KR20100096511A - Apparatus and method for unsharp masking removing impulse noise - Google Patents

Abstract

PURPOSE: An unsharp masking apparatus for removing impulse noise and a method thereof are provided to reduce noise and emphasize detailed information such as an edge, thereby providing an image with improved picture quality. CONSTITUTION: A high frequency component extracting unit(110) generates the first image made of a high frequency component extracted from an original image inputted from the outside by a high pass filter. A noise cancellation unit(120) removes noise of the original image by a band pass filter to generate the second image. A picture quality improving unit(130) adds the pixel value of each pixel configuring the first image to the pixel value of each pixel configuring the second image to generate a result image.

Description

Apparatus and method for unsharp masking removing impulse noise}

The present invention relates to an unsharp masking apparatus and method for removing impulse noise, and more particularly, to an apparatus and method for improving image quality by emphasizing high frequency components of an image.

Unsharp masking is a technique of improving image quality by acquiring high-frequency components of an image and adding them to the original image to highlight details such as edges present in the image. Unsharp masking is a technique widely used in various fields of image processing, and various algorithms have been studied.

1 is a diagram illustrating an algorithm of linear unsharp masking (LUM) that is generally used.

Referring to FIG. 1, the high frequency component of the original image x (n) is extracted by a high pass filter (HPF), and the degree to which the high frequency component is emphasized in the image is adjusted according to the weight λ. When the high frequency component emphasized on the original image x (n) is added, the sharpness of the boundary and detail information of the image is increased to obtain a result image y (n) with improved image quality. Since the linear unsharp masking technique linearly recognizes high frequency components, even noise existing in an image is recognized and emphasized as high frequency components. Therefore, a problem may occur in that the image quality is deteriorated by linear unsharp masking.

To solve this problem, permutation weighted median (PWM) is one of the unsharp masking schemes that is robust to noise, and the image enhancement method is robust to random noise. Similar results to unsharp masking, but improved performance in the presence of noise. However, there is a disadvantage in that if there is an impulse noise in the image, it cannot be effectively removed.

In general, the impulse noise present in the image can be removed by a median filter. However, the median filter is not suitable as a preprocessing for unsharp masking because it causes blur distortion of an image like a low pass filter (LPF). Therefore, there is a need for a new unsharp masking technique that can effectively remove impulse noise without generating blur distortion by a median filter, and improve image quality by emphasizing the details of an image.

An object of the present invention is to provide an unsharp masking apparatus and method for improving image quality by emphasizing details while removing impulse noise from an image.

In order to achieve the above technical problem, the apparatus and method for unsharp masking according to the present invention comprises: a high frequency component extracting unit for generating a first image made of a high frequency component extracted from an original image received from an external device by a high pass filter; A noise removing unit for generating a second image by removing noise existing in the original image by a global pass filter; And an image quality improvement unit generating a resultant image by adding pixel values of each pixel constituting the first image and pixel values of each pixel constituting the second image.

According to another aspect of the present invention, there is provided an unsharp masking method, comprising: a high frequency component extracting step of generating a first image including a high frequency component extracted from an original image input from an external device by a high pass filter; A noise removing step of generating a second image by removing noise existing in the original image by a global pass filter; And an image quality improvement step of generating a resultant image by adding a pixel value of each pixel constituting the first image and a pixel value of each pixel constituting the second image.

According to the unsharp masking apparatus and method for removing impulse noise according to the present invention, by applying a combination of a global pass filter for removing impulse noise from an image and a high pass filter for extracting high frequency components of the image, the noise is applied to the original image. Is reduced, and details such as edges are emphasized to obtain an image having improved image quality. In addition, the complex response process can be reduced by constructing a global pass filter and a high pass filter robust to noise by adjusting the frequency response by a linear combination of a plurality of weighted median filters.

Hereinafter, exemplary embodiments of an unsharp masking apparatus and method for removing impulse noise according to the present invention will be described in detail with reference to the accompanying drawings.

