KR100772405B1 - Methods for adaptive noise reduction based on global motion estimation and video processing system therefore - Google Patents

Abstract

The present invention relates to an improved temporal noise reduction method and system for estimating global motion and adjusting the overall gain of temporal filtering. The temporal noise attenuation technique is applied to two video frames, the input video frame currently mixed with noise and the previous filtered video frame stored in memory. In this method, noise estimation is first performed to estimate variance / standard deviation of noise. Then, motion estimation is applied to obtain motion vectors that represent the relative motion between the pixels of the noisy current input picture frame and the corresponding pixels of the previous noise attenuated picture frame. Global motion estimation is performed to estimate the camera motion of the image sequence through the motion vector. When reliable global motion estimation is made, the overall gain of temporal filtering is reduced by adjusting the estimated noise level. Therefore, motion blur is prevented.

Global motion, noise reduction, motion vector, motion estimation, noise reduction

Description

Methods for adaptive noise reduction based on global motion estimation and video processing system therefore}

1 is a flowchart illustrating one embodiment of a method of temporal noise reduction based on global motion estimation.

2 is a functional block diagram of one embodiment of a time noise attenuation system based on global motion estimation.

3 (a)-(d) are examples of calculations of noise level adjustment parameters of the present invention.

 The present invention generally relates to image processing and, more particularly, to noise reduction of video sequences.

 In many video display systems, such as television receivers, video enhancement by noise attenuation is performed to obtain a noise-free video sequence that is essential in video display. Various noise reduction methods have been developed, but very few are used in real products because they introduce unnecessary artifacts into the frame. Most existing noise reduction methods use spatial (two-dimensional) noise reduction, temporal noise reduction, and three-dimensional (3D) noise reduction (ie, a combination of two-dimensional and temporal noise reduction). Can be divided into four categories.

Spatial noise reduction is filtering by dividing every pixel of the current image frame into a small local window. Such filters are usually treated as convolution filters based on kernels. Examples of such filters are a mean filter, a Gaussian filter, a median filter, and a sigma filter. The mean filter is the simplest and most intuitive way to soften the image and reduce the noise. The mean value of the small space window is calculated as the filtered result. In general, a 3X3 square kernel is used for a simple implementation. However, the mean filter causes severe image bleeding.

Gaussian filtering uses a "bell-shaped" kernel to remove noise. Gaussian filtering means weighted averaged execution of pixels within a small space window. However, Gaussian filtering also causes bleeding (the severity of the bleeding can be controlled by adjusting the standard deviation of the Gaussian distribution).

Median filtering is a nonlinear method. It lists the pixels in a small space window and takes their median as the filtered result. Median filters do not generate new unrealistic pixel values and maintain sharp boundaries. Also, the value of aliasing pixels will not affect the filtered result. However, as the number of input pixels increases, the cost of computation required to align the pixels becomes too expensive for a practical implementation.

To solve such a problem, several edge-oriented spatial filtering algorithms have been developed. However, these algorithms require expensive equipment and cause artifacts, especially when edge-detection fails in a noisy image. Another algorithm transforms the image on the frequency axis to attenuate high frequency components. Since the details of the image are also high frequency components, such methods also blur the screen.

The temporal noise attenuation first examines the motion information between the current image frames and their neighboring frames. The pixels are divided into a motion area and an empty area. In the blank area, the filter applies pixels of the current frame and neighboring frames along the time axis. In the motion domain, the time filter does not work to avoid motion blur. In general, temporal noise attenuation is better than spatial noise attenuation, keeping the details and boundaries. However, the performance of filtering depends on the accuracy of motion detection. If the motion detection fails, motion blur occurs. Such a disadvantage limits the application of time noise attenuation.

 Therefore, there is a need for a method and system for noise reduction to reduce blurring of motion.

In order to solve the above problem, the present invention aims to provide an improved temporal noise reduction method and system for estimating global motion and adjusting the overall gain of temporal filtering. .

