KR20060098952A - A device for removing the noise on the image and a method for removing the noise on the image - Google Patents

Abstract

Disclosed are a video signal noise canceller and a video signal noise canceller, which are free from distortion of image quality such as blur and blur. The video signal noise canceller includes a noise estimation block, a motion estimation block, and a mixer block. The noise estimation block estimates a noise level (NL), which is an amount of noise included in an input image signal CF (t). The motion estimation block considers the spatial correlation between the macroblock to be processed of the voltage level NL and the video signal CF (t) and neighboring macroblocks, and thus the video signal CF (t). And generating a MAD for the reference video signal RF (t-1), a reference frame, which is processed by removing noise from the video signal input before the current video signal CF (t). The mixer block calculates a weight w having a value in a predetermined range by using the MAD and the noise level NL, and reflects the weight on the reference image and the input image to output the image OF (t). ) The video signal noise removing method includes a noise estimation step, a motion estimation step, and a mixing step.

Video signal noise reduction, motion estimation,

Description

A device for removing the noise on the image and a method for removing the noise on the image}

BRIEF DESCRIPTION OF THE DRAWINGS In order to better understand the drawings cited in the detailed description of the invention, a brief description of each drawing is provided.

1 is a block diagram of a noise removing image processing apparatus according to an exemplary embodiment.

2 illustrates a method of dividing an input image into macroblocks.

3 illustrates a method of obtaining a weighted average value of motion vectors of neighboring macroblocks of a macroblock currently being processed.

4 illustrates the basic concept of motion compensation.

5 shows a method for obtaining MAD values.

The present invention relates to video signal processing, and more particularly, to a method for removing noise of a video signal.

The video signal data is transmitted via a medium such as airwaves or a cable, and the transmitted video signal data is used after being processed through certain steps. Since the transmission or processing of the video signal data is exposed to various types of noise, it is natural to add noise to the original video signal. The noise embedded in the video signal data adversely affects the picture quality or resolution of the video signal to be reproduced.

Hereinafter, advantages and disadvantages of methods for removing noise included in a conventionally used video signal will be described.

The one dimensional spatial noise reduction method is a method of removing noise in consideration of spatial correlation between images. This method removes noise of a pixel to be processed in consideration of characteristics of pixels positioned around the pixel to be processed. That is, an arbitrary pixel value is compared with its surrounding pixel values, and if any pixel value is larger than the reference pixel value or smaller than the reference pixel value, it is determined that the arbitrary pixel value is noisy and the data is processed. do.

For example, a constant search window including a pixel to be processed is defined, and a maximum value and a minimum value are obtained from values of the remaining pixels except for the pixel value to be processed among the pixels in the area. Then the value of the pixel to be processed is compared with the maximum and minimum values. If the value of the pixel to be processed is greater than the maximum value, the value of the pixel to be processed is reduced to the maximum value, and if the value of the pixel to be processed is smaller than the minimum value, the value is increased to the minimum value and is between the maximum value and the minimum value. Maintains the pixel value as it is.

However, the method has a disadvantage in that blurring of the image occurs when the number of surrounding images is reflected, and noise is not removed well when the number of surrounding images is reflected. In addition, since temporal correlation is not considered, noise that can be removed only by considering temporal correlation cannot be removed.

The one dimensional temporal noise reduction method is a method of removing noise in consideration of temporal correlation between images. That is, a predetermined threshold value is obtained by using the difference between the pixels corresponding to each other between the image previously input and processed and the current input image, and the pixel is currently processed using the threshold value. This method removes the noise involved. However, the screen is not only shaken because the pixel value is determined by simply considering the previously input image, and if only this method of considering only temporal correlation is used, there is a disadvantage in that noise due to spatial correlation cannot be removed.

The two-dimensional noise reduction method is a method of removing noise in consideration of both methods, and can overcome the above-described problems. In other words, it removes noise considering both temporal and spatial correlation.

