KR101048530B1 - Image Noise Reduction Device and Method Using Jet Mask - Google Patents

Abstract

Disclosed are an apparatus and a method for removing image noise using a jet mask. The image noise removing apparatus comprises an ID map generator for generating ID maps of the entire image by making structural feature information of the pixels to be restored and neighboring pixels into IDs, and converts the noise image into the generated ID maps using a jet mask. And a noise remover configured to remove noise of the noise image on a pixel-by-pixel basis using the changed ID map and the changed ID map.

Jet mask, image, noise, rejection

Description

Image Noise Reduction Apparatus and Method Using Jet Mask {IMAGE DENOISING APPARATUS AND METHOD USING Z-MASK}

The present invention relates to an apparatus and method for removing image noise using a jet mask, and more particularly, to generate an ID map by using structural feature information of pixels and neighboring pixels, and to convert a noise image into an ID map using a jet mask. The present invention relates to an apparatus and a method for changing noise to eliminate noise.

In general, noise corrupts the information in the image, reducing the visual information that is conveyed to the viewer. Such noise can be processed in the frequency domain and spatial domain of the image. When it is removed from the frequency domain, noise and texture are included in the oscillating form. Is going on.

In the spatial domain, a typical noise reduction method calculates the similarity between the pixels to be restored and the neighboring pixels to remove the noise. The difference between the pixels or the pixel value of the original image is generated. Due to this difference, the restored value of some pixels severely damaged by noise is calculated differently from the original pixel values. To solve these problems, the noise reduction method using NL (Non-Local) Means shows the restoration result that is robust to noise by using non-local pixels in the search window even if the pixel to be restored is damaged by noise. There is a disadvantage that it takes a lot of time.

The present invention provides an image noise canceling apparatus and method capable of reconstructing an image similar to the original image using a jet mask and improving a speed of removing noise.

The present invention also provides an apparatus and method for removing noise by changing a noise image into an ID map using a jet mask kernel.

The present invention also provides an apparatus and method for removing noise by variably applying a variable in consideration of peripheral information of a restored pixel.

An image noise removing device according to an embodiment of the present invention includes an ID map generator for generating ID maps of an entire image by making structural feature information of a pixel to be restored and neighboring pixels into an ID, and a noise image using a jet mask. A changer for changing the generated ID map to a generated ID map, and a noise remover for removing noise of the noise image by a pixel unit using the changed ID map.

According to an aspect of the present invention, the ID map generator may generate an ID map by giving an ID to each pixel of the noise image with a jet mask kernel.

In addition, according to an aspect of the present invention, the noise removing unit may remove the noise of the noise image by using the average value of the weights applied to the pixels to be restored and the neighboring pixels as a recovery value of each pixel.

In addition, according to an aspect of the present invention, the weight is determined through the Euclidean distance between the ID, the higher the similarity can be given more.

According to an aspect of the present invention, the ID map generator may generate the ID map by variably selecting the number of pixels used for restoration according to the structural feature information.

In addition, the method for removing noise according to an embodiment of the present invention may include generating ID maps of the entire image by making structural ID information of a pixel to be restored and neighboring pixels in the ID map generator as an ID. Changing a noise image into an ID map generated by the ID map generator by using a jet mask, and removing the noise of the noise image by a pixel unit using the changed ID map by the noise removing unit; Include.

According to an aspect of the present invention, the generating of the ID map may generate the ID map by assigning an ID to each pixel of the noise image by the jet mask kernel in the ID map generator.

According to an aspect of the present invention, the removing of the noise is performed by the noise removing unit using a weighted average value of the pixels to be neighbored to the pixel to be restored as a restoration value of each pixel. Noise can be removed.

In addition, according to an aspect of the present invention, the weight is determined through the Euclidean distance between IDs, the higher the similarity can be given more.

According to an aspect of the present invention, in the generating of the ID map, the ID map generator may generate the ID map by variably selecting the number of pixels used for restoration according to the structural feature information.

According to the present invention, by removing the noise using a jet mask in the noise image can be reconstructed more similar to the original image that does not contain noise.