According to one embodiment of the invention, the temporal noise attenuation is applied to the current input image frame which is noisy and the previous filtered image frame stored in the memory, two image frames. In the method of the present invention, noise estimation is first performed to estimate the noise variance and the standard deviation of the input image sequence. Then, motion estimation is performed to obtain a motion vector representing the relative motion between the pixels of the current frame that are noisy and the corresponding pixels of the previous noise attenuated frame. Global motion estimation is performed to estimate the camera motion of the image sequence in the motion vector. Once a reliable global motion vector is obtained, the overall gain of temporal fitering is reduced to adjust the estimated noise level. Thus, motion blur is prevented.

Hereinafter, various embodiments, features, and advantages of the present invention will be described in detail with reference to the accompanying drawings.

1 shows a step 10 of applying motion estimation to a noisy current input video frame and a previous noise attenuated frame; Obtaining (12) a motion vector representing a relative motion between the pixels of the noisy current input image frame and the pixels of the previous noise attenuated frame; Performing (14) global motion estimation using the motion vector to estimate camera motion of an image sequence; Adjusting a noise level to reduce blurring due to movement (16); And applying temporal noise attenuation using the estimated global motion to produce a noise attenuated output frame to the current input image frame and the previous attenuated frame that are noisy. Show an example flowchart. The above steps are applied repeatedly for the next incoming frame in the image sequence.

'를 't'시점에서의 상기 입력 영상 프레임이라 하고, 세로 좌표를 i, 가로 좌표

를 j라 할 때

를 좌표 (i,j)에 대응하는 픽셀값이라 한다. 입력 영상 시퀀스에 부가적, 독립적으로 동일하게 분포하는 잡음으로서 분산

를 갖는 영(zero) 평균 가우시안(Gaussian) 잡음이 발생한다고 가정하면 상기

픽셀은 다음과 같이 표시된다. 2 is an example of a motion-compensated temporal noise reduction system 100 implementing the above method according to an embodiment of the present invention in a functional block diagram. To systematically illustrate the time noise attenuation problem and the system embodiment of the present invention in FIG.

'Is the input image frame at the' t 'time, the vertical coordinate is i, the horizontal coordinate

When j is

Denotes a pixel value corresponding to the coordinate (i, j). Distributed as noise, additionally and independently equally distributed over the input image sequence

Assuming a zero mean Gaussian noise with

The pixels are represented as follows.

`,

`,

는 잡음의 방해가 없는 본래의 픽셀값을 나타내고

는 가우시안 분포의 잡음 성분을 나타낸다.here

Denotes the original pixel value without disturbing noise

Denotes the noise component of the Gaussian distribution.

The system 100 shown in FIG. 2 includes a noise estimation module 102, a motion estimation module 104, a global motion estimation module 106, and a noise level adjuster ( noise level adjustment midule 108, memory 110, and motion-compensated temporal filtering module 112.

의 픽셀들과 이에 대응하는 잡음 감쇄된 프레임

의 픽셀들 간의 상대적인 움직임을 나타내는 움직임 벡터들을 추정한다. 상기 움직임 벡터를 통해서 상기 전역 움직임 추정기(106)는 영상 시퀀스의 카메라 움직임을 추정한다. 만약 신뢰할 수 있는 전역 움직임이 추정되면, 상기 잡음 레벨 조정기(108)는 잡음의 분산/표준편차를 조정하여, 잡음 감쇄된 프레임

를 생성하는 상기 움직임 보상 시간 필터(112)에서 잡음 감쇄가 적게 일어나도록 한다. 따라서, 더 적은 움직임 번짐이 발생하게 된다. 상기 메모리(110)은 다음에 오는 프레임을 위해 이전의 잡음 감쇄된 프레임으로서 상기 잡음 감쇄된 프레임

를 저장한다. The noise estimator 102 estimates the variance / standard deviation of the noise of the input image sequence g. The motion estimator 104 is a noisy current frame

Pixels with corresponding noise attenuated frame

Estimate the motion vectors representing the relative motion between the pixels. Through the motion vectors, the global motion estimator 106 estimates the camera motion of the image sequence. If reliable global motion is estimated, the noise level adjuster 108 adjusts the variance / standard deviation of the noise to reduce the noise attenuated frame.