However, the performance difference is large depending on the algorithm applied to implement this method, and it includes both the shortcomings of the one-dimensional noise canceling method considering the spatial correlation and the shortcomings of the one-dimensional noise canceling method considering the temporal correlation. The disadvantage is that it is likely to occur.

SUMMARY OF THE INVENTION The present invention has been made in an effort to provide an apparatus for removing video signal noise free from distortion of image quality such as blur and blur of an image.

Another object of the present invention is to provide a method for removing video signal noise without distortion of image quality such as blur and blur of an image.

According to an aspect of the present invention, there is provided an apparatus for removing video signal noise, including a noise estimation block, a motion estimation block, and a mixer block.

The noise estimation block estimates a noise level (NL), which is an amount of noise included in an input image signal CF (t). The motion estimation block considers the spatial correlation between the macroblock to be processed of the voltage level NL and the video signal CF (t) and neighboring macroblocks, and thus the video signal CF (t). And generating a MAD for the reference video signal RF (t-1), a reference frame, which is processed by removing noise from the video signal input before the current video signal CF (t). The mixer block calculates a weight w having a value in a predetermined range by using the MAD and the noise level NL, and reflects the weight on the reference image and the input image to output the image OF (t). )

The video signal noise removing method according to the present invention for achieving the another technical problem comprises a noise estimation step, a motion estimation step and a mixing step.

The noise estimating step estimates a noise level which is an amount of noise included in an input video signal. In the motion estimation step, noise is removed from the video signal input before the video signal and the current video signal in consideration of the spatial relationship between the macroblock to be processed and the neighboring macroblocks of the video signal. And generating a MAD of the processed reference video signal by using the MAD and the noise level to calculate a weight having a predetermined range of values, and applying the weight to the reference video and the input video. By reflecting, an output image OF (t) is generated.

DETAILED DESCRIPTION In order to fully understand the present invention, the operational advantages of the present invention, and the objects achieved by the practice of the present invention, reference should be made to the accompanying drawings that illustrate preferred embodiments of the present invention.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. Like reference numerals in the drawings denote like elements.

The present invention relates to a technique for obtaining a motion of an image by using motion estimation and to remove noise included in a current image input by using calculated motion information of the image.

1 is a block diagram of a noise removing image processing apparatus according to an exemplary embodiment.

Referring to FIG. 1, the noise removing image processing apparatus 100 includes a noise estimation block 110, a motion estimation block 120, a mixer block 10, and a storage 140.

The noise estimation block 110 estimates a noise level NL, which is an amount of noise included in an input image signal CF (t).

For example, the noise estimation is performed by dividing an input image into predetermined macroblock units, obtaining a standard deviation of macroblocks that are determined to be the same among each macroblock in a screen, and then extracting the final 10 standard deviations. Take an average.

2 illustrates a method of dividing an input image into macroblocks.

Referring to FIG. 2, after dividing the input image into units of 5 × 2 macroblocks, the standard deviation of macroblocks determined to be the same among each 5 × 2 macroblocks in the screen is obtained, and then the final 10 standards are obtained. Take the average of the deviations.

In other words, a minimum standard deviation value is selected from macroblocks having the same number for each of the ten points in the macroblock, and an average value of the ten minimum standard deviations is calculated.

The motion estimation block 120 includes a first motion estimation block 121, a second motion estimation block 122, a motion vector correction block 123, and a MAD block 124.

The first motion estimation block 121 calculates an average weight of motion vectors of macroblocks located around the macroblock to be processed.

3 illustrates a method of obtaining a weighted average value of motion vectors of neighboring macroblocks of a macroblock currently being processed.

Referring to FIG. 3, a weighted average value between three macroblocks MB1 to MB3 and a current macroblock around a macroblock to be processed may be expressed as Equation 1.

mvx (*) is a motion vector in the x direction of the current macroblock (Current MB) to be processed, and is a weighted average value of the motion vectors in the x direction of three neighboring macroblocks MB1 to MB3 located in the vicinity. mvy (*) is a motion vector in the y direction of the macroblock (Current MB) to be processed, and is a weighted average value of the motion vectors in the y direction of three neighboring macroblocks MB1 to MB3 located nearby.