In addition, according to the present invention, it is possible to shorten the time to perform the noise removal by improving the noise removal speed using a jet mask in the noise image.

Hereinafter, an apparatus and a method for removing image noise using a jet filter will be described in detail with reference to the accompanying drawings.

1 is a diagram illustrating a configuration of an apparatus for removing video noise according to an embodiment of the present invention.

Referring to FIG. 1, the image noise removing apparatus 100 according to an embodiment of the present invention includes an ID map generator 110, a changer 120, and a noise remover 130.

The ID map generator 110 generates an ID map by assigning IDs to respective pixels of the noise image using the jet mask kernel. For example, the ID map generator 110 assigns four IDs to each pixel with the jet mask kernel, and generates each ID map by expressing each ID as a digit sequence in which two 2-digit integers are consecutive. Can be.

2 is a view showing an example of a jet mask according to the present invention.

Referring to FIG. 2, the first mask 210 is a 7 × 7 mask centered on the pixel i to be reconstructed, and the second mask 220 is located at a portion where contrast (hatched) of the first mask 210 is included. The mask is a corresponding mask, and the third mask 230 is a jet mask calculated through the second mask 220.

3 is a diagram illustrating the form of four IDs assigned to a pixel.

라고 정의한다.

는 순서대로 화소i와 가까운 거리에 놓여 있으며, 제트 마스크의 값을 이용하여 수학식 1에 의해 계산된다.Referring to FIG. 3, a set of four IDs around the pixel i is defined as i_ id , and when each ID is defined as i _k , each two-digit integer constituting the ID is defined.

It is defined as.

Is in close proximity to the pixel i in order, and is calculated by Equation 1 using the value of the jet mask.

Z (n) is calculated by the value of the mask shown in the second mask 220, and the method of calculating the Z (n) value is expressed by Equation (2).

In Equation 2, α is a variable that determines the number of digits of Z (n) . For example, α = 10.5 can be applied to fit all Z (n) to two digits. Pixel i and neighboring independent pixels are difficult to robustly express the pixel i and the adjacent structural information due to the influence of noise. Therefore, Z (n) calculated as an average of four pixels in Equation 2 may be used as a unit representing structural information neighboring the pixel i .

의 전체적인 구조를 살펴보면, 아이디를 잡음 영상의 모든 화소에 부여하면 전체적인 아이디 맵을 생성할 수 있다. The ID map generator 110 is a jet mask kernel, which is assigned to four IDs assigned to the pixel i .

Looking at the overall structure of, the ID is assigned to all the pixels of the noise image to generate an overall ID map.

For example, the ID map generator 110 may generate the ID map by using a large number of pixels in a region where the structural characteristic information is simple, and by using a small number of pixels in a region where the structural characteristic information is complicated. In the conventional NL-Means algorithm, all pixels of the search window are used irrespective of regions in which the structural features of the image are complicated or simple, but the present invention can variably select the number of pixels used for reconstruction according to the structural features. To perform this process, first, the ID map generator 110 calculates a structural similarity with respect to the periphery of the pixel to be reconstructed.

In Equation 3, i_id is a numeric string representing IDs (IDs) of pixels i to be restored, and i_dis represents Euclidean distances for respective IDs of pixels i. If the value of i_dis is low, the peripheral area of pixel i is structurally flat or symmetrical. If the value of i_dis is high, the peripheral area of pixel i is structurally fine and asymmetrical.

In Equation 4, i_id represents a peripheral ID of pixel i, j_id represents a peripheral ID of pixel j, and w represents a weight of pixel j with respect to pixel i. G uses one of the parts symmetrical with respect to the center of the two-dimensional Gaussian function, and β is a reference value for determining pixels to be excluded from reconstruction.

As described above, the image noise device 100 according to the present invention uses a large number of pixels in a region having a simple structural characteristic, and uses a small number of pixels in a region having a complicated structural characteristic so that a low similarity pixel is used for reconstruction. The problem can be solved, and the process can more effectively restore the noise image.

The changer 120 changes the noise image into the generated ID map using a jet mask. That is, the changer 120 changes the noise image into an ID map generated by the ID map generator 110 using the jet mask as a preprocessing process when removing noise.