Figure 2 is a block diagram showing the configuration of a preferred embodiment of the unsharp masking device according to the present invention.

Referring to FIG. 2, the unsharp masking apparatus according to the present invention includes a high frequency component extracting unit 110, a noise removing unit 120, and an image quality improving unit 130.

The high frequency component extractor 110 generates a first image including high frequency components extracted from the original image by the high pass filter.

As described above, when unsharp masking is performed to improve image quality, the noise of the original image is mistaken for a high frequency component and amplified. there is a problem. In order to solve this problem, the high frequency component extraction unit 110 designs a high pass filter that is robust to noise by combining a plurality of weighted median filters. The weighted median filter has a characteristic of removing impulse noise while preserving a boundary area which is detailed information in an image by setting weights. Therefore, the first image generated by the high frequency component extractor 110 becomes an image unaffected by noise.

3 is a diagram illustrating the principle of a weighted median filter. Referring to FIG. 3, different weights are assigned to pixel values of each pixel of a window filter applied to an image. That is, the pixel value of the pixel of the filter corresponding to the position of the pixel to which the filter is to be applied in the original image is given a greater weight than the surrounding pixels. In the case of FIG. 3, the weight given to the pixel located in the middle of the filter is set to '3'. This means that there are three pixels having a pixel value of '9' in the original image. Therefore, if the pixel values of the pixels constituting the original image are aligned, the median value is '9'. On the other hand, when the median filter with no weight is applied, the median value found by aligning pixel values of pixels constituting the original image is '8'.

The weighted median filter may have various frequency response characteristics according to the weighted weight, and various filters may be configured by designing a desired finite impulse response (FIR) filter coefficient by linearly combining a plurality of weighted median filters. . In this case, the FIR filter refers to a filter having a finite length of an impulse function, which is a characteristic function of the filter by performing filtering with only certain values of an input signal. As described above, a filter in which a plurality of weighted median filters are linearly combined is called a linear combination of weighted median (LCWM) filter.

4 is a diagram illustrating a configuration of an LCWM in which a plurality of weighted median filters are linearly combined. Referring to FIG. 4, x (n) is a pixel value of a pixel constituting the original image, y _LCWM (n) is a pixel value of a pixel constituting the image obtained by applying a filter to the original image, and WM ₁ to WM _k are k weighted median filter coefficients and α ₁ to α _k represent coefficients multiplied by the weighted median filter coefficients of WM ₁ to WM _k , respectively.

5 shows an example of a weighted median filter mask used for an LCWM when the size of the LCWM is 5 × 5. Areas indicated by dots in FIG. 5 are pixels having a weight set to 1, and indicate positions of pixels referred to when designing an LCWM. The pixels in the remaining areas are set to zero weight and are not used in the design of the LCWM. Six WM _i are obtained for six filter masks for pixels located in the center of a 5x5 size image, and alpha _i is multiplied for each and linearly combined. The designed LCWM has a large frequency characteristic depending on the value of α _i . For example, if the α _i value is set to [1.403, 0.941, 0.033, 0.005, 0.519, -1.902], the LCWM operates as a low pass filter with a cutoff frequency of 0.4.

Since the LCWM used by the high frequency component extractor 110 should show the same frequency characteristics as the high pass filter, an α _i value corresponding thereto is used. For example, when using the six filter masks shown in FIG. 5, the α _i value may be set to a value of [1.5034, 1.8092, -0.0173, -0.0004, 2.501, -5.7958], respectively. However, this value is an optimal value set for the LCWM to operate as the high pass filter by the high frequency component extraction unit 110, and other values for allowing the LCWM to show the frequency characteristics of the high pass filter may be set to the α _i value. have.

When the six filter masks shown in FIG. 5 are used as WM _i and the value α _i , which is a coefficient multiplied by each WM _i , is set to [1.5034, 1.8092, -0.0173, -0.0004, 2.501, -5.7958], the final The resulting LCWM operates as a high pass filter with the following filter coefficients.

