KR19990039993A - Adaptive 3D Median Filter and Its Noise Reduction Method - Google Patents

Abstract

The present invention relates to an adaptive three-dimensional median filter and a noise canceling method thereof, and more particularly, to determine whether a nonlinear rank filter is selected by calculating whether a frame is damaged and the number of pixels in the damaged frame. It is designed to reduce the amount of computation to enable real-time image processing and to preserve intact pixels. The present invention includes a first nonlinear rank filter block 220 that removes impulse noise when the noise figure of the current image is less than a predetermined value, and a second nonlinear that removes impulse noise when the noise figure of the current image is greater than a predetermined value. The first nonlinear rank filter block 220 is selected when the number of damaged pixels calculated by the rank filter block 230 and the first and second nonlinear rank filters 220 and 230 is smaller than a predetermined value. A pixel selection block 240 for controlling the switching unit 250 to select the second non-linear rank filter block 230 when it is larger than a predetermined value, a value of a signal of a previous frame, a signal of a next frame, and the current frame; Selects a signal of the current frame if the difference values are smaller than a predetermined value, and selects a frame controlling the switching unit 270 to select an output signal of the switching unit 250 if the difference is greater than a predetermined value.Characterized in block 260.

Description

Adaptive 3D Median Filter and Its Noise Reduction Method

TECHNICAL FIELD The present invention relates to image reconstruction, and more particularly, to an adaptive three-dimensional median filter for impulse noise cancellation and a method for removing the noise thereof.

The development of image transmission and compression technology has made it possible to speed up image transmission and store large amounts of images. However, image damage has been a serious problem due to noise during image transmission.

In order to prevent such image damage, a method using a median filter, a weighted median filter, an FMH filter, and the like has been conventionally proposed.

The method using the median filter is the most common method to restore a pixel damaged by noise by taking a constant window around the current pixel and replacing the current pixel value with the median of the corresponding values.

The method using the weighted median filter recovers pixels damaged by noise by taking a constant window around the current pixel and weighting the corresponding value to find the intermediate value.

The method using the FIR median hybrid (FMH) filter subdivides a constant window around the current pixel into several pieces, takes an FIR filter on each of them, and mediates the value to restore pixels damaged by noise.

However, the conventional technology has a problem in that the processing speed is reduced by a large amount of computation, and the image is blurred by affecting pixels that are not damaged by noise.

In other words, the method of using a general median filter is not effective when there is a lot of noise, there is a large amount of calculation, and there is a problem that the image is blurred by replacing intact pixels with other values.

In addition, the method of using a weighted median filter reduces the blurring of the image but increases the amount of calculation relatively sharply according to the weight, thereby lowering the processing speed.

In addition, the method using the FMH filter has a large amount of calculation and it is difficult to find an optimal FIR filter.

Accordingly, the present invention can reduce the amount of computation required to remove the impulse noise by real-time image processing by determining whether a nonlinear rank filter is selected by calculating whether the frame is damaged and the number of pixels in the damaged frame in order to improve the conventional problem. The purpose of the present invention is to provide an adaptive three-dimensional median filter and its noise canceling method, which are designed to preserve the undamaged pixels.

1 is an exemplary view showing an input pixel of the present invention.

2 is a block diagram showing an embodiment of the present invention.

3 is an exemplary view calculating a noise figure for each frame in the present invention.

Figure 4 is a waveform diagram showing a signal-to-noise ratio comparing the present invention and the prior art.

5 is an exemplary view showing image restoration in the present invention.

Explanation of symbols on the main parts of the drawings

210,280: frame memory 220,230: nonlinear rank filter block

240: pixel selection block 250, 270: switching unit

260 frame selection block

In order to achieve the above object, the present invention provides a first non-linear rank block that is selected when the noise figure of the current image is smaller than a predetermined value to remove impulse noise in the image, and is selected when the noise figure of the current image is larger than a predetermined value. A second nonlinear rank block for removing impulse noise in the image, a pixel selection block for selecting one of the first and second nonlinear rank blocks by calculating the number of damaged pixels in the image, and a threshold for a given amount of change in the image frame It is characterized by consisting of a frame selection block that selects the current frame when smaller than the value.

The second nonlinear rank filter block is configured by connecting two nonlinear rank filters in series.

In addition, the present invention is to determine whether there is a damaged pixel in the frame of the current image in order to achieve the above object, the step of determining the noise figure of the current image when there is a damaged pixel in the frame, and the current image And removing the impulsive noise according to the noise figure of.

The determining of the damaged pixel may include determining that the damaged pixel exists when the change amount of the result value of the previous frame and the next frame and the value of the current frame are larger than the threshold value.

