse noise is superimposed by digital signal processing.

[0002]

2. Description of the Related Art Hitherto, it has been known to use a median filter as an image processing apparatus for removing noise from two-dimensional image data on which impulse noise is superimposed. This median filter is a filter that sequentially processes all input signals, rearranges the processed input data sequence in ascending order or descending order, and outputs the median value in a filter window of a certain size. Can remove impulse noise.

[0003] However, since this median filter also deforms small changes in the signal, it has been proposed to remove impulse noise using a center weighted median (CWM) filter. . The CWM filter is an extended version of the median filter. The CWM filter assigns weights to the processing points and outputs the median value when the processing points are arranged in ascending or descending order.

Further, as a conventional image processing apparatus, an apparatus using a median filter based on a fuzzy rule has been proposed (Transactions of the Institute of Electronics, Information and Communication Engineers A Vol. J78-AN).
o.2, pp.123-131, February 1995). The median filter based on this fuzzy rule inputs the input image data to the detector through the median filter and directly to the detector, and determines the presence or absence of impulse noise in the input image data based on the fuzzy rule. Is determined to be present, the output of the median filter is used as output image data, and when it is determined that no impulsive noise is present, the input image data is output as it is.

[0005]

However, all of the above-mentioned conventional image processing apparatuses tend to process the entire image, remove impulse noise, and simultaneously degrade the original signal. In addition, portions that are characteristic in shape of the image, that is, edges and lines composed of one pixel are not taken into consideration, so that they are deteriorated.

[0006] The present invention has been made in view of the above points,
It is an object of the present invention to provide an image processing apparatus capable of removing impulsive noise without impairing the shape characteristics of an image.

[0007]

SUMMARY OF THE INVENTION In order to achieve the above object, the present invention has been made to solve the above problem.
When receiving two-dimensional image data as an input signal and sequentially processing the data of all pixels, the pixel to be processed is impulsive noise based on data calculated from the pixel to be processed and a plurality of surrounding pixels. In some cases, learning is performed using a learning image so that the output when the noise is not impulsive noise approaches the second value, and the output differs depending on the presence or absence of impulsive noise based on the learning result. An impulse noise detector that outputs a logical impulse noise detection signal, and two-dimensional image data of an original image and an impulse noise detection signal are received as input signals, and the impulse noise detection signal indicates that there is no impulse noise. Output, the input image data of the corresponding pixel is output as it is, and when impulse noise is present, a plurality of pixels around the pixel currently being processed are output. Using the element as a reference point, an input signal value to be fuzzy controlled is calculated from the signal value of the reference point and a predetermined signal value in the filter window, and learning is performed on the input signal value based on a learning image. And an impulse noise eliminator for calculating a value of pixel data to be output from a value obtained by performing the fuzzy control.

According to the present invention, fuzzy control is provided in which two-dimensional image data is prepared in consideration of the shape characteristics of an image, and the image data is optimally learned by a fuzzy rule using a learning image in an impulse noise detector. And when the pixel to be processed is impulsive noise,
Learning is performed using an image for learning so that the output when the noise is not the impulse noise approaches the second value, thereby outputting an impulse noise detection signal. When the impulse noise detection signal indicates that there is impulsive noise, the impulse noise remover sets a plurality of pixels around the pixel currently being processed as a reference point, and a signal value of the reference point and a filter window. The fuzzy control is performed on the input signal value calculated from the predetermined signal value in the above, and the obtained values are made to compete with each other, and the output signal is obtained based on the signal value of the winner. As a result, features in the shape of the image can be preserved, and a visually good image can be obtained.

[0009]

Next, an embodiment of the present invention will be described.

FIG. 1 is a block diagram showing an embodiment of an image processing apparatus according to the present invention. In this embodiment, an impulse noise detector 10 to which two-dimensional image data as an original image is inputted, and two-dimensional image data as an original image and an output detection signal of the impulse noise detector 10 are inputted,
Impulse noise remover 20 for outputting processed image data
It is composed of

As shown in the block diagram of FIG. 2, the impulse noise detector 10 includes an input generator 11 to a fuzzy controller, a fuzzy controller 12 to which an output signal of the input generator 11 is input, and a fuzzy controller 12. An output signal of the controller 12 is input, and is constituted by a binarizer 13 for binarizing the input signal.

The impulse noise eliminator 20 shown in FIG.
As shown in the block diagram of FIG. 3, an input generator 21 to which an output detection signal of the impulse noise detector 10 and two-dimensional image data of an original image are input, and an output signal of the input generator 21 to be input. Fuzzy controller 22 and a competitor 2 which receives the output signal of fuzzy controller 22 and performs a competition process
3 and an output image generator 24 for generating an output image from the output signal of the competitor 23.

