KR20140124180A - Method and apparatus for removing noise in image - Google Patents

Abstract

The present invention relates to a method and an apparatus for removing noise included in an image by converting an image in a spatial domain into data in a frequency domain. The present invention provides a method for removing noise included in one image comprises the steps of: (a) converting the image into data in a frequency domain; (b) setting a threshold using the data in the frequency domain; (c) thresholding the data in the frequency domain using the threshold; and (d) restoring an original image by inverting the thresholded data.

Description

TECHNICAL FIELD [0001] The present invention relates to a method and apparatus for removing noise in an image,

The present invention relates to a method and apparatus for removing noise included in an image, and more particularly, to a method and apparatus for converting an image in a spatial domain into a frequency domain to remove noise included in the image.

When a transmitter converts an image such as a moving image or a still image into an electric signal and then transmits the signal through a communication channel, noise is mixed in the image during transmission, and when the receiver receives the image, A phenomenon appears. As described above, when a distorted image due to noise is displayed on a display device, the screen is not smooth, which is disruptive to the eyes. In addition, when compressing and storing an image according to a moving picture compression standard such as Moving Picture Experts Group (MPEG), compression efficiency is deteriorated due to the influence of the noise.

For this reason, studies are actively conducted to remove noise from an image. In the early stage, noise is removed in spatial domain and noise is removed in time domain in order to remove image noise. However, in recent years, various techniques for converting an image into a frequency domain have been developed. It has been found that the technique is simple and effective when removing noise in the frequency domain, and a method of removing noise in the frequency domain is widely used.

As an example of a method of removing noise using frequency conversion, there is a method of removing noise using a parsess of an image. This method is a method of recovering an image by finding the spatiality of an image using an over complete DCT basis, and then calculating a basis and a sparse coefficient. This method has a large amount of computation and therefore has a limitation in actual implementation.

An object of the present invention is to provide a noise cancellation method and an apparatus for removing noise included in an image by converting an image into a frequency domain.

According to an aspect of the present invention,

A method of removing noise included in one image, the method comprising: (a) converting the image into frequency domain data; (b) setting a threshold using data in the frequency domain; (c) thresholding data in the frequency domain using the threshold value; And (d) restoring the original image by inversely transforming the threshold-processed data.

In order to solve the above problems,

A method of removing noise included in a moving picture, the method comprising: (a) converting a first image frame among a plurality of image frames constituting the moving image into frequency domain data; (b) setting a threshold value for the first image frame using the data in the frequency domain; (c) threshold-processing the data in the frequency domain using the threshold value; (d) inversely transforming the threshold-processed data to restore the original image; And (e) verifying that the last one of the plurality of video frames is a video frame, repeating the steps (a) to (e) until the last video frame is obtained.

In order to solve the above problems,

A frequency converter for converting the image data into frequency domain data; A threshold setting unit configured to set a threshold value using data in the frequency domain; A threshold value processor for thresholding the frequency domain data using the threshold value; And an image reconstruction unit for reconstructing the original image by inversely transforming the threshold-processed data.

As described above, the present invention converts an image into a frequency domain using a Discrete Cosine Transform (DCT) method, which is widely used in conventional image compression. Therefore, it is not necessary to add a new frequency conversion function or a new circuit to a video processing apparatus such as a semiconductor integrated circuit, and as a result, noise included in the video can be economically removed.

Also, as a result of the test using the test image, it has been proved that the noise cancellation performance according to the present invention is significantly superior to other methods.

1 is a flowchart illustrating a noise canceling method of an image according to an exemplary embodiment of the present invention.
FIG. 2 (a) shows the original image, and FIG. 2 (b) shows a state where the discrete cosine transform is performed on the original image and is converted into the frequency domain.
FIG. 3 (a) shows a state in which the original image is divided into 8 8 blocks, and FIG. 3 (b) shows a state in which discrete cosine transformation is performed on the original image.
4 is a flowchart illustrating a method of removing noise of an image according to another embodiment of the present invention.
5 is a block diagram of an image noise removal apparatus according to the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily carry out the present invention. Like reference numerals in the drawings denote like elements.

The meaning of the terms described in the present invention should be understood as follows. The word " comprises "or" having "and / or " having ", when used in this specification, are to be construed as including the plural number of expressions, unless the context clearly dictates otherwise, Elements, parts, or combinations thereof, and does not preclude the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof do.

1 is a flowchart illustrating a noise canceling method of an image according to an exemplary embodiment of the present invention. That is, FIG. 1 is a flowchart illustrating a noise removal method of a still image. Referring to FIG. 1, a method for removing noise in an image includes first through fourth steps S111 through S141.

In the first step S111, one image, i.e., still image, is converted into frequency domain data. For this, a discrete cosine transform (DCT) method can be used.