The second motion estimation block 122 removes noise from a video signal input before the current video signal CF (t) and processes the reference video signal RF (t-1), a reference frame, and the current video signal. A sum of absolute difference (SAD) for processing the current video signal CF (t) is generated using (CF (t)). Here, SAD is a variable calculated to indicate the degree of similarity of the macroblocks to be compared. The SAD is an accumulated value obtained by obtaining absolute values of difference values of pixels constituting the macroblock. If the accumulated SAD value is small, the two macroblocks to be compared are similar to each other, and if the values are large, the two macroblocks to be compared are different from each other.

Algorithms for obtaining a motion vector using the SAD between the current input image CF (t) and the reference image RF (t-1) are various, and most algorithms must perform significant calculations and use memory. Its disadvantage is that it requires a lot of bandwidth.

The algorithm used in the present invention uses multi-resolution motion estimation to overcome this disadvantage.

The multi-step motion estimation method is a method of estimating motion by dividing into a plurality of steps in obtaining a motion vector required for compressing a current video signal.

In the multi-step motion estimation method, at the highest level, for example, at level 3, the input image third reduced signal having a resolution of 1/16 compared with the original input image, and 1 compared with the original reference image A third motion vector MV3 estimated from the third search area of the widest range is generated using data obtained by coding the third reduced signal of the reference image having a resolution of / 16.

In the next step, Level 2, the input reduced image second reduction signal having a 1/4 resolution compared to the original input image and the reference reduced image second reduced signal having a 1/4 resolution compared to the original reference image are respectively coded. Using the data and the third motion vector MV3, the second motion vector MV2 is estimated in the reduced second search area compared to the third search area (level 2).

In the final step, Level 1, the first motion vector is reduced in the first search area, which is reduced compared to the second search area by using the data and the second motion vector MV2, which are coded with the original input image and the original reference image, respectively. (MV1) is estimated (level 1).

In level 1, the SAD is calculated at the same position as the macroblock currently processed in order to prevent falling into a local minimum which may be caused by incorrect estimation of the motion vector at level 2 and level 3.

The motion refinement block 123 is corrected by considering the noise level NL, the weighted average values of the motion vectors (mvx and mvy), and the spatial correlation and noise level of the macroblock to be processed. Determine the motion vector (MV).

In the motion correction step, the weighted averages (mvx and mvy, Weighted Average) of motion vectors of neighboring macroblocks are used. It means that it has low frequency components such as colored walls. In this case, even a small amount of noise affects the image quality, and thus, the estimated noise should be removed in a considerable amount. On the other hand, if the weighted average values mvx and mvy are large, this means that the data difference between pixels constituting the screen is large and has a high frequency component. Since the noise generated in this case has a relatively low impact on the image quality, the degree of removing the noise may be smaller than in the case described above.

As described above, when the weighted average values (mvx and mvy) of the motion vectors of the neighboring macroblocks are small, the noise level is largely corrected. In contrast, when the weighted average values (mvx and mvy) are large, the noise level (Noise) is large. Reduce the level. Then, the difference in the noise levels after the correction will be wider than the difference in the noise levels estimated before the correction.

As described above, the motion vector correction block 123 corrects the noise level NL using the weighted average values mvx and mvy, and adjusts the range of SAD values to be searched using the corrected noise level NL. Generate the appropriate motion vector (MV).

The noise level NL is convenient to normalize using the weighted average values mvx and mvy.

4 illustrates the basic concept of motion compensation.

Referring to FIG. 4, the search proceeds only for a search point having a constant SAD value among the measured SAD values, where the range of the constant SAD value means a SAD value between two dotted lines.