Next, the operation of this embodiment will be described.

First, when image data on which impulse noise is superimposed is input to the impulse noise detector 10 shown in FIGS. 1 and 2, the impulse noise detector 10 performs an operation according to the flowchart of FIG. , The presence or absence of impulsive noise is detected. That is, the impulse noise detector 10 receives the image data (step 10).
1) The input image data is processed for each pixel, and it is determined whether or not it is impulsive noise. When all the pixels have been detected, the processing of the impulse noise detector 10 is completed. I do. It is identified as follows:

Assuming that a currently selected pixel on a two-dimensional image is x (i, j), eight pixels around it, x (i-1, j-1) and x (i, j), -1), x
(I + 1, j-1), x (i-1, j), x (i + 1,
j), x (i-1, j + 1), x (i, j + 1), x
Considering (i + 1, j + 1), the fuzzy controller 1 of FIG.
2 input data strings inpk _k (where k = 0)
To 5) are calculated based on the following equation.

[0016]

(Equation 1) Here, x _n1 and x _n2 indicate the signal value closest to x (i, j) and the signal value next to it, respectively, in the filter window under consideration, and med (i, j) indicates the signal value in the filter window. The median value of

The above input data sequence inpk _k is input to the fuzzy controller 12. Fuzzy controller 12, the input data sequence Inpk _k, optimally learning computes the output gr based on the signal for learning (step 103). Here, since gr = 1 when there is impulsive noise and gr = 0 when there is no impulsive noise, the intermediate value "0.5" is set as the threshold value, and the value gr is set to the threshold value. It is determined whether the value is greater than “0.5” (step 1).
04), if large, the pixel (i, j) is determined to be impulsive noise (step 105); if not large, the pixel (i, j) is determined to be the original signal instead of impulsive noise (step 106). ), And stores the determination result in an internal memory (step 107).

Then, the fuzzy controller 12 performs the processing of steps 101 to 107 for all pixels (steps 101 to 109). The data output from the fuzzy controller 12 is supplied to the binarizer 13 shown in FIG.
Alternatively, it is binarized to "1", output as impulsive noise data (impulse noise detection signal) (step 110), and stored in the memory.

The impulse noise data detected by the impulse noise detector 10 as described above is input to the impulse noise remover 20 shown in FIGS. 1 and 3 together with the pixel data of the original image to remove the noise. Is done. next,
FIG. 5 shows the operation of the impulse noise eliminator 20.
It will be described together with the flowchart of FIG.

In the impulse noise eliminator 20,
Processing is performed for each pixel to remove impulse noise.
The input generator 21 of FIG.
Receiving as input signals the impulse noise data obtained in step 2 and the two-dimensional image data of the original image (step 201);
It is determined whether or not the impulse noise data corresponding to the pixel of the currently processed two-dimensional image data is “0”, that is, whether or not there is any impulse noise (step 20).
2) If "0", the corresponding pixel is not impulsive noise, so the image data of the input pixel is output as it is as output image data (the output is equivalent to the input) and the process proceeds to the next pixel (step) 210).

On the other hand, if the impulse noise data corresponding to the pixel currently being processed is "1", the image data corresponding to the pixel is input (step 203) to remove the impulse noise. Pixel x currently being processed
Eight pixels x (i-1, j-1) around (i, j),
x (i, j-1), x (i + 1, j-1), x (i-
1, j), x (i + 1, j), x (i-1, j + 1),
x (i, j + 1) , x (i + 1, j + 1) to calculate the input Inpj _k to fuzzy controller 22 which is a reference point, respectively (step 204). The reference point is x (i-1, j-
In the case of 1), the input to the fuzzy controller 22 is as follows.

Inpj₀= | X_m1-X_m2｜ impj₁= | X_m1-X (I-1, j-1) | impj_Two= | X_m2-X (I-1, j-1) | impj_Three= | X_m3-X (I-1, j-1) | where x_m1Is the city near the reference point in the filter window
Near the signal value of the reference point of the two points where the street distance is 1
Signal value, x_m2, X_m3Is 2 where the city distance is 1
The point within the filter window excluding the point is closest to the signal value of the reference point.
Signal value and the next closest signal value. In addition, the city area distance
Is when there are two points (x1, y1) and (x2, y2)
| X1-x2 | + | y1-y2 |.

That is, when the reference point is x _standard , the input to the fuzzy controller 22 is as follows.