An image generated by shooting an object is stored in a memory as an image signal or transmitted to another device. When a video signal is transmitted to another apparatus, for example, from a semiconductor integrated circuit to another semiconductor integrated circuit or via a network to a central processing unit (CPU) or a microcontroller, A noise generated from a network, a signal line, or the like may be included. The noise reduction method of the present invention can be applied to remove the noise. And converts the received video signal into a frequency domain. In this way, the discrete cosine transform method is executed to convert the image data into the frequency domain. When the image data is subjected to the discrete cosine transform, the data in the spatial domain is transformed into the data in the frequency domain. A method of transforming data in a spatial domain into data in a frequency domain includes a discrete cosine transform (DFT), a discrete sine transform (DST), a wavelet transform, A contourlet transform, and a curllet transform.

In order to perform the discrete cosine transform, the image data included in the image signal is divided into a plurality of blocks. Referring to FIG. 3 (a), the image data may be divided into 8 8 blocks (64 blocks). Each block includes various coefficients, e.g., the same or different coefficients.

Next, discrete cosine transform is performed on the plurality of divided blocks. When the discrete cosine transform is performed, the video data of the spatial domain shown in Fig. 2A is converted into the frequency domain data as shown in Fig. 2B. Referring to FIG. 2 (b), it can be seen that data is concentrated on the lower frequency side (DC part). In other words, there is a portion 211 which rises up to the upper left (DC portion), where the data is the largest value, and the remaining portion (AC portion) is generally flat. This means that the average color value of the 8th and 8th blocks becomes the DC component, and the part that is different from the remaining average value becomes the AC component. This also means that the value of DC is large because adjacent pixels are of a similar color. The human eye is also sensitive to low frequency components (DC), but not to high frequency components (AC).

As described above, when the image is converted into the frequency domain by using the discrete cosine transform method, an energy concentration phenomenon occurs in which the data converges to a low frequency (DC) domain.

The result of discrete cosine transformation is shown in Fig. 3 (b) as a coefficient. Referring to FIG. 3 (b), it can be seen that the coefficient of a block in a row 1 column (low frequency, DC) has a much larger value 1480 than a coefficient in other matrixes. As described above, when the image is subjected to discrete cosine transform, a coefficient having a high data value is concentrated in a low frequency region. That is, an energy concentration phenomenon occurs in which the energy is concentrated on the DC side.

This can be expressed as follows.

[Equation 1]

y = x + n

Here, y represents a video signal, i.e., a block signal including noise, x represents a signal that minimizes a mean square error (MSE) with respect to an original video signal, and n represents noise.

When the signal and the noise are independent, Expression (1) is expressed in terms of energy.

&Quot; (2) "

Ey = Ex + En

Here, Ey represents the energy of the video signal including noise, Ex represents the energy of the video signal without noise, and En represents the energy of noise.

Therefore, the energy (Ex) of the noise-free video signal is expressed by the following equation (3).

&Quot; (3) "

Ex = Ey - En

In the second step S121, a threshold value of the image data in the frequency domain is set.

In order to set the threshold value, the noise variance for the coefficients in the frequency domain is first calculated. In order to calculate the noise variance, the average value for 64 coefficients in the frequency domain is calculated, the average value is subtracted from each of the 64 coefficients, and the sum is squared.

Specifically, the noise variance can be calculated as " average of sum of squares of (coefficient-average value) ". That is, the noise variance is an indicator of how close the coefficients are to the mean.

In Equation (3), the noise energy (En) can be modeled as Equation (4) below.

&Quot; (4) "

Here,? Represents a noise variance.

K is a noise constant, which can be set to a constant value irrespective of the type of video signal or the degree of noise. This can be confirmed through experiments. As a result of the experiment, the noise constant (k) is about 1.34.

The value of the noise energy may be set to the threshold value.

In the third step S131, thresholding is performed on the image data in the frequency domain.

The threshold value processing method includes a hard thresholding method and a soft thresholding method, and a hard threshold value processing method can be partially applied to the present invention.

In order to perform the threshold processing, Equation (3) may be used. That is, the coefficients of the 64 blocks in the frequency domain are summed in order of decreasing energy, and the corresponding blocks are removed before the sum reaches the threshold value.

For example, in FIG. 2 (b), the coefficients having small numerical values are arranged in order from the beginning, and each of the coefficients is summed in order. That is, the coefficients having small numerical values are listed as follows: 0.3, 0.7, -1.0, 1.0, 1.1, 1.4, 1.6, 1.8, -1.9, -2.0, -2.1, 2.1, 2.4, 2.5, , -3.0, -3.2, 3.2, 3.4, 3.8, 4.0 ... And their energies, that is, the values obtained by squaring them are 0.09, 0.49, 1, 1.21, 1.96, 2.56, 3.24, 3.8, 4.0, 4.41, 4.41, 5.76, 6.25, 7.84, 7.84, 7.84, 9.0, , 10.24, 11.56, 14.4, 16.0. . Therefore, from 0.09 to 14.4 out of the squared values, 119.14 is obtained, and when 0.09 to 16.0 is added, 135.14 is obtained.