The noise figure determination step is characterized by determining by counting the number of damaged pixels present in the previous frame.

The impulsive noise removing step is characterized in that when the noise figure is greater than the threshold, the second nonlinear rank filter block is executed, and when the noise figure is smaller, the first nonlinear rank filter block is executed.

Hereinafter, the present invention will be described in detail with reference to the drawings.

2 is a block diagram showing an embodiment of the present invention, as shown therein, a frame memory 210 storing an image signal x (u, v, t + 1), and a signal of a previous frame y (u , v, t-1)), the signal of the next frame (x (u, v, t + 1)) and the output signal (x (u, v, t)) of the frame memory 210 are received. The first nonlinear rank filter block 220 removes impulse noise when the noise figure is less than a predetermined value, the signal of the previous frame (y (u, v, t-1)), and the signal of the next frame (x (u). (v, t + 1)) and the output signal (x (u, v, t)) of the current frame to receive two non-linear rank filters (i) to remove impulse noise when the noise figure of the current image is greater than a predetermined value. 231 and 232, a second nonlinear rank filter block 230, a switching unit 250 for selecting one of the first and second nonlinear rank filters 220, 230, and the first, second Calculate the number of corrupted pixels calculated by the two nonlinear rank filters 220, 230 To select the first non-linear rank filter block 220 when smaller than the predetermined value, and to select the second non-linear rank filter block 230 when larger than the predetermined value. And a switching unit 270 which selects one of an output signal x (u, v, t) of the frame memory 210 and an output signal _xm (u, v, t) of the switching unit 250. ), The frame memory 280 for storing the output signal y (u, v, t) of the switching unit 270, the signal (y (u, v, t-1) of the previous frame), and the next Compare each value of the signal x (u, v, t + 1) of the frame with the signal x (u, v, t) of the current frame and if the difference values are smaller than a predetermined value, the frame memory. Select the output signal x (u, v, t) of 210 and switch to select the output signal _xm (u, v, t) of the switching unit 250 when the difference value is greater than a predetermined value; The frame selection block 260 controls the unit 270.

Referring to the operation and effect of the embodiment of the present invention configured as described above are as follows.

When the image signal of the frame unit is input to the apparatus of the present invention, it is temporarily stored in the frame memory 210 and directly to the first nonlinear rank filter block 220, the second nonlinear rank filter block 230, and the frame selection block 260. In addition, the data is temporarily stored in the frame memory 210 and input to the blocks 220, 230, and 260.

The frame memory 280 may output the output signal y (u, v, t-1) of the previous frame to the first nonlinear rank filter block 220, the second nonlinear rank filter block 230, and the frame selection block ( 260).

That is, the present invention requires 9 signals of the current frame, a signal of the previous frame, and a signal of the next frame as shown in FIG. 1, and needs 11 windows to calculate the result of the current frame.

This is expressed as in Equation (1) below.

--- Equation (1)

At this time, the frame selection block 260 is a signal of the current frame (x (u, v, t)), a signal of the next frame (x (u, v, t + 1)) and a signal of the previous frame (y (u, v, t-1)), the absolute value of the difference between the current frame and the previous frame.backward_difference = abs (x (u, v, t)-y (u, v, t)) and the difference between the current frame and the next frame. Calculate whether the current frame is damaged by comparing whether the absolute value of forward_difference = abs (x (u, v, t)-x (u, v, t + 1)) is smaller than the threshold Th.

Accordingly, when the difference values are all smaller than the threshold value Th, it is determined that the current frame is not damaged by noise and selects the signal x (u, v, t) of the current frame output from the frame memory 210. On the contrary, if any one of the difference values is larger than the threshold Th, the switching unit 270 determines that the current frame is damaged by noise, and thus the output signal of the switching unit 250 (x _m (u, v, t)). The switching unit 270 is controlled to select.

Here, using the frame correlation, the threshold value given the amount of change of each value of the previous frame (y (u, v, t-1)) and the next frame (x (u, v, t + 1) and the value of the current frame If it is determined to be smaller than (Th) it can be seen that the value of the current frame is not damaged by the noise, so that the switching unit 270 selects the signal of the current frame to reduce the amount of calculation of the overall filter.

Experimentally the given threshold Th is '10'.

That is, the operation of the frame selection block 260 as described above is performed by the same process as the following program.

backward_difference = abs (x (u, v, t) -y (u, v, t));

forward_difference = abs (x (u, v, t) -x (u, v, t + 1));

if (backward_difference <Th && forward_difference <Th)

y (u, v, t) = x (u, v, t);

else

y (u, v, t) = x _m (u, v, t);

If it is determined that the current frame is damaged by noise and the switching unit 270 is switched to the switching unit 250, one of the nonlinear rank filter blocks 220 and 230 is compared with the threshold of the noise figure of the previous frame. Will be selected.

