KR101753365B1 - Method and apparatus for reducing noise of image using non local means - Google Patents

Abstract

The present invention relates to an image noise reduction method and apparatus capable of effectively removing noise of an image as well as enhancing noise removal performance of an image using a NL means (Non-Local means) The method includes the steps of transforming a noise-containing image into a Discrete Cosine Transform (DCT) region, selecting a coefficient in a block of the discrete cosine transformed region, Calculating a correction value by applying a non-local means method, and applying the calculated correction value to a spatial region.

Description

Technical Field [0001] The present invention relates to a method and apparatus for efficiently removing image noise using non-local means,

[0001] The present invention relates to a method and apparatus for eliminating image noise, and more particularly, to a method and apparatus for eliminating image noise using NL means (non-local means) And more particularly, to a method and apparatus for eliminating image noise.

Various methods have been applied to remove irregular noise when added to the image.

Generally, in removing noise of an image, after detecting the degree of noise of an image, if the noise is determined to be high through the detected noise level, the image is strongly filtered. If it is determined that the noise is small, .

The stronger filtering of the image means that the noise of the pixel is removed by using a large number of peripheral pixels of the pixel to be removed. On the contrary, the weakening of the filtering of the image means that the surrounding of the pixel It means that the noise of the pixel is removed using fewer pixels.

In order to remove the noise of an image, a method of removing image noise using a low-pass filter and a median filter, a method of removing an image noise using a wavelet filter, (bilateral filter).

The image noise cancellation method using the low pass filter is the most classical method, and it is possible to remove the noise of the image, but also the high frequency component of the image is lost together, and blur is likely to occur. In addition, the use of the median filter can reduce the blur that occurs when using the low-pass filter to some extent, but it has a disadvantage that it does not work well against the cluster noise.

The Binary Lateral filter eliminates image noise by smoothing the image while preserving the edge component of the image by obtaining the output value using the weight considering the similarity between the neighboring pixels as well as the distance between the neighboring pixels.

As a noise reduction method of the transform domain, a wavelet filter method is mainly used. By dividing the image into subbands using a wavelet filter and analyzing the characteristics of the image, a high frequency band signal is applied to the characteristics of the image Therefore, the image noise is reduced by reducing the noise of the image. Since the noise of the image mostly includes the high frequency component and the degree of noise included in each subband is different, the characteristic of each image signal is statistically analyzed for each subband, By appropriately reducing the high-frequency signal according to the local characteristics, it is possible to effectively reduce the noise of the image, and the high-frequency component can be well preserved compared to the conventional low-pass filter, It is difficult to accurately analyze the statistical characteristics of each video signal.

On the other hand, the NL means method measures the similarity of a local portion of an image input to a computer device and calculates a weight according to the similarity of the local characteristics, thereby obtaining a weighted sum, thereby efficiently removing noise This method removes image noise by using the following relation.

[Formula 1]

을 만족하는 값을 가진다.Here, v (i) and v (j) represent the brightness of the pixel i and the pixel j, and the weight w (i, j) represents the similarity between the pixel i and the pixel j, 1 and

Lt; / RTI >

In the above equation, it is ideal to obtain the weight for the entire image, but the calculation amount is reduced by calculating the weight for the pixels within a certain range in consideration of the calculation amount. This weight can be obtained by using the similarity between the brightness vectors v (N _i ) and v (N _j ) of the pixel vector composed of the surrounding pixels of the two pixels i and j,

[Formula 2]

In this case, Z (i) represents a coefficient for normalization and h is a value which is 4 to 10 times the standard deviation of noise.

[Formula 3]

In the NL means method, instead of comparing only the brightness value of one pixel, the geometric shape of neighboring pixels is compared and the weight is calculated according to the similarity, so that the noise can be removed while maintaining the geometric shape of the image.

In the NL means method, a patch-based NL means method for calculating weights for several pixels at once can be expressed as the following relation.