Assuming that the noise variance is 10, the threshold value becomes 134 (1.34 ㅧ 10 ㅧ 10), which is a threshold value processing method, and is sequentially listed from the smallest coefficient among the coefficients. That is, the coefficients, for example, coefficients corresponding to 0.3 to 3.8, which are sequentially arranged from 0.3 to 1480 and whose values summed by squaring these values are smaller than the threshold value 134 are all made to be 0 (zero) The coefficients among the coefficients included in the value whose sum of squares exceeds the threshold value 134, for example, 4.0 to 1480, are maintained.

In the image data converted into the frequency domain, the image signal is distributed at a low frequency, and the noise is distributed more in the high frequency than in the image signal. In FIG. 3 (b), the video signal is mainly included in the DC region, and the noise is distributed in the AC region. Therefore, as described above, by performing threshold processing using the threshold value, the noise distributed in the AC region can be removed.

Table 1 below shows the results of applying the noise reduction method according to the present invention. Table 1 uses the Peak Signal to Noise Ratio (PSNR) for performance comparison.

Type of input image Noise image Invention Barbara Video σ = 5 34.13 dB 37.90 dB σ = 15 24.61 dB 31.68 dB Lena video σ = 5 34.13 dB 38.90 dB σ = 15 24.61 dB 32.25 dB

According to Table 1, when the present invention is applied, PSNR of about 3 to 8 [dB] is improved according to the magnitude of noise variance. That is, it can be seen that the noise cancellation performance is improved by that much.

In a fourth step S 141, the original image is reconstructed by performing an inverse discrete cosine transform on the threshold-processed image signal. In order to prevent block artifacts that may occur in the inverse discrete cosine transform process, the blocks are shifted by moving the positions of the blocks smaller than the blocks. That is, a weighted average is applied.

In order to reduce the blocking phenomenon caused by the absence of overlapping blocks, the weighted average is obtained by dividing the blocks by a certain pixel interval (1, 2, 4, 8, ...; In the case of 8-pixel intervals, non-overlapping blocks are executed. This is not a weighted average). When the pixel interval is 1, a maximum of 64 overlaps occur.

4 is a flowchart illustrating a method of removing noise of an image according to another embodiment of the present invention. That is, FIG. 4 is a flowchart illustrating a noise cancellation method of moving images. Referring to FIG. 4, the noise cancellation method of moving picture includes first through fifth steps S411 through S451.

A moving picture is composed of a plurality of image frames. That is, in the case of a still image, the discrete cosine transform is performed on the still image. In case of a moving image, discrete cosine transform is performed on each of a plurality of image frames constituting the moving image. The process of performing the discrete cosine transform on each image frame is the same as performing the discrete cosine transform on the still image. That is, the first to fourth steps S411 to S421 of FIG. 4 are the same as the first to fourth steps S111 to S141 of FIG. 1, and in order to avoid redundant explanations thereof, S411 to S441) will not be described in detail.

However, when the discrete cosine transform is performed on a moving image, since the noise variance values are different for each image frame, the threshold value is different for each image frame accordingly. That is, the threshold value is not fixed, and a noise removal method suitable for each image frame is applied. In this way, according to another embodiment of the present invention, since noise is removed by applying different threshold values to each image frame, the noise included in the moving image can be efficiently removed.

In a fifth step S451, it is determined whether the image frame in which the first through fourth steps S411 through S441 of the plurality of image frames is completed is the last image frame. If it is not the last image frame, the image frame is changed (S452), and the first through fifth steps S411 through S451 are performed for the other image frames.

The flowcharts shown in Figs. 1 and 4 may be executed by a program, a logic circuit, an ASIC, a firmware, or the like stored in the data processing apparatus. Therefore, the present invention is not limited to the above-described embodiments.

The steps described to illustrate FIG. 1 and FIG. 4 may be performed in a different order than described above, unless the context clearly dictates a particular order. That is, each step may proceed substantially the same as the described order, or may proceed simultaneously or in reverse order.

5 is a block diagram of an image noise canceller 501 according to the present invention. 5, an image noise canceller 501 includes a frequency transform unit 511, a threshold value setting unit 521, a threshold value processing unit 531, and an inverse transform unit 541. The image noise canceller 501 may be constituted by a single data processor such as a semiconductor integrated circuit or a microcontroller having a function of being able to process software stored therein or provided from the outside, , The data processing apparatus may include all of the elements 511 to 541 shown in FIG.