Impj ₀ = | x _m1 -x _m2 | inpj ₁ = | x _m1 -x _standard | impj ₂ = | x _m2 -x _standard | impj ₃ = | x _m3 -x _standard | Fuzzy controller 2
2 calculates the output by learning the above input optimally using the image for learning (step 205). The output of the fuzzy controller 22 obtained by this calculation has a value close to "1" when the average of the value of the reference point x _standard and x _m2 is considered to be the output, and otherwise a value close to "0".

A total of eight outputs from the fuzzy controller 22 using the eight pixels around the pixel x (i, j) for which the impulse noise detection signal indicates the presence of the impulse noise as reference points x _standard are as follows: is supplied to the contention unit 23 in FIG. 3, where competition operation among average of the reference point and x _{m @ 2} of the fuzzy controller 22, the largest value is selected and output (step 206).

Next, the output image generator 24 of FIG. 3 determines whether or not the value selected by this competition operation (this is referred to as a winner) is larger than a predetermined reference value (step 207). If the winner is greater than the reference value, an output pixel data value is calculated from the signal value of the reference point (step 208). If the winner does not exceed the reference value, the output is calculated as the median value of the pixel value in the filter window. (Step 209).

Subsequently, the process proceeds to the next pixel, and it is determined whether or not these operations have been completed for all the pixels (steps 210 and 211), and the processing of steps 201 to 210 is executed for all the pixels. The output obtained is the output of the image processing apparatus. Then, the pixel data output from the output image generator 24 is displayed as an image on the display (step 212). This makes it possible to remove impulse noise without visually deteriorating the two-dimensional image.

[0028]

Next, an example of the above embodiment will be described. A character T as shown in FIG. 6 is used as an image for simulation, and impulse noise is superimposed on 5% of the pixels of the entire image (that is, the impulse noise occurrence rate is 5%). ± 2 as noise
FIG. 7 shows a deteriorated image when 56 is superimposed.

The result of processing the two-dimensional image data of the deteriorated image shown in FIG. 7 in the embodiment shown in FIG. 1 is based on the mean square error (MSE) between the original image and one of the image evaluation criteria. : Mean Square Error) was “0.387”, and the obtained image was as shown in FIG. In contrast, the two-dimensional image data of the deteriorated image shown in FIG.
Is "84.8", the obtained image is as shown in FIG. 9, and the MSE of the image data obtained through the conventional CWM filter is "7.482".
As shown in FIG. 11, the MSE as a result of further processing by the median filter based on the conventional fuzzy rule is “3.906”, and the obtained image is as shown in FIG.

In general, the smaller the value of MSE, the higher the quality of the output image. As can be seen from the above value of MSE, the processing according to the embodiment of the present invention has the most effective effect of removing impulse noise. In addition, it was confirmed that the image could be restored closest to the original image.

The image "gir" of a woman shown in FIG.
l ″ is used as an image for simulation, and in this case as well, FIG. 13 shows a degraded image when the impulse noise occurrence rate is 5% and the value ± 256 is superimposed as the impulse noise.

When the two-dimensional image data of the deteriorated image shown in FIG. 13 is processed by the embodiment shown in FIG.
SE was "2.369", and the obtained image was as shown in FIG. On the other hand, the MSE of the image data obtained by passing the two-dimensional image data of the deteriorated image shown in FIG. 13 through the conventional median filter is “28.78”, and the obtained image is as shown in FIG. MSE of the image data obtained through the CWM filter of “12.
76 ", the image at that time is as shown in FIG.
Further, the MSE of the result of processing by the median filter based on the conventional fuzzy rule was “6.109”, and the obtained image was as shown in FIG. That is, it has been confirmed that, in the case of this "girl", the present embodiment can remove the most impulsive noise without impairing the shape characteristics of the image.

[0033]

As described above, according to the present invention,
Eliminates impulse noise without impairing the shape characteristics of images such as edges and lines composed of one pixel,
The original image can be best restored.

[Brief description of the drawings]

FIG. 1 is a block diagram of an embodiment of an image processing apparatus according to the present invention.

FIG. 2 is a block diagram of an example of an impulse noise detector in FIG.

FIG. 3 is a block diagram of an example of an impulse noise remover in FIG. 1;

FIG. 4 is a flowchart for explaining the operation of the impulse noise detector of FIG. 2;

FIG. 5 is a flowchart for explaining the operation of the impulse noise eliminator of FIG. 3;

FIG. 6 is an example of an original image for simulation.

FIG. 7 shows an impulse noise sound generation rate of 5 in the image of FIG.
%, And a deteriorated image superimposed with a size of ± 256.