As the corrected noise level (α × NL) increases, the number of search points included in the motion compensation region increases, and for the search points in the correction region, the weighted average value of the motion vectors of the neighboring blocks and the macroblock to be processed Since a search point having a small difference in motion vectors is determined as a final motion vector (RMV), the motion vector of neighboring macroblocks is adaptively followed. The human eye responds more sensitively to fixed images than to fast-changing images. The above-described method does not deteriorate image quality in fast-moving regions, greatly improves image quality and reduces noise in fixed regions, and reduces noise. The trembling of the screen, which is a big side effect, can be greatly reduced.

Magnitude of Absolute Difference (MAD) block 124 sums the absolute value (MAD) of the difference between the reference image and the input image with respect to a certain region based on the pixel at the position indicated by the motion vector (MV) obtained after the motion compensation. Obtain

5 shows a method for obtaining MAD values.

Referring to FIG. 5, a MAD of each pixel of a unit macroblock (Box represented by shade) constituting a part of a current input image is obtained using a reference image, and MAD. Satisfies Equation 2.

Here, x and y represent two-dimensional positions of pixels to be processed. a and b represent the horizontal comparison range, and α and β represent the vertical comparison range. Cur (*) denotes an input image, and Ref (*) denotes a reference image. mv _i and mv _j mean a motion vector of a macroblock including a pixel to be currently processed.

The mixer block 130 calculates a weight w having a value between 0 (zero) and 1 (one) by using the calculated MAD, and reflects the weight on the reference image and the input image to output the image (OF). (t)) In obtaining the weight w, it is important to normalize the MAD value by the noise level.

The high noise level means that the input image contains a lot of noise. In this case, the output image is generated by compressing the output image that reflects the reference image relatively more than the input image. When generating the output image from which the noise included in the current input image is removed by referring to the two image signals of the current input image and the reference image, it reflects more of the reference signal from which the noise is removed than the current image containing a lot of noise. As a result, the noise of the output image is reduced.

On the contrary, when the noise level is small, the output image reflecting the input image larger than the reference image is generated and compressed. If the current input image contains little noise, the output image generated by reflecting the current input image not only has a low noise but is most similar to the current input image.

As described above, when the current input image contains a lot of noise, the present invention generates and compresses an output image reflecting a relatively large number of reference images, thereby reducing the identity with the actual input image but significantly removing the noise. On the other hand, when the current input image contains less noise, by generating and compressing an output image that reflects the current input image relatively much, it is possible to obtain compressed data that significantly reduces the noise as well as the actual input image. There is an advantage.

The output image OF (t) or (x, y) can be displayed as shown in Equation (3).

Output video (x, y) = w x Input video (x, y) + (1-w) x Reference video (x, y)

The value of the weight w is a value that can be arbitrarily adjusted according to the intention of the user.

The storage 140 stores the output image OF (t), and outputs the previously stored reference image RF (t-1) to the motion estimation block 120 and the mixer block 140.

As described above, the optimum embodiment has been disclosed in the drawings and the specification. Although specific terms have been used herein, they are used only for the purpose of describing the present invention and are not intended to limit the scope of the invention as defined in the claims or the claims. Therefore, those skilled in the art will understand that various modifications and equivalent other embodiments are possible. Therefore, the true technical protection scope of the present invention will be defined by the technical spirit of the appended claims.

As described above, the noise removing image processing apparatus and the noise removing image processing method according to the present invention are to remove and compress various noises generated during the transmission and processing of an image signal. There is an advantage that side-effects that occur when using a conventional method such as blurring of the part or the entire screen and blurring of the screen do not occur.