Less noise attenuation occurs in the motion compensation time filter 112 that generates. Thus, less motion blur occurs. The memory 110 is the noise attenuated frame as a previous noise attenuated frame for the next frame.

Save it.

를 추정한다. 본 명세서에 참조로서 삽입되며, 공통 양도된 미국 특허출원 번호 10/991,265 "영상 시퀀스로부터 잡음 분산을 추정하는 방법(Methods to estimate noise variance from a video sequence)"과 미국 특허출원 번호 11/100,369 "시간 잡음 추정의 신뢰도 추정(Reliability estimation of temporal noise estimation)", 미국 특허출원번호 11/187,528 "영상 시퀀스로부터 분산을 추정하는 확장된 방법(Extended method to estimate noise variance from a video sequence)" 은 이러한 잡음 추정기(102)의 예들을 제공한다. 참조문헌에 기재된 잡음 추정기(102)의 실시예에 의하면, 잡음 추정기(102)는 우선 잡음이 섞인 현재 프레임

와 이전의 잡음이 섞인 프레임의

을 비교하여 두 개의 연속적인 프레임 내 로컬 윈도우(local window) 상에서의 차이를 계산하여 이미지 구조를 제거한다. 그런 다음, 로컬 차의 분포도(distribution of local difference)로부터 잡음의 분산/표준 편차를 추정한다. 이 분야에서 통상의 기술을 가진 자는 로컬 차를 구하는 방법에 있어서 MAE(mean absolute error), MSE(mean square error), DLM(difference of local mean) 또는 다른 방법 적용과 이의 변형이 가능하다.The noise estimator 102 performs noise variance and noise standard deviation of an image sequence.

Estimate US Patent Application No. 10 / 991,265 "Methods to estimate noise variance from a video sequence" and US Patent Application No. 11 / 100,369 "times, incorporated herein by reference Reliability estimation of temporal noise estimation ", US patent application Ser. No. 11 / 187,528," Extended method to estimate noise variance from a video sequence " Examples of 102 are provided. In accordance with an embodiment of the noise estimator 102 described in the reference, the noise estimator 102 first determines the current frame of mixed noise.

And the previous noise of the frame

Compare and calculate the difference on the local window in two consecutive frames to remove the image structure. The variance / standard deviation of the noise is then estimated from the distribution of local difference. One of ordinary skill in the art can apply a mean absolute error (MAE), mean square error (MSE), difference of local mean (DLM) or other methods in the method of obtaining a local difference, and variations thereof.

의 픽셀들과 이에 대응하는 이전의 잡음 감쇄된 프레임

의 픽셀들 사이의 상대적인 움직임을 나타내는 움직임 벡터를 추정한다. 움직임 추정 방법과 관련하여 제한이 없다. 상기 움직임 추정의 하나의 예로서, 프레임 내의 정해진 검색 범위(search area)의 모든 로케이션(location)을 검색하고, 매칭 레지듀얼 에러(matching residual error)가 최소화되는 위치를 선택하는 풀서치 블록 매칭 알고리즘(full search block matching algorithm)이 있다. 이 분야에서 통상의 기술을 가진 자에게 알려진 바와 같이 세단계 서칭(three step searching), 다이아몬드 서칭(diamond searching), 계층적/다분해 움직임 추정(hierarchical/multi-resolution motion estimation), 오브젝트 베이스 움직임 추정(object-based motion estimation) 등과 같은 다른 움직임 추정 방법이 사용될 수도 있다.The motion estimator 104 is a noisy current frame

Pixels and corresponding previous noise canceled frame

Estimate a motion vector that represents the relative motion between the pixels of. There is no limitation with regard to the motion estimation method. As an example of the motion estimation, a full search block matching algorithm for searching all locations of a given search area in a frame and selecting a location where a matching residual error is minimized ( full search block matching algorithm). Three step searching, diamond searching, hierarchical / multi-resolution motion estimation, object-based motion estimation, as known to those skilled in the art. Other motion estimation methods, such as object-based motion estimation, may be used.

