KR20010003018A - Recursive noise reduction system - Google Patents

Abstract

PURPOSE: A system of removing a recurring noise is provided to remove a background noise such as a gauss noise and a uniform noise distributed in a moving images while minimizing a loss of the moving images. CONSTITUTION: A filter(10) separates a moving image and a noise that is included in a video input and a recursive data. A histogram part(20) values the separation information of the moving image and the noise from the filter(10) and obtains the factor of a recursive algorithm. A cross-fade part(30) adaptively biases the video input and the recursive data of the recursive algorithm using the factor of the histogram thereby removing a recursive noise such as a gauss noise and a uniform noise while minimizing a loss of the moving images.

Description

Recursive noise reduction system

The present invention relates to a cyclic noise removal system that suppresses the loss of information generated when removing background noise distributed in a video.

When the noise generated during the video signal processing is classified into statistical characteristics, it is classified into impulse noise, Gaussian noise, and uniform noise.

Background noise is randomly generated in the screen when displaying a video, and when the random noise pixel is calculated as a probability, most of the Gaussian or uniform noise having the form as shown in FIG. do.

For reference, impulse noise refers to intensively generated at a specific amplitude as shown in FIG.

In order to remove background noise from a video screen, a method using a noise removal filter is collectively adopted, but when background noise is removed by a filter, there is a problem that video information is also lost along with the noise.

The present invention is to remove background noise such as gaussian or uniform noise distributed in the video while minimizing the loss of the video information.

The present invention extracts the degree of motion and noise, quantifies the extracted information, and adaptively biases the video input and the cyclic data to obtain a noise removing effect. And a histogram for digitizing information separated by the filtering means to obtain coefficients, and a crossfade means for biasing video input and cyclic data using coefficients obtained from the histogram.

1 is a background noise and impulse noise state diagram

2 is an exemplary circuit diagram of the present invention.

3 is a circuit diagram of another embodiment of the present invention.

4 is a filtering state diagram of the average filter of the present invention.

[Explanation of symbols on the main parts of the drawings]

1,2: line memory and circuit 3,4: delay

5,6: difference signal output unit 10: filter unit

11,12: average filter 20: histogram

23,24: histogram 30: crossfade

32: crossfade

FIG. 2 is an exemplary circuit diagram of the present invention, and includes a filter unit 10 for separating video and noise included in video input and cyclic data, and a histogram unit 20 for obtaining coefficients by digitizing separation information of the filter unit 10. And a crossfade section 30 for biasing the video input and the circulating data using the coefficients obtained by the histogram section 20.

The filter unit 10 includes an average filter 11 for filtering a video input, and an average filter 12 for filtering recursive data input stored and input to the memory 13.

The histogram unit 20 includes a difference signal output unit 21 which obtains a difference signal by inputting outputs Cm1 and Cm2 of the average filters 11 and 12, and a center pixel m1 of video input and cyclic data. Histogram 1 for dividing the pixel amplitudes of the difference signal output unit 22 that obtains the difference signal by inputting (m2) and the pixel amplitudes of the difference signal output units 21 and 22 into eight levels, and then obtaining the sum of the respective levels. , 2 (23) and (24).

The crossfade section 30 is a sum signal of the difference signal sum-cm3 (sum-m3) of the outputs sum-Cm3 (sum-m3) of the histograms 1,2 (23) and 24 (sum-m3). ) Is obtained from the difference signal output unit 31, the sum of the difference signal output unit 31, the output (cm3) (m3) of the difference signal output units 21 and 22, and the histogram 1 (23). It consists of a crossfade 32 which biases the video input and the cyclic data by inputting the sum of the upper level values (video information) and the center pixels m1 (m2) of the video input and the cyclic data input.

The output from the crossfade 32 is obtained by the following equation.

mxfade = {[Limit (sumdiff × mxfade-coeff / 256)] × hibucket} / 256

K = Limit {[m3 × (QUANT-mxfade) + (m3 × mxfade) / QUANT}

Video-Out = m2 + (m1-m2) × crossfade-yuv

(mxfade: crossfade coefficient in frames

K: crossfade coefficient in pixels)

The video-out of the present invention is biased to the cyclic data when it is determined that noise is distributed in the currently input frame, and vice versa to the data of the current input frame.

However, if noise and video exist simultaneously in the input frame, the output should be calculated considering both noise and video.

In this case, the following formula (1) is used, and the crossfade coefficient K should be calculated considering noise and video at the same time.

(Equation-1) Video-Out = m2 + (m1-m2) × K

Video-Out: Output

m1: Frame data currently input

m2: rotated frame data

K: crossfade coefficient (bias coefficient)

In order to obtain the crossfade coefficient K as described above, an algorithm capable of accurately measuring the amount of noise present in a frame and accurately detecting a video generated between frames is required, which is implemented based on the following equation.