[Formula 4]

The weight w (P (i), P (j)) represents the brightness of the patch P (i), where P (i) And P (j), and using the brightness information of neighboring pixels as in the pixel-by-pixel NL means method, noise can be removed in units of patches.

Although the image noise cancellation technique using the NL means described above is superior in performance, it requires a large amount of computation to calculate the local similarity.

In addition, in relation to a technique for eliminating noise in an image, Japanese Unexamined Patent Application Publication No. 10-2011-0068645 relates to a block-based image noise cancellation method and apparatus, A similarity degree between reference blocks in a search range for a block and a current block is determined and a weight for a reference block for the current block is determined based on the similarity to remove noise in the current block.

However, since the similarity between a plurality of reference blocks and a current block needs to be calculated in the above conventional art, there is a problem that the amount of calculation may increase when noise is removed from the entire image.

Published Patent No. 10-2011-0068645 (published on June 22, 2011)

SUMMARY OF THE INVENTION The present invention has been conceived to solve the problems described above, and it is an object of the present invention to provide a noise reduction method and a noise reduction method which reduce a calculation amount in image noise removal by using NL means (non-local means) And an object of the present invention is to provide an efficient image noise removal method and system using the NL means method which can improve the image noise.

According to an aspect of the present invention, there is provided an efficient image noise canceling method using the NL means technique, including: converting a noise-containing image into a Discrete Cosine Transform (DCT) region; A step of selecting some coefficients in the block of the discrete cosine transformed region, a step of calculating a correction value by applying the NL means method to the selected partial coefficient, the step of applying the calculated correction value to a spatial region And a control unit.

In the method of removing image noise according to the present invention, a coefficient having a predetermined energy concentration degree or more may be selected in the step of selecting the partial coefficient.

The method of the present invention further includes calculating a correction value by calculating a weighted sum for a patch located at the center and included in a block of the discrete cosine transformed region in the step of calculating the correction value, .

According to another aspect of the present invention, there is provided a method of removing image noise, comprising: converting a noise-containing image into a redundant transform domain using a lapped transform; Calculating a correction value by applying the NL means method to the selected partial coefficient, and applying the calculated correction value to the spatial region.

According to an aspect of the present invention, there is provided an image noise removing apparatus including a DCT transform module for transforming a noise-containing image into a discrete cosine transform (DCT) domain, And a correction value calculation module for calculating a correction value by applying an NL means method to some coefficients selected by the target pixel selection module, And the correction value calculated in the calculation module is applied to the spatial region.

According to another aspect of the present invention, there is provided an apparatus for removing noise in an image, the apparatus comprising: a redundancy conversion module for converting a noise-containing image into a lapped transform region; And a correction value calculation module for calculating a correction value by applying the NL means method to some coefficients selected by the target pixel selection module, wherein the correction value calculation module calculates a correction value And the correction value is applied on the spatial region.

According to the efficient image noise cancellation method and system using the non-local means technique of the present invention, the weight is calculated by non-local means technique using only some coefficients in the transform domain, Can be improved.

Also, according to the present invention, if similarity of each block is calculated using only a few coefficients in the lapped transform region, there is an advantage that performance of the NL means can be improved due to high energy concentration of the redundant transform.

FIG. 1 is a block diagram schematically illustrating a configuration of an efficient image noise removing apparatus using an NL (Non-local) means method according to an embodiment of the present invention.
FIG. 2 is a flowchart illustrating an efficient method of removing an image noise using the NL means method according to the present invention.
3 is an exemplary diagram illustrating an application of the NL means method of the present invention in a transform domain;
4 is a block diagram schematically illustrating the configuration of an efficient image noise removing apparatus using a non-local means (NL) method according to another embodiment of the present invention.
FIG. 5 is a flowchart illustrating an efficient image noise removal method using the NL means method according to the present invention.
FIGS. 6 and 7 are illustrations showing exemplary results of eliminating image noise by applying the NL means method of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a detailed description of preferred embodiments of the present invention will be given with reference to the accompanying drawings. In the following description of the present invention, detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear.