라 하고, 선택된 전역 움직임벡터의 카운트(count) 값을

라고 하자.

이 모든 움직임벡터의 세어진 횟수라고 할 때, 파라미터(parameter)

는

로 계산된다. 파라미터

가 클수록, 더 많은 블록들/픽셀들이 전역 움직임을 따르고, 상기 전역 움직임 벡터

는 더 신뢰할 수 있게 된다. 그런 다음 상기 전역 움직임 벡터

와 상기 파라미터

는 전역 움직임 추정기(106)에 의해 잡음 레벨 조정기(noise level adjustment module)(108)로 전달된다.The global motion estimation module 106 first calculates a histogram of the motion vector estimated by the motion estimator 104. The motion vector counted the most in the histogram is selected as the global motion vector. The global motion vector

The count value of the selected global motion vector

Let's say

Given the counted number of all motion vectors,

Is

Is calculated. parameter

The larger is, the more blocks / pixels follow the global motion and the global motion vector

Becomes more reliable. Then the global motion vector

And the above parameters

Is transmitted by the global motion estimator 106 to the noise level adjustment module 108.

를 검사한다. 만약 전역 움직임 벡터

가 영(zero)이라면, 카메라의 움직임이 없는 것을 나타내고 잡음 레벨 조정기(108)는 아무런 조정도 실행하지 않는다. 만약 상기 전역 움직임 벡터

가 영(zero)이 아니라면 추정된 잡음의 표준 편차

는

로 조정된다.The noise level adjuster 108 first determines the global motion vector.

Check it. If global motion vector

If is zero, it indicates no camera movement and noise level adjuster 108 makes no adjustments. If the global motion vector

If is not zero, the standard deviation of the estimated noise

Is

Is adjusted.

의 예를 도시한다.

과

를

을 만족하는 두 임계치라고 한다. 만약

이면 검출된 카메라의 움직임은 신뢰할 수 없는 것이고, 화면에 전역 움직임(gloabal motion)은 없는 것으로 추정된다. 따라서

가 1.0으로 설정되고, 아무런 조정도 실행되지 않는다. 만약

이면 검출된 카메라 움직임은 매우 신뢰할 수 있는 것이고,

는

을 만족하는 상수

로 설정된다. 만약

이면,

는 도 3(a)에 도시된 바와 같이

에 대하여 선형적인(linearly) 값을 가질 수 있고, 도 3 (b)-(d)에 도시된 바와 같이 비선형적으로 보간된(non-linearly interpolated) 값을 가질 수도 있다.

에 따라 단순 감소하기만 한다면 (monotonically decreasing)

가 보간되는 방법에 대한 제한은 없다.3 (a)-(d) show an adjustment parameter

Shows an example.

and

To

Are called two thresholds. if

It is assumed that the detected camera movement is unreliable, and there is no global motion on the screen. therefore

Is set to 1.0, and no adjustment is made. if

The camera movement detected is very reliable

Is

Constant that satisfies

Is set to. if

If,

As shown in Fig. 3 (a)

It may have a linear value with respect to and may have a non-linearly interpolated value as shown in FIGS. 3 (b)-(d).

Monotonically decreasing

There is no restriction on how it is interpolated.