The first and second non-linear rank filter blocks 220 and 230 include a ROM to output the filtered signal when the pixel is damaged.

The ROM (ROM) sequentially lists 10 pixels excluding its own value of x (u, v, t) among 11 pixels inputted as in Equation (1) to obtain Equation (2) below. Determine the value of the ROM (m (u, v, t)) as shown in equation (3).

---- Formula (2)

----- Formula (3)

First, the first nonlinear rank filter block 220 is performed when the noise figure of the previous frame is smaller than the given threshold TL, and the operation thereof is given by Equation 4 below.

------ Formula (4)

At this time, in order to obtain the value of the constant (a), the difference is calculated by performing the same operation as in Equation (5) below.

--- Equation (5)

Where k = 1, ..., 5.

Accordingly, by comparing the five values obtained by the above equation (5) with the given five 'TL _k ' values, if any one is large, 'a' becomes '0'.

At this time, the threshold value TL is experimentally obtained and the value is 'TL ₁ ' = 2, 'TL ₂ ' = 10, 'TL ₃ ' = 18, 'TL ₄ ' = 40, 'TL ₅ ' = 50 .

Accordingly, the output signal of the first nonlinear rank filter block 220 ( ) Is the output signal of the ROM filter when the value of the constant (a) is '0'. ) And '1', the output signal of the frame memory 210 ( )

In this case, '1' means that the pixel is not damaged by noise.

Then, the second nonlinear rank filter block 230 is composed of two nonlinear rank filters 231 and 232, and is performed when the noise figure of the previous frame is larger than a given threshold TH. )

In this case, many of the eleven pixels are more likely to be damaged by impulse noise, so the maximum and minimum values are discarded.

------ Formula (6)

In this case, the second nonlinear rank filter block 230 determines the value of the constant b through the operation as shown in Equation (7) below to select the output signal (x _h (u, v, t)).

--- Equation (7)

Where k = 2, ..., 5.

Accordingly, by comparing the four values obtained by the above equation (7) with the given four thresholds TH _k , if any one is large, the constant b becomes '0'.

At this time, the threshold value TH was experimentally obtained, and the values are 'TH ₂ ' = 2, 'TH ₃ ' = 10, 'TH ₄ ' = 30, and 'TH ₅ ' = 40.

Therefore, the output signal of the nonlinear rank filter block 230 ) Is the output signal m (u, v, t) of the ROM filter when the value of 'b' is '0', and the output signal x (of the frame memory 210 when it is '1'). u, v, t)).

In this case, when the first and second nonlinear Ram filter blocks 220 and 230 determine whether the pixels are damaged by noise, the pixel selection block 240 calculates the number of damaged pixels in the previous frame to determine the current frame. Predict the noise level.

That is, the pixel selection block 240 calculates this number because the pixels selected by the constant (a or b) as '0' in the first and second nonlinear rank filter blocks 220 and 230 are pixels damaged by impulses. The calculated number indicates the percentage of corrupted pixels in the frame as a percentage.

Accordingly, the pixel selection block 240 selects the output signal x _l (u, v, t) of the first nonlinear rank filter block 220 when the percentage is less than or equal to the predetermined value, and the second nonlinearity when the percentage is greater than or equal to the predetermined value. The switching unit 250 is controlled to select the output signal x _h (u, v, t) of the rank filter block 230.

For example, if the percentage is 12% or less, the output signal x _l (u, v, t) of the first non-linear rank filter block 220 is selected and if it is 12% or more, the output signal of the non-linear rank filter block 230 is greater than 12%. (x _h (u, v, t)) is selected and output through the switching unit 270.

That is, when the original image as shown in FIG. 5 (a) is damaged by noise as shown in FIG. 5 (b), the present invention determines whether the pixels in the frame are damaged or not, and thus, the first and second non-linear rank filter blocks 220 ( By selecting one of the output signals x _h (u, v, t) and (x _l (u, v, t)) of 230, a reconstructed image is finally output as shown in FIG.

The output signal y (u, v, t) of the switching unit 270 is stored in the frame memory 280 to determine whether the next frame is damaged, thereby selecting the nonlinear rank filter blocks 220 and 230 and selecting the frame. Output is made to block 260.

Meanwhile, according to the present invention performing the above operation, when performing a performance evaluation on 300 frames of a 176 * 144 standard QCIF image, the noise figure per frame is shown in FIG. 3 and the peak value signal-to-noise ratio compared to the prior art. (PSNR; Peak Signal to Noise Ratio) is shown in FIG.

This calculation amount of the present invention is about 25% of the general median filter.