The noise removing unit 130 removes the noise of the noise image in units of pixels by using the changed ID map. For example, the noise removing unit 130 may remove the noise of the noisy image by using an average value, which is a weight value applied to the pixels to be restored and neighboring pixels, as a recovery value of each pixel. The weight of each pixel is determined by the Euclidean distance between the IDs, and the higher the similarity, the larger the weight.

4 is a diagram illustrating an example of a noise removal result.

Referring to FIG. 4, the first image 410 is a Lena original image, the second image 420 is a noise image including noise in the original image, and the third image 430 is reconstructed by the NL-Means method. The fourth image 440 is an image reconstructed by the noise canceling method according to the present invention.

As such, the image noise removing apparatus 100 according to the present invention can provide a clearer image reconstructed closer to the Lena original image as well as shorten the time for removing noise compared to the image reconstructed by the NL-Means method. You can.

5 is a diagram illustrating another example of the noise removal result.

Referring to FIG. 5, the first image 510 is an original airplane image, the second image 520 is a noise image including noise in the original image, and the third image 530 is reconstructed by the NL-Means method. The fourth image 540 is an image reconstructed by the noise canceling method according to the present invention.

As such, the noise canceling apparatus 100 according to the present invention may provide a reconstructed image 540 that is sharper than the image 530 reconstructed by the NL-Means method and more similar to the original airplane image 510. Not only that, but it also reduces the time to remove the noise.

6 is a flowchart illustrating an operation of an image noise removing method according to an embodiment of the present invention.

1 to 6, in operation S610, the image noise removing apparatus 100 generates ID maps of the entire image by making structural feature information of pixels to be restored and neighboring pixels as IDs. That is, the image noise removing apparatus 100 generates ID maps for the entire image by making the structural feature information of the pixels to be restored and the neighboring pixels through the ID map generator 110 as an ID (S610).

For example, in operation S610, the image noise removing apparatus 100 may generate the ID map by assigning an ID to each pixel of the noise image by the jet mask kernel through the ID map generator 110.

As another example, in operation S610, the image noise removing apparatus 100 may generate the ID map by variably selecting the number of pixels used for restoration according to the structural feature information through the ID map generator 110. .

In another example, in operation S610, the image noise removing apparatus 100 uses a large number of pixels in an area where the structural characteristic information is simple, and uses a small number of pixels in an area where the structural characteristic information is complicated. You can create a map.

The complexity of the noise canceling method according to the present invention is shown in Equation 5, and the complexity of the NL-Means method is shown in Equation 6.

Here, N, W, and M represent horizontal and vertical sizes of an image, a search window, and a mask, and O represents a complexity.

인 경우에는 NL-Means 알고리즘을 이용한 방법보다 본 발명에 따른 잡음 제거 방법이 더 효율적이다. 일반적으로 잡음에 강건한 마스크는 M≥7이 사용되며, 본 발명에 따른 잡음 제거 방법은 NL-Means 알고리즘을 이용한 방법보다 효율적인 수행시간을 보장할 수 있다. if,

In case of, the noise canceling method according to the present invention is more efficient than the method using NL-Means algorithm. In general, M≥7 is used as a mask that is robust against noise, and the noise cancellation method according to the present invention can ensure more efficient execution time than the method using the NL-Means algorithm.

In operation S620, the image noise removing apparatus 100 changes the noise image into the generated ID map using a jet mask. That is, the image noise removing apparatus 100 changes the noise image into the ID map generated by the ID map generator 110 through the changer 120 using the jet mask (S620).

In operation S630, the image noise removing apparatus 100 removes noise of the noisy image on a pixel-by-pixel basis using the changed ID map. That is, the image noise removing apparatus 100 removes the noise of the noisy image in units of pixels by using the changed ID map through the noise removing unit 130 (S630). For example, in operation S630, the image noise removing apparatus 100 uses the noise removing unit 130 as a reconstruction value of each pixel using an average value applied to weights of the pixels to be restored and neighboring pixels as a reconstruction value of each pixel. It can remove the noise for. The weight is determined by the Euclidean distance between the IDs, the higher the similarity can be given more.