Claims (21)

A noise estimating block estimating a noise level NL, which is an amount of noise included in an input image signal CF (t); The video signal CF (t) and the current video signal CF in consideration of the spatial correlation between the macroblock to be processed and the neighboring macroblocks of the voltage level NL and the video signal CF (t). (t)) a motion estimation block for generating a MAD for a reference video signal (RF (t-1), Reference Frame) processed by removing noise from a previously input video signal; And A mixer for calculating a weight w having a predetermined range of values using the MAD and the noise level NL, and generating an output image OF (t) by reflecting the weight on a reference image and an input image. Noise reduction image processing apparatus comprising a block. The method of claim 1, wherein the noise level is, The input image is divided into macroblock units having 5 × 2 pixels, a standard deviation of pixels located at the same position among a plurality of macroblocks is obtained, and finally an average of 10 (5 × 2) standard deviations is obtained. Noise canceling image processing apparatus, characterized in that the obtained value. According to claim 1, wherein the motion estimation block, A first motion estimation block for calculating average weights (mvx and mvy) of motion vectors of macroblocks located in the vicinity of the macroblock to be processed; The input using the reference image (RF (t-1), Reference Frame) and the input image CF (t) processed by removing noise from an image signal input before the input image CF (t). A second motion estimation block for generating a sum of absolute difference (SAD) for processing the image CF (t); A motion vector correction block for determining a motion vector (MV) corrected using the noise level (NL), weighted average values (mvx and mvy) of the motion vector, and the SAD; And And a MAD (Multitude of absolute difference) block for obtaining a sum MAD of the difference between the reference image and the input image with respect to a predetermined area based on the pixel at the position indicated by the corrected motion vector. Noise canceling image processing apparatus, characterized in that. The method of claim 3, wherein the average weights (mvx and mvy),

, And the weighted average value mvx (x, y) is a motion vector in the x direction of the current macroblock (Current MB) to be processed. The weighted average value of the motion vectors, and the weighted average value mvy (x, y) is the motion vector in the y direction of the current macroblock (Current MB) to be processed, y of the three peripheral macroblocks (MB1 to MB3) located in the vicinity Noise canceling image processing device, characterized in that the weighted average value of the motion vector in the direction. The method of claim 3, wherein the SAD, A variable that is calculated to indicate the degree of similarity between the macroblocks of the input image to be compared with the macroblocks of the reference image. The absolute value of the difference values of the pixels constituting the macroblock is accumulated and then accumulated. And if the accumulated SAD value is small, two macroblocks to be compared are similar to each other, and if larger, the two macroblocks to be compared are different from each other. The method of claim 5, wherein the second motion estimation block, In obtaining a motion vector necessary for compressing the input image, a multi-resolution motion estimation method is used, which is a method of estimating a motion by dividing it into a plurality of steps. Processing unit. The method of claim 3, wherein the motion vector correction block, When the weighted average values mvx and mvy of the motion vector are smaller than a predetermined value, the noise level is largely corrected. On the contrary, when the weighted average values mvx and mvy are large, the noise level is noised. Level) and adjust the range of SAD values to be searched using the corrected noise level to generate the corrected motion vector (MV). The method of claim 3, wherein the MAD,