The frequency conversion unit 511 converts the image data included in the input video signal y into spatial domain data in the frequency domain. And divides the image data into a plurality of blocks before converting the image data into frequency domain data. Referring to FIG. 3 (a), the image data may be divided into 8 8 blocks. Each block has various coefficients. In order to convert the divided image data into frequency domain data, a discrete cosine transform method is used. When the image data is moving image data, the frequency converter converts each of the plurality of image frames constituting the image data into frequency domain data.

The threshold value setting unit 521 sets a threshold value using the data in the frequency domain. In order to set the threshold value, the noise variance for the coefficients in the frequency domain is first calculated. In order to calculate the noise variance, the average value for 64 coefficients in the frequency domain is calculated, the average value is subtracted from each of the 64 coefficients, and the sum is squared. The noise energy En may be modeled as Equation (4), where the noise constant k may be set to a constant value regardless of the type of the image signal or the degree of noise. The noise constant k is about 1.34. The value of the noise energy may be set to the threshold value.

The threshold value processing unit 531 thresholds the frequency domain data using the threshold value. In order to process the threshold value, Equation (3) can be used. That is, the coefficients of the 64 blocks in the frequency domain are summed in order of decreasing energy, and the corresponding blocks are removed before the sum reaches the threshold value. In the image data converted into the frequency domain, the image signal is distributed at a low frequency, and the noise is distributed more in the high frequency than in the image signal. In FIG. 3 (b), the video signal is mainly included in the DC region, and the noise is distributed in the AC region. Therefore, as described above, by performing threshold processing using the threshold value, the noise distributed in the AC region can be removed. According to Table 1, when the present invention is applied, PSNR of about 3 to 8 [dB] is improved according to the magnitude of noise variance. That is, it can be seen that the noise cancellation performance is improved by that much.

The image reconstruction unit 541 inversely transforms the threshold-processed data to restore the original image signal x. The image reconstruction unit 541 uses an inverse discrete cosine transform method to perform the inverse transform. In order to prevent a blocking phenomenon that may occur in the inverse discrete cosine transform, blocks are shifted by moving the positions of the blocks smaller than the blocks. That is, a weighted average is applied.

Although the present invention has been described with reference to the embodiments shown in the drawings, it is to be understood that various modifications and equivalent embodiments may be made by those skilled in the art without departing from the scope of the present invention. Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

Claims (15)

A method for removing noise included in one image,
(a) converting the image into frequency domain data;
(b) setting a threshold using data in the frequency domain;
(c) threshold-processing the data in the frequency domain using the threshold value; And
(d) reconstructing the original image by inversely transforming the threshold-processed data. The method according to claim 1,
Wherein a discrete cosine transform method is used to transform the image into the frequency domain data, and an inverse discrete cosine transform method is used to perform the inverse transform. The method according to claim 1,
And dividing the image into a plurality of blocks before converting the image into a frequency domain. The method according to claim 1,
Wherein the data transformed into the frequency domain comprises 8 8 blocks. The method according to claim 1,
Wherein the noise variance of the data in the frequency domain is calculated before setting the threshold value. 6. The method of claim 5,
Wherein the threshold value is set by multiplying the noise variance by a noise constant. The method according to claim 1,
Wherein the threshold processing step sums the lowest coefficient among the data in the frequency domain and sequentially sums up the blocks, and removes the blocks having the summed coefficients before the sum exceeds the threshold. The method according to claim 1,
Wherein the weighted average method is applied to the inverse transform process. The method according to claim 1,
Wherein the one image is a still image. The method according to claim 1,
Wherein the steps (a) to (d) are executed in one processing device. A method for eliminating noise included in a moving picture,
(a) converting a first image frame of the plurality of image frames constituting the moving image into frequency domain data;
(b) setting a threshold value for the first image frame using the data in the frequency domain;
(c) threshold-processing the data in the frequency domain using the threshold value;
(d) inversely transforming the threshold-processed data to restore the original image; And
(e) checking if the last one of the plurality of image frames is a first image frame,
Wherein the steps (a) to (e) are repeated until the last image frame is obtained. 12. The method of claim 11,
Wherein a discrete cosine transform method is used to transform the image into the frequency domain data, and an inverse discrete cosine transform method is used to perform the inverse transform. A frequency converter for converting the image data into frequency domain data;
A threshold setting unit configured to set a threshold value using data in the frequency domain;
A threshold value processor for thresholding the frequency domain data using the threshold value; And
And an image restoring unit for restoring the original image by inversely transforming the threshold-processed data. 14. The method of claim 13,
Wherein a discrete cosine transform method is used to transform the image into the frequency domain data, and an inverse discrete cosine transform method is used to perform the inverse transform. 14. The method of claim 13,
Wherein the threshold processing unit sums the lowest coefficient among the data in the frequency domain and sequentially sums the data, and removes the blocks having the summed coefficients before the sum exceeds the threshold.

Images