As described in detail above, the present invention can restore a video damaged or corrupted by image transmission or storage in a fast manner, thereby enabling real-time image processing and removing the noise by a simple hardware configuration to prevent the image from being blurred. It has an effect.

Claims (17)

A first nonlinear rank block that is selected when the noise figure of the current image is less than a predetermined value to remove impulse noise in the image, and a second nonlinear that is selected when the noise figure of the current image is more than the predetermined value to remove impulse noise in the image A pixel selecting means for selecting one of the first and second nonlinear rank blocks by calculating a rank block, the number of damaged pixels in the image, and a frame for selecting a current frame when the amount of change in the image frame is smaller than a given threshold. Adaptive three-dimensional median filter characterized in that the selection means. The method of claim 1, wherein the frame selecting means selects and outputs the current frame signal when the absolute value of the difference between the current frame and the previous frame and the absolute value of the difference between the current frame and the next frame are smaller than the threshold Th and the pixel when the frame is larger. Adaptive three-dimensional median filter for selectively outputting the output signal by the selection means. Here, the threshold Th is experimentally obtained and the value is '10'. The adaptive three-dimensional median filter according to claim 1, wherein the first and second non-linear rank filter blocks are each provided with ROMs for outputting values such as the following equations. The adaptive three-dimensional median filter according to claim 1, wherein the first nonlinear rank filter block is configured to perform an operation as in the following equation. The adaptive three-dimensional method as claimed in claim 4, wherein the constant (a) is determined to be '0' if any one is larger than the experimentally obtained threshold value (TL _k ) of the five values given by the following equation. Median filter. Where k = 1, ..., 5 and 'TL ₁ ' = 2, 'TL ₂ ' = 10, 'TL ₃ ' = 18, 'TL ₄ ' = 40 and 'TL ₅ ' = 50. The adaptive three-dimensional median filter according to claim 1, wherein the second nonlinear rank filter block is formed by connecting two nonlinear rank filters in series. 7. The adaptive three-dimensional median filter according to claim 1 or 6, wherein the second nonlinear rank filter block performs an operation as in the following equation. The adaptive value 3 according to claim 7, wherein the constant (b) is determined to be '0' if any one is larger by comparing the threshold value TH _k obtained experimentally with the four values given by the following equations. Dimensional median filter. Where k = 2, ..., 5 and 'TH ₂ ' = 2, 'TH ₃ ' = 10, 'TH ₄ ' = 30 and 'TH ₅ ' = 40. The pixel selection block of claim 1, wherein the pixel selection block selects the first nonlinear rank filter block when the constant (a or b) is less than or equal to a predetermined value by calculating the number of pixels selected as '0' in the first and second nonlinear rank filter blocks. And the second nonlinear rank filter block is selected if more than a predetermined value. 10. The adaptive three-dimensional median filter of claim 9, wherein the predetermined value compared to the number of damaged pixels is 12% as a percentage. A first step of determining whether there is a damaged pixel in a frame of the current image; a second step of determining a noise figure of the current image when there is a damaged pixel in the frame; and an impulsive noise according to the noise figure of the current image And a third step of removing the noise of the adaptive three-dimensional median filter. 12. The adaptive three-dimensional median of claim 11, wherein the first step determines that there is a damaged pixel when a change amount of the result value of the previous frame and the next frame and the value of the current frame is larger than a predetermined threshold value. How to filter out noise. The method of claim 12, further comprising selectively outputting a signal of a current frame when it is determined that there are no damaged pixels. 12. The method of claim 11, wherein the second step is to calculate and determine the number of damaged pixels present in the previous frame. 12. The adaptive three-dimensional method as claimed in claim 11, wherein the third step is to perform an operation as shown in Equation (2) below when the noise figure is greater than the threshold value and perform an operation as shown in Equation (1) as below. How to remove noise from median filter. ---- (One) ---- (2) 16. The adaptive three-dimensional method as claimed in claim 15, wherein the constant (a) is determined to be '0' if any one is larger than the experimentally obtained threshold value (TL _k ) of the five values given by the following equation. How to remove noise from median filter. Where k = 1, ..., 5 and 'TL ₁ ' = 2, 'TL ₂ ' = 10, 'TL ₃ ' = 18, 'TL ₄ ' = 40 and 'TL ₅ ' = 50. 16. The adaptive value 3 according to claim 15, wherein the constant (b) is determined to be '0' if any one is larger by comparing the threshold value TH _k obtained experimentally with the four values given by the following equation. Noise Reduction Method of Dimensional Median Filter. Where k = 2, ..., 5 and 'TH ₂ ' = 2, 'TH ₃ ' = 10, 'TH ₄ ' = 30 and 'TH ₅ ' = 40.