FIG. 8 is an image obtained by processing the image of FIG. 7 according to the embodiment of the present invention;

FIG. 9 is an image obtained as a result of processing the median filter on the image of FIG. 7;

FIG. 10 is an image obtained as a result of processing by the CWM filter on the image of FIG. 7;

11 is an image obtained as a result of processing a median filter based on a fuzzy rule for the image of FIG. 7;

FIG. 12 is another example of an original image for simulation.

FIG. 13 shows an impulse noise generation rate of 5 in the image of FIG.
%, A weighted image having a size of ± 256.

14 is an image obtained by processing the image of FIG. 13 according to the embodiment of the present invention;

15 is an image obtained as a result of processing the median filter on the image of FIG. 13;

FIG. 16 is an image obtained as a result of processing by the CWM filter on the image of FIG. 13;

17 is an image obtained as a result of processing a median filter based on a fuzzy rule for the image of FIG. 13;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 10 Impulse noise detector 11 Input generator to fuzzy controller 12 Fuzzy controller 13 Binarizer 20 Impulse noise remover 21 Input generator 22 Fuzzy controller 23 Competitor 24 Output image generator

Claims (3)

[Claims] When receiving two-dimensional image data of an original image on which an impulse noise is superimposed as an input signal and sequentially processing data of all pixels, a plurality of pixels to be processed and a plurality of surrounding pixels are used. Based on the calculated data, learning is performed using a learning image so that the output of the pixel to be processed is close to the first value when the pixel to be processed is impulsive noise and the second value when the pixel is not impulsive noise. An impulse noise detector that outputs an impulse noise detection signal having a different logical value depending on the presence or absence of the impulse noise based on the learning result; two-dimensional image data of the original image and the impulse noise detection signal When the impulse noise detection signal indicates that there is no impulse noise, the input image data of the corresponding pixel is output as it is, When noise is present, a plurality of pixels around the pixel currently being processed are used as reference points, and an input that is fuzzy-controlled based on the signal value of the reference points and a predetermined signal value in the filter window. An impulse noise eliminator that calculates a signal value and performs a fuzzy control for learning the input signal value based on a learning image to calculate a value of pixel data to be output from a value obtained. An image processing apparatus comprising: 2. The impulse noise detector receives, as an input signal, two-dimensional image data on which the impulse noise is superimposed, and sequentially processes data of all pixels to determine a pixel to be processed. x (i, j), surrounding pixels x (i-1, j-1), x (i, j)
-1) x (i + 1, j-1), x (i-1, j), x
(I + 1, j), x (i-1, j + 1), x (i, j +
1), x (i + 1, j + 1), six input values are generated from two values whose signal values are close to the pixel (x, j) to be processed in the filter window, and a median value in the filter window An input generator to the fuzzy controller, and a first value when the pixel to be processed is impulsive noise, using the six input values generated by the input generator. A fuzzy controller for learning using an image for learning so that the output at the time approaches a second value, and a binarizer for setting the output of the fuzzy controller to "0" or "1" and storing the output 2. The image processing apparatus according to claim 1, comprising: 3. The impulse noise remover receives two-dimensional image data of the original image and the impulse noise detection signal as input signals, and the impulse noise detection signal indicates that there is no impulse noise. Means for outputting the input image data of the corresponding pixel as it is; and when the impulse noise detection signal indicates the presence of impulse noise, the pixel x (i,
Assuming that a plurality of pixels around j) are a plurality of reference points x _standard , the following equation is obtained: inpj ₀ = | x _m1 −x _m2 | inpj ₁ = | x _m1 −x _standard | inpj ₂ = | x _m2 −x _standard | inpj ₃ = | x _m3 −x _standard | (where x _m1 is a signal value whose value is close to the signal value of the reference point among two points in the filter window near the reference point and whose city distance is 1; x _m2 and x _m3 are input generations that calculate a plurality of input values based on the signal value closest to the reference point and the signal value closest to the reference point among the points in the filter window excluding two points where the city distance is 1. For each of the plurality of input values from the input generator, the final output is determined based on the value of the reference point of the input value and the most signal value at a point in the filter window near a city distance not equal to 1. The first value if it is the average of the signal values near the point, otherwise A fuzzy controller that learns based on a learning image so as to output a value close to a second value, a competing device that competes a plurality of outputs from the fuzzy controller and selects a maximum value, An output image generator that calculates an output when an output value of the fuzzy controller selected by the competitor exceeds a predetermined reference value, and outputs a median value in the filter window when the output value does not exceed a predetermined reference value; 2. The method according to claim 1, further comprising:
The image processing apparatus according to any one of the preceding claims.