Where x and y represent the position of the pixel to be processed in two dimensions, a and b represent the comparison range of the horizontal, α and β represent the comparison range of the vertical, and Cur (*) Denotes the input image, Ref (*) denotes the reference image, and mv _i and mv _{j denote} motion vectors of a macroblock including a pixel to be processed. . The method of claim 1, wherein the output image (OF (t) or (x, y)) of the mixer block, Output video (x, y) = w x Input video (x, y) + (1-w) x Reference video (x, y) And a value of the weight (w) is a value that can be arbitrarily adjusted according to the intention of the user to use. The noise canceling image processing apparatus of claim 1, And a storage device for storing the output image (x, y) and outputting the reference image to the motion estimation block and the mixer block. A noise estimation step of estimating a noise level which is an amount of noise included in an input video signal; The reference video signal processed by removing noise from the video signal and the video signal input before the current video signal in consideration of the spatial correlation between the voltage level and the macroblock to be processed and the neighboring macroblocks. A motion estimation step of generating a MAD for the; and A mixing step of calculating a weight having a value in a predetermined range using the MAD and the noise level, and generating an output image OF (t) by reflecting the weight on the reference image and the input image; Noise reduction image processing method, characterized in that. The method of claim 11, wherein the noise level, The input image is divided into macroblock units having 5 × 2 pixels, a standard deviation of pixels located at the same position among a plurality of macroblocks is obtained, and finally an average of 10 (5 × 2) standard deviations is obtained. Noise removal image processing method characterized in that the obtained value. The method of claim 11, wherein the motion estimation step, A first motion estimation step of calculating average weights (mvx and mvy) of motion vectors of macroblocks located in the vicinity of a macroblock to be processed; The input using the reference image (RF (t-1), Reference Frame) and the input image CF (t) processed by removing noise from an image signal input before the input image CF (t). A second motion estimation step of generating a sum of absorptive difference (SAD) for processing the image CF (t); A motion vector correction step of determining a motion vector (MV) corrected using the noise level (NL), weighted average values (mvx and mvy) of the motion vector, and the SAD; And And calculating a MAD (Multitude of absolute difference) which is a sum of absolute values of the difference between the reference image and the input image in a predetermined region based on the pixel at the position indicated by the corrected motion vector (MV). Noise reduction image processing device. The method of claim 13, wherein the average weights (mvx and mvy),

The weighted average value mvx (x, y) is a motion vector in the x direction of the current macroblock (Current MB) to be processed, and is the x direction of three peripheral macroblocks MB1 to MB3 located in the vicinity. Is the weighted average of the motion vectors of and the weighted average mvy (x, y) is the motion vector in the y direction of the current macroblock (Current MB) to be processed, and the three neighboring macroblocks MB1 to MB3 located in the vicinity Noise canceling image processing method characterized in that the weighted average value of the motion vector in the y direction. The method of claim 13, wherein the SAD, A variable that is calculated to indicate the degree of similarity between the macroblocks of the input image to be compared with the macroblocks of the reference image. The absolute value of the difference values of the pixels constituting the macroblock is accumulated and then accumulated. And if the accumulated SAD value is small, the two macroblocks to be compared are similar to each other. On the contrary, if the SAD values are large, the two macroblocks to be compared are different from each other. The method of claim 15, wherein the second motion estimation step, In obtaining a motion vector necessary for compressing the input image, a multi-resolution motion estimation method is used, which is a method of estimating a motion by dividing it into a plurality of steps. Treatment method. The method of claim 13, wherein the motion vector correction block, When the weighted average values mvx and mvy of the motion vector are smaller than a predetermined value, the noise level is largely corrected. On the contrary, when the weighted average values mvx and mvy are large, the noise level Level) and adjust the range of SAD values to be searched using the corrected noise level to generate the corrected motion vector (MV). The method of claim 13, wherein the MAD,

Where x and y represent the position of the pixel to be processed in two dimensions, a and b represent the comparison range of the horizontal, α and β represent the comparison range of the vertical, and Cur (*) Denotes the input image, Ref (*) denotes the reference image, and mv _i and mv _{j denote} a motion vector of a macroblock including a pixel to be currently processed. . The method of claim 11, wherein the output image (OF (t) or (x, y)) of the mixer block, Output video (x, y) = w x Input video (x, y) + (1-w) x Reference video (x, y) And a value of the weight w is a value that can be arbitrarily adjusted according to the intention of the user. The noise canceling image processing apparatus of claim 1, And storing and outputting the output image (x, y) and outputting the reference image to the motion estimation block and the mixer block. A noise estimation step of estimating a noise level which is an amount of noise included in an input video signal; The reference video signal processed by removing noise from the video signal and the video signal input before the current video signal in consideration of the spatial correlation between the voltage level and the macroblock to be processed and the neighboring macroblocks. A motion estimation step of generating a MAD for the; and The mixing step of calculating a weight having a value in a predetermined range using the MAD and the noise level and generating the output image OF (t) by reflecting the weight on the reference image and the input image is performed. Storage means including instructions or programs that can be.

Images