)은 움직임 보상 시간 필터(motion-compensated temporal filtering module)(112)에 공급된다. 조정된 잡음 레벨

에 근거하여, 움직임 보상 시간 필터(112)는 잡음 감쇄된 프레임

를 얻기 위해 현재 프레임

와 이전의 필터링 된 프레임

에 움직임 보상 시간 필터링을 수행한다. 본 명세서에 참조로서 삽입되며, 공통 양도된 미국 특허출원 출원번호 11/106,998 "움직임 보상 시간 잡음 감쇄 방법(A method of motion compensated temporal noise reduction"과 2004년 미국 특허출원 출원번호 11/025,173 "비디오 시퀀스에서의 시간 잡음 감쇄 방법(A method of temporal noise reduction in video sequence)"은 이러한 움직임 보상 시간 필터(112)의 예들을 제공한다. 참조문헌에 기재된 움직임 보상 시간 필터(112)의 실시예에 따르면, 움직임 보상 시간 필터(112)는

로 표시되는 현재의 잡음이 있는 프레임 및

로 표시되는 메모리에서 읽혀지는 이전의 필터된 프레임 상에 시간축을 따라 필터링하는데, 이때 움직임 벡터를 이용하여, 비-움직임 영역에 있는 픽셀들은 시간축을 따라 필터링되는 반면, 움직임 영역에 있는 픽셀들은 필터링되지 않는다. 또한, 또다른 실시예에 따르면, 비-움직임 영역에서의 고품질의 필터링된 프레임과 움직임 영역에서의 필터링을 스위치 오프(switching off)함으로써 야기되는 필터링되지 않는 프레임과의 비대칭성을 해결하고자, 잡음이 섞인 현재 프레임의 픽셀들과 이전의 잡음 감쇠된 프레임에 있어서 대응하는 픽셀들의 가중치 평균을 계산함으로써 움직임 보상 시간 필터링을 할 수 있다. 상기 기재한 잡음 감쇄를 잡음-적응성(noise-adaptive)이라 한다. 만약 검출된 잡음 레벨이 더 높으면, 더 많은 잡음 감쇄가 수행된다. 만약 상기 전역 움직임 벡터가 신뢰할 수 있는 것으로 추정되고, 잡음 레벨이 잡음 레벨 조정기(108)에서 감소된다면, 더 적은 잡음 감쇄가 움직임 번짐을 방지하기 위해 수행될 것이다. Adjusted noise level determined by the noise level adjuster 108 (i.e., standard deviation of the estimated estimated noise)

Is supplied to a motion-compensated temporal filtering module 112. Adjusted Noise Level

Based on the motion compensation time filter 112, the noise canceled frame

Current frame to get

Previous filtered frame with

Perform motion compensation time filtering. Commonly assigned US patent application Ser. No. 11 / 106,998 "A method of motion compensated temporal noise reduction" and 2004 US patent application Ser. No. 11 / 025,173 "Video sequences, incorporated herein by reference. A method of temporal noise reduction in video sequence "provides examples of such motion compensation time filter 112. According to an embodiment of the motion compensation time filter 112 described in the reference, The motion compensation time filter 112

The current noisy frame indicated by

Filters along the time axis on a previous filtered frame that is read from the memory, denoted with a motion vector, where pixels in the non-motion region are filtered along the time axis, while pixels in the motion region are not filtered. Do not. Further, according to another embodiment, to solve the asymmetry between the high quality filtered frame in the non-motion region and the unfiltered frame caused by switching off the filtering in the motion region, Motion compensation time filtering can be performed by calculating the weighted average of the pixels of the current frame that are mixed and the corresponding pixels in the previous noise attenuated frame. The noise attenuation described above is called noise-adaptive. If the detected noise level is higher, more noise attenuation is performed. If the global motion vector is assumed to be reliable and the noise level is reduced in the noise level adjuster 108, less noise attenuation will be performed to prevent motion blur.

One of ordinary skill in the art will appreciate that the present invention can be used for both progressive and interlaced videos. Odd and even regions of an interlaced image can be processed into two separate progressive video sequences; Alternatively, two regions may be merged into a single frame before being processed.

The best embodiments have been disclosed in the drawings and specification above; However, other embodiments are possible, and the spirit or scope of the claimed invention should not be limited to the details of the optimal embodiments described above.

In one embodiment, the present invention provides an improved time noise reduction method and system by estimating global motion and adjusting the overall gain of time filtering.