Embodiments in accordance with the concepts of the present invention can make various changes and have various forms, so that specific embodiments are illustrated in the drawings and described in detail in this specification or application. It is to be understood, however, that it is not intended to limit the embodiments according to the concepts of the present invention to the particular forms of disclosure, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention.

FIG. 1 is a block diagram schematically showing the configuration of an efficient image noise system using an NL (Non-local) means technique according to the present invention, FIG. 2 is a block diagram illustrating an efficient image noise canceling method using the NL means technique according to the present invention FIG.

3 is an exemplary diagram illustrating an exemplary application of the NL means method for some coefficients of the present invention in a transform domain.

The method according to the invention will generally be done via a computing device.

Referring to the drawings, an efficient image noise removing apparatus 10 using an NL (Non-local) means according to an embodiment of the present invention includes a DCT transform module 100, a target pixel selection module 110, Module 120, as shown in FIG.

The apparatus according to the present invention first receives an image to be noise-canceled through an input unit, and then stores it in a memory (not shown).

Thereafter, the DCT transform module 100 converts the stored image into a Discrete Cosine Transform (DCT) region (S101). Then, the target pixel selection module 110 selects some coefficients (represented by a gray lattice) having a high energy concentration as shown in FIG. 3 (b) in the block of the discrete cosine transformed region through the DCT transform module (S102).

The coefficient of high energy concentration occurs according to the characteristic of DCT conversion. In the DCT conversion, energy concentration phenomenon occurs in which data (signal) is driven at a low frequency. According to such an energy concentration phenomenon, data is crowded in a portion represented by a gray lattice (pixel) in FIG. 3 (b), and only a lattice (pixel) in which a data rounding phenomenon above a predetermined level occurs is selected.

Thereafter, the correction value calculation module 120 calculates using only the coefficient of the pixel selected through the target pixel selection module 110 using [Equation 4].

Next, the correction value calculation module 120 applies the calculated result to the spatial area (S104).

As described above, there is a disadvantage in that a large amount of calculation is required in order to calculate the local similarity in the image noise cancellation method using the NL means, but in the present invention, the image is transformed into the transform domain, By calculating the local similarity, image noise cancellation using NL means can be efficiently implemented.

FIG. 4 is a block diagram schematically illustrating the configuration of an efficient image noise system using an NL (Non-local) means according to another embodiment of the present invention, and FIG. 5 is a block diagram illustrating an NL means technique according to another embodiment of the present invention. FIG. 2 is a flowchart illustrating an efficient method for removing an image noise.

As shown in the figure, the image noise removing apparatus 20 may include an overlap conversion module 200, a target pixel selection module 210, and a correction value calculation module 220.

The redundancy conversion module 200 converts an image stored in a memory (not shown) into a redundancy conversion area by applying a lapped transform method (S201).

Then, the target pixel selection module 210 selects some coefficients having a high energy concentration as shown in FIG. 3 (b) in the block of the overlap-transformed region through the redundancy conversion module 200 (S202).

Thereafter, the correction value calculation module 220 calculates correction values using only the coefficients of the pixels selected through the target pixel selection module 210 using the following relational expression (S203). At this time, the performance of the NL means can be improved due to the high energy concentration of the redundant transform.

Next, by applying the resultant value calculated by the correction value calculation module 220 to the spatial region (S204), it is possible to provide a noise-free image.

FIG. 6 and FIG. 7 illustrate exemplary results of eliminating image noise by applying the NL means technique of the present invention.

In order to analyze the performance of the image noise cancellation technique using the NL means of the present invention, as shown in FIGS. 6 and 7, the performance of the NL means method under various conditions using the brightness components of the 768 × 512 Kodak standard image Respectively.