The high frequency component extraction unit 110 generates the first image composed of the high frequency components extracted from the original image by designing the LCWM as a high pass filter by the above method. The high pass filter designed by the LCWM is robust to noise and is not affected by the impulse noise, and as shown in FIG. 1, the noise is not amplified even by increasing the λ value for adjusting the emphasis level of the image details.

The noise removing unit 120 generates a first image by removing noise existing in the original image received from the outside by the global pass filter.

The weighted median filter used by the noise canceller 120 to construct the global pass filter is the same as the weighted median filter used by the high frequency component extractor 110 to construct the high pass filter. The LCWM obtained by multiplying the filter coefficients of the weighted median filter of WM ₁ to WM _k by α ₁ to α _k, which are coefficients for determining the frequency response of the filter, and summing all of them, acts as a global pass filter. In this case, α _i is appropriately set to a value that allows the LCWM to show the frequency response of the global pass filter.

For example, six filter masks shown in FIG. 5 are used for the design of LCWM, and the coefficient α _i value multiplied by each WM _i is finally set to [1.4966, 1.1908, 0.0173, 0.0004, 0.499, -2.2042]. The LCWM is designed to operate as a global pass filter with the following filter coefficients.

The noise removing unit 120 generates a second image by removing the impulse noise existing in the original image by the designed global pass filter.

Finally, the image quality improvement unit 130 generates a result image by adding pixel values of each pixel constituting the first image and pixel values of each pixel constituting the second image.

6 is a diagram illustrating an algorithm for generating a resultant image having improved image quality from an original image by the noise removing unit 110, the high frequency component extracting unit 120, and the image quality improving unit 130. Referring to FIG. 6, x (n) represents an original image, and y (n) represents a result image having improved image quality. First, the noise removing unit 110 generates a first image by applying a high pass filter to the original image, and the high frequency component extractor 120 generates a second image by applying a noise removing filter that is a global pass filter to the original image. do. Next, the image quality improvement unit 130 generates a result image by adding pixel values of pixels constituting the first image and the second image, respectively. In this case, λ is a coefficient for adjusting the degree to which the detail information of the image is emphasized by the high pass filter, and is preferably set to 1. Looking at the resulting image y (n), it can be seen that the noise of the image is reduced and the edges are emphasized.

The first image generated by the high frequency component extractor 110 is an image including detailed information such as an edge extracted from the original image. In addition, the second image generated by the global pass filter configured by the noise removing unit 120 is an image in which impulse noise is removed from the original image. Therefore, the resultant image generated by combining the first image and the second image effectively removes the noise, and the detailed information of the original image is highlighted. Accordingly, an image having an improved image quality compared to an image obtained by the conventional unsharp masking method may be obtained.

7 is a flowchart illustrating a preferred embodiment of the unsharp masking method for removing impulse noise according to the present invention.

Referring to FIG. 7, first, the high frequency component extractor 110 generates a first image including high frequency components extracted from an original image input from the outside by a high pass filter robust to noise (S710). Next, the noise removing unit 120 generates a second image by removing noise existing in the original image by the global pass filter (S620). In this case, LCWM designed by linearly combining a plurality of weighted median filters may be used as the high pass filter and the global pass filter. The LCWM determines the frequency response of the filter by adjusting the value of the coefficient that is multiplied when combining the plurality of weighted median filters.

Finally, the image quality improvement unit 130 generates a result image by adding pixel values of each pixel constituting the first image and pixel values of each pixel constituting the second image (S730). The generated result image removes the impulse noise existing in the original image, and the detail information of the original image is emphasized, thus showing an improved image quality compared to the original image.

Hereinafter, the experimental results performed to evaluate the performance of the unsharp masking apparatus and method for removing impulse noise according to the present invention will be described. The experiment was performed by MATLAB, and an 8-bit black and white image was used as the original image.

First, the result of applying linear linear filtering to the result of applying LCWM to the original image with noise is compared.