(A) Moving picture and noise information moving picture in frame unit: In FIG. 2, cm3 which is the difference signal between cm1 and cm2 output of average filter 11 and 12 represents moving picture information.

The degree of moving picture between frames is sumdiff, which is the difference between the output (sum-cm3) of the histogram 1 (23) and the output (sum-m3) of the histogram 2 (24), which is a measure of noise.

(Equation-2) shown below is a crossfade coefficient in units of frames.

(Mx-2) mxfade = sumdiff x hibucket

(B) Video and noise in pixels.

As shown in Equation 2, when the video and noise information in the frame unit are applied as the bias of the output, the video cannot be applied to the change of the video.

However, using the following Equation-3, the crossfade coefficient that changes in units of pixels can be calculated to compensate for the above disadvantage.

K = (m3 × (QUANT-mxfade) + cm3 × mxfade) / QUANT

Where QUANT is a constant and 256 is used in this algorithm.

As described above, in the present invention, the crossfade coefficient K is biased to m1 and m2 to the degree of motion picture and noise as shown in (Equation-1). This can be prevented.

The present invention is also implemented by another embodiment as shown in FIG. 3, wherein a video input is applied to a line memory and a circuit 1 for extracting surrounding pixels based on a center pixel among input signals, and a cyclic frame data input diagram is also provided. It is applied to the line memory and the circuit 2 for extracting 9 pixels around the center pixel.

Nine pixels extracted from the line memory and the circuit (1) (2) are applied to the average filter (11) (12) so that nine pixels based on the center pixel are filtered using respective coefficients.

In this case, the filtering coefficients of the average filters 11 and 12 are as shown in FIG. 4 based on the center pixel.

The retarders 3 and 4 are added to the data path for the center pixel of the video input or the cyclic data input in order to take into account the information delay occurring in the filtering process of the average filter 11 and 12 described above.

The difference signal output unit 5 obtains a difference between the outputs of the average filters 11 and 12, and the difference signal output unit 6 obtains a difference with respect to the center pixel passing through the delay units 3 and 4. As shown in FIG.

The difference signal output m3 (cm3) of the difference signal outputs 5 and 6 is applied to the histogram 20 to divide each input pixel amplitude by eight levels and obtain the sum of each level. The histogram 20 Is supplied to the mxfade generating section 8.

The mxfade generator 8 generates mxfade, which is a crossfade coefficient in units of frames, using sum-diff and Hibucket values, which are outputs of the histogram 20.

At this time, generate mxuvfade for color.

The crossfade generator 7 generates a crossfade coefficient K using the output of the mxfade generator 8 and the output m3 of the difference signal generator 5 and cm3. The crossfade coefficient K is applied to the crossfade 32 to determine the video output by the center pixel m1 (m2).

According to the present invention, it is possible to remove background noise while minimizing video information loss by adaptively biasing video input and cyclic data by extracting and digitizing video and noise.

That is, the present invention minimizes the generation of the image information as compared with the case of applying the filter collectively when removing the background noise.

Claims (8)

In the removal of the background noise distributed in the video, comprising a filtering means for separating the video and noise information, and the histogram for obtaining the coefficient of the cyclic algorithm by digitizing the separated video and noise information from the filtering means And a crossfade means for biasing the circulating data of the video input and the cyclic algorithm by using coefficients. The cyclic noise reduction system of claim 1, wherein the video input and the cyclic data input are respectively filtered to detect a moving image, and the difference is calculated. 2. The cyclic noise reduction system according to claim 1, wherein the difference between the output of the average filter and the data before the filter input is determined to detect the noise. 4. The cyclic noise reduction system of claim 2 or 3, which uses a histogram having eight levels to extract video and noise information in units of frames. The cyclic noise removal system according to claim 2 or 3, wherein both the difference between the filter input signal and the difference between the filter output signal are used to reflect the video and noise information on a pixel basis. The cyclic noise reduction system of claim 4, wherein mxfade = sumdiff × hibucket is implemented to reflect the information of frame units in the output. mxfade: Crossfade count in frames sumdiff: Difference between sum-cm3 and cm3 (noise information) hibucket: Sum of highest level values of histogram 1 (video information) 6. The method of claim 5, wherein K = (m3 × (QUANT−mxfade) + (m3 × mxfade) / QUANT) in order to reflect the information of the pixel unit in the output. K: crossfade coefficient in pixels m3: the difference between m1 and m2 mxfade: crossfade coefficient QUANT: constant In order to remove the noise, a coefficient calculated in the form of K = (m3 × (QUANT-mxfade) + (m3 × mxfade) / QUANT) is implemented in the form of Video-Out = m3 + (m1-m2) × K. Cyclic Noise Reduction System.