To evaluate the noise image restoration method, white Gaussian noise with noise standard deviation of 20 was used, and the variable h that controls the result value in NL-Means was applied to 2.12 and used in the image noise removing method according to the present invention. Each of the values for the calculated variables is represented by Equation 7.

In general, the performance evaluation of the reconstructed image uses a PSNR (Peak Signal-to Noise Ration), and is defined as Equation (8).

Here, I ₀ denotes an original image, _I denotes a reconstructed image, and Ω denotes an entire pixel set of the image.

Table 1 shows the PSNR comparison results for the NL-Means algorithm and the method according to the invention.

NL-Means Method according to the invention Boat 28.8 dB 29.0 dB Lena 31.3 dB 31.6 dB Baboon 24.5 dB 24.5 dB Papper 31.6 dB 31.9dB Airplane 30.3 dB 31.6 dB Average 29.3 dB 29.7 dB

Referring to Table 1, the resultant image reconstructed by the image noise removing method according to the present invention may be reconstructed more similarly to the original image than the resultant image reconstructed by the NL-means algorithm.

Table 2 shows the performance comparison of the NL-means algorithm and the execution time for the method according to the present invention and compares them with the average value of the execution time used to apply noise to the 50 natural images (492 × 748). .

NL-Means Method according to the invention Average execution time (unit s) 290.1 112.3

Referring to Table 2, the image noise removing method according to the present invention can reduce the execution time by about 61.3% compared with the case where the NL-means algorithm is applied.

As such, the image noise removing method according to the present invention can not only reconstruct an image more similar to the image quality of the original image by removing the noise of the image using a jet mask, but also improve the processing speed of removing the noise.

Embodiments according to the present invention can be implemented in the form of program instructions that can be executed by various computer means can be recorded on a computer readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. Program instructions recorded on the media may be those specially designed and constructed for the purposes of the present invention, or they may be of the kind well-known and available to those having skill in the computer software arts. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tape, optical media such as CD-ROMs, DVDs, and magnetic disks, such as floppy disks. Magneto-optical media, and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine code generated by a compiler, but also high-level language code that can be executed by a computer using an interpreter or the like. The hardware device described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

As described above, the present invention has been described by way of limited embodiments and drawings, but the present invention is not limited to the above embodiments, and those skilled in the art to which the present invention pertains various modifications and variations from such descriptions. This is possible.

Therefore, the scope of the present invention should not be limited to the described embodiments, but should be determined not only by the claims below but also by the equivalents of the claims.

1 is a view showing the configuration of a noise removing device according to an embodiment of the present invention.

2 is a view showing an example of a jet mask according to the present invention.

3 is a diagram illustrating the form of four IDs assigned to a pixel.

4 is a diagram illustrating an example of a noise removal result.

5 is a diagram illustrating another example of the noise removal result.

6 is a flowchart illustrating an operation of a noise removing method according to an embodiment of the present invention.

Claims (10)

An ID map generator for generating ID maps of the entire image of the noisy image by assigning IDs to the pixels to be restored by making structural feature information of the pixels to be reconstructed and neighboring pixels in the noisy image; And A noise removing unit for removing the noise of the noise image by the pixel unit using the ID map. Image noise reduction device comprising a. delete The method of claim 1, The noise removing unit, And removing the noise of the noisy image by using a weighted average value of the pixels to be reconstructed and neighboring pixels as a reconstruction value of each pixel. The method of claim 3, The weight is, And the higher the similarity, the larger the similarity is determined by the Euclidean distance for the IDs given to the pixels to be restored. The method of claim 1, The ID map generation unit, And variably selecting the number of pixels used for restoration according to the structural feature information to generate the ID map. An image noise removing method in a noise removing device including an ID map generator and a noise remover, Generating, by the ID map generator, IDs of the entire image of the noisy image by assigning IDs to the pixels to be restored by making structural feature information of the pixels to be reconstructed from the noisy image and neighboring pixels into a plurality of IDs; ; And In the noise removing unit, removing the noise of the noise image in units of pixels using the ID map. Image noise reduction method comprising a. delete delete delete delete

Images