8A to 8D are diagrams showing an example of comparing the performance of the present invention and conventional linear unsharp masking. 8A is an image without noise, and FIG. 8B is an original image in which noise occurs in the image of FIG. 8A. Figure 8c shows the result of applying the low pass filter, high pass filter and linear unsharp masking to the original image, respectively, Figure 8d shows the low pass filter, high pass filter designed by LCWM to the original image and the unsharp masking according to the present invention The result of applying each apparatus is shown. When comparing the first image of FIG. 8C with the first image of FIG. 8D, when the existing low pass filter is applied to the original image, the noise existing in the original image is not completely removed, but the LCWM is designed as the low pass filter and applied to the original image. You can see that all the noise of the original image is removed. In addition, when comparing the second image of FIG. 8C with the second image of FIG. 8D, when the existing high pass filter is applied to the original image, noise existing in the original image is also emphasized, so that the detailed information of the original image cannot be recognized, but the LCWM is a high pass filter. In case of designing and applying to the original image, only the details of the original image except the noise were highlighted. 8C and 8D show the results of applying linear unsharp masking and unsharp masking by LCWM according to the present invention, respectively. In the case of unsharp masking by LCWM, the detail of the original image is emphasized while noise is emphasized. The result shows that the removal performance is improved compared to the conventional linear unsharp masking.

9A to 9D are diagrams illustrating another example in which the present invention is compared with the performance of the conventional linear unsharp masking. 9A is an image without noise, and FIG. 9B is an original image in which noise occurs in the image of FIG. 9A. Figure 9c shows the results of applying a low pass filter, a high pass filter and a linear unsharp masking to the original image, respectively, Figure 9d shows a low pass filter, a high pass filter designed by LCWM to the original image and the unsharp masking according to the present invention The result of applying each apparatus is shown. As in the case of FIGS. 8A to 8D, when the unsharp masking by LCWM is used, a resultant image having an improved image quality compared to the conventional linear unsharp masking can be obtained.

10A to 10D and 11A to 11D are diagrams illustrating still another example in which the present invention is compared with the performance of the conventional linear unsharp masking. 10C and 10D, and FIGS. 11C and 11D show the same results as FIGS. 8C and 8D, and the unsharp masking by LCWM has better image quality improvement performance than the conventional linear unsharp masking.

In addition, the present invention was evaluated in comparison with the conventional linear unsharp masking (hereinafter referred to as LUM) method and permutation weighted median (hereinafter referred to as PWM).

FIG. 12 is a diagram illustrating a result of applying LUM, PWM, and unsharp masking according to the present invention to an image having no noise. When there is no noise in the image, it can be seen that all three methods produce result images of almost the same quality.

13A to 13F illustrate a result of applying a LUM to an original image and an original image having 5% impulse noise, respectively, a result of applying a 3 × 3 median filter and a LUM to the original image, and applying a PWM to the original image. Figure 3 shows the result of applying a 3x3 median filter and PWM to and the result of applying unsharp masking by LCWM.

If LUM and PWM are applied while ignoring the impulse noise existing in the original image, noise is emphasized as shown in FIGS. 13B and 13D. When the 3x3 median filter is applied to remove the impulse noise, as shown in FIGS. 13C and 13E, the impulse noise is removed, but the median filter acts like a lowpass filter, resulting in blur distortion in the image. However, if unsharp masking by LCWM is applied to the original image, it is possible to obtain an image having improved image quality compared to the conventional methods by removing the impulse noise of the image and preserving details such as edges as shown in FIG. 13F. . Comparing the box portions shown in FIGS. 13C, 13E, and 13F, it can be seen that the detailed information of the image is more clearly preserved when unsharp masking by LCWM is applied.

14A to 14F illustrate a result of applying a 3 × 3 median filter and a LUM to an original image and an original image having 12% impulse noise, respectively, as a result of applying a 5 × 5 median filter and a LUM to the original image, The result of applying the 3 × 3 median filter and PWM, the result of applying the 5 × 5 median filter and PWM to the original image, and the result of applying unsharp masking by LCWM.