First, Table 1 and Table 2 show the results of analyzing the performance of the NL means in the conventional pixel unit and patch unit.

h Comparison block cap parrot 4σ 3 x 3 31.56 32.99 5σ 5 × 5 31.95 32.65 7σ 7 × 7 31.52 32.46

h Comparison block cap parrot 4σ 4 × 4 31.92 32.49 5σ 6 x 6 31.81 32.20 7σ 8 x 8 31.67 32.29

As shown in Tables 1 and 2, when noise is removed in units of patches, the noise reduction performance is lowered, but the difference is not large compared with the case where noise is removed on a pixel-by-pixel basis.

Table 3 and Table 4 show the results of performance analysis using only some coefficients in the DCT domain and the lapped transform domain according to the present invention.

h Comparison block cap parrot 4σ 4 × 4 32.22 32.93 7σ 8 x 8 32.23 33.17

h Comparison block cap parrot 4σ 4 × 4 32.19 32.73 7σ 8 x 8 32.35 33.13

NL means using some coefficients of the DCT domain, using 5 coefficients for 4 × 4 DCT and 16 coefficients for 8 × 8 DCT. In the Lapped transform domain, 6 coefficients are used for the 4 × 4 lapped transform and 16 coefficients are used for the 8 × 8 lapped transform.

As shown in FIG. 3 (b), a weight is obtained by using only five or six coefficients in the case of a block size of 4 × 4, and a weight is obtained by using only 16 coefficients in the case of an 8 × 8 block It can be seen that the noise cancellation performance is superior to the result obtained by weighting using the total coefficient.

Since the noise of the image is a high-frequency component, it can be determined that a part of the noise of the image is removed when a weight is obtained by excluding only a part of high-frequency components as shown in FIG. 3 (b).

In addition, Table 5 and Table 6 show the results obtained by eliminating the noise of the image for the four cases in which the relative positions of the 2 × 2 patches are different, and then obtaining the averages.

h Comparison block cap parrot 4σ 4 × 4 32.49 33.33 7σ 8 x 8 32.48 33.36

h Comparison block cap parrot 4σ 4 × 4 32.63 33.19 7σ 8 x 8 32.64 33.33

Table 5 shows the average of four offsets after applying the NL means using some coefficients of the DCT domain. Table 6 shows the results of the four offsets after applying the NL means using the partial coefficients of the lapped transform region. And the average value is obtained.

As described above, when noise is removed using only some coefficients, it can be seen that there is no significant difference from the result of removing noise by applying the NL means to the existing total coefficients, and it can be confirmed that only local coefficients There is a characteristic that calculation can be reduced by calculation.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims. will be. Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

10, 20: Image noise canceller
100: DCT conversion module
200: Redundant conversion module
110, 210: target pixel selection module
120, 220: correction value calculating module

Claims (10)

delete delete delete A method for eliminating image noise,
Converting an image including a noise into a redundant transform region using a lapped transform;
Selecting a partial coefficient having a predetermined energy concentration degree or more in a block of the redundantly transformed region;
Calculating a correction value by calculating a weighted sum for a patch located at a center and included in a block of the redundantly transformed region by applying a non-local means method to the selected partial coefficient; And
And applying the calculated correction value on the spatial region,
Wherein the noise cancellation performance is improved while reducing the amount of calculation by calculating local similarities only for some coefficients with a high energy concentration in the redundant transform domain.
delete delete delete delete delete An image noise removing apparatus using the image noise removing method according to claim 4,
A redundancy conversion module for converting an image including noise into a lapped transform region;
A target pixel selection module for selecting some coefficients in a block of a region subjected to redundancy conversion by the redundancy conversion module; And
And a correction value calculation module for calculating a correction value by applying a non-local means method to some coefficients selected by the target pixel selection module,
And applying the correction value calculated by the correction value calculating module to the spatial region.

Images