Since the impulse noise is more severe than in the case of FIG. 13, the noise is not removed by the 3 × 3 median filter. Therefore, when the LUM and the PWM are applied, the noise is amplified as shown in FIGS. 14B and 14D and the image quality of the image is degraded. However, if the 5 × 5 median filter is used to remove all the impulse noise, as shown in FIGS. 14C and 14E, a lot of detail information of the image is lost, resulting in severe image degradation. However, when unsharp masking by LCWM is applied, the image quality improvement is excellent even for images with high impulse noise.

Although the preferred embodiments of the present invention have been shown and described above, the present invention is not limited to the specific preferred embodiments described above, and the present invention belongs to the present invention without departing from the gist of the present invention as claimed in the claims. Various modifications can be made by those skilled in the art, and such changes are within the scope of the claims.

1 is a diagram illustrating an algorithm of linear unsharp masking (LUM) commonly used.

Figure 2 is a block diagram showing the configuration of a preferred embodiment of the unsharp masking device according to the present invention,

3 shows the principle of a weighted median filter;

4 is a diagram illustrating a configuration of an LCWM in which a plurality of weighted median filters are linearly combined;

5 shows an example of a weighted median filter mask used for LCWM when the size of LCWM is 5 × 5;

6 is a diagram illustrating an algorithm for generating a resultant image having improved image quality from an original image by the high frequency component extractor 110, the noise remover 120, and the image quality improver 130.

7 is a flowchart illustrating a preferred embodiment of the unsharp masking method for removing impulse noise according to the present invention;

8a to 8d is a view showing an example comparing the performance of the present invention and the conventional linear unsharp masking,

9a to 9d are diagrams showing another example comparing the performance of the present invention and conventional linear unsharp masking;

10a to 10d is a view showing another example comparing the performance of the present invention and the conventional linear unsharp masking,

11a to 11d is a view showing another example comparing the performance of the present invention and the conventional linear unsharp masking,

12 is a diagram illustrating a result of applying LUM, PWM, and unsharp masking according to the present invention for an image without noise;

13A to 13F illustrate a result of applying a LUM to an original image and an original image having 5% impulse noise, respectively, a result of applying a 3 × 3 median filter and a LUM to the original image, and applying a PWM to the original image. Figure 3 shows the result of applying a 3 × 3 median filter and PWM to the unsharp masking by LCWM, and

14A to 14F illustrate a result of applying a 3 × 3 median filter and a LUM to an original image and an original image having 12% impulse noise, respectively, as a result of applying a 5 × 5 median filter and a LUM to the original image, The result of applying the 3 × 3 median filter and PWM, the result of applying the 5 × 5 median filter and PWM to the original image, and the result of applying unsharp masking by LCWM.

Claims (6)

A high frequency component extracting unit configured to generate a first image including high frequency components extracted from an original image received from an external device by a high pass filter; A noise removing unit for generating a second image by removing noise existing in the original image by a global pass filter; And And an image quality improvement unit generating a resultant image by adding pixel values of each pixel constituting the first image and pixel values of each pixel constituting the second image. The method of claim 1, And the high pass filter comprises a linear combination of a plurality of weighted median filters. The method according to claim 1 or 2, The all-pass filter is an unsharp masking device, characterized in that configured by linearly combining a plurality of weighted median filter. A high frequency component extracting step of generating a first image made of a high frequency component extracted from an original image received from the outside by a high pass filter; A noise removing step of generating a second image by removing noise existing in the original image by a global pass filter; And And an image quality improvement step of adding a pixel value of each pixel constituting the first image and a pixel value of each pixel constituting the second image to generate a resultant image. The method of claim 4, wherein And the high pass filter comprises a linear combination of a plurality of weighted median filters. The method according to claim 4 or 5, The all-pass filter is an unsharp masking method comprising a linear combination of a plurality of weighted median filters.

Images