In one embodiment of the present invention, temporal noise attenuation is applied to two image frames: the current frame with noise and the previous filtered frame stored in memory. In this method, noise estimation is first performed to estimate the noise variance / standard deviation of the input image sequence. Then, motion estimation is applied to obtain a motion vector that represents the relative motion between the pixels of the noisy current frame and the corresponding pixels of the previous noise attenuated frame. Global motion estimation is applied to estimate the camera motion of the image sequence from the motion vector. Once a reliable global motion is obtained, the overall gain of temporal filtering is adjusted to reduce the estimated noise level. Movement blur is prevented in this way.

Claims (25)

In the video noise reduction method of the video frame sequence, Motion estimation on the current noisy image frame and the previous noise attenuated frame to obtain a motion vector representing the relative motion between the pixels of the noisy current frame and corresponding pixels of the previous noisy frame. Executing; Performing global motion estimation to estimate a global motion vector of an image sequence using the motion vector; And determining a noise attenuated frame output based on the global motion vector. The method of claim 1, Noise-Attenuated Picture Frames

Noisy current image frame using the estimated global motion vector to determine output

And previous noise canceled picture frames

And performing a temporal noise attenuation at the video noise reduction method. The method of claim 2, Determining whether the global motion vector is reliable, and if so, further comprising reducing the overall gain of the temporal filtering by reducing estimated motion blur by adjusting the estimated noise level. . The method of claim 1, wherein performing global motion estimation comprises: Calculating a histogram of the motion vector; And estimating a global motion vector of the image sequence by using the histogram. 5. The method of claim 4, wherein performing global motion estimation And selecting the motion vector having the maximum count value in the histogram as the global motion vector. 6. The method of claim 5, wherein performing global motion estimation The global motion vector selected from the histogram

Count value of

And the count value of the entire motion vector

Is called a parameter

To

Further comprising calculating the parameter

The larger is, the more pixels follow the global motion and the global motion vector

Means that is more reliable. The method of claim 6, Parameter to reduce motion blur

And performing a noise level adjustment based on the video noise reduction method. The method of claim 7, wherein The parameter

Determining whether the global motion vector is reliable based on the video signal, and if so, adjusting the estimated noise level using an adjustment parameter to reduce the overall gain of temporal filtering to reduce motion blur. Attenuation method. 9. The method of claim 8, wherein the global motion vector

If is zero, it means that there is no camera movement and thus no adjustment is further performed. The method of claim 9,

Is the standard deviation of the adjusted estimated noise,

Is an adjustment parameter, the global motion vector

If is not zero, the standard deviation of the estimated noise

To

The video noise reduction method further comprises the step of adjusting by. The method of claim 10,

and

end

When the threshold meets if

If, then the detected global motion vector indicates unreliable, so that no adjustment is performed.

Setting to 1; if

Then the detected global motion vector represents very reliable,

To

Predefined

Setting to; if

If,

Using the simple reduction function of

An adjustment parameter comprising interpolating

The image noise reduction method further comprises the step of determining. An image processing system for image noise reduction of an image frame sequence, A noise estimator for estimating a noise level of an input image sequence g; The pixels of the current frame g that are noisy and the previous noise attenuated frame corresponding thereto.

A motion estimator for estimating a motion vector representing the relative motion between the pixels of the apparatus; A global motion estimator which estimates global motion of the image sequence using the motion vector and estimates reliability of the global motion vector; A noise level adjuster for adjusting a noise level based on the reliability of the global motion vector; And a motion compensation time filter for filtering said current input video frame with noise by said adjusted noise level. The method of claim 12, The current input image frame in which the noise is mixed using the estimated global motion

And the previous noise canceled image frame

A noise attenuated frame by performing time noise attenuation on the

And a noise attenuator for determining the output. The method of claim 13, Means for determining if the estimated global motion is reliable, and if so, adjusting the estimated noise level to reduce the overall gain of the temporal filtering to reduce motion blur. 13. The system of claim 12, wherein the global motion estimator And obtaining a histogram of the motion vector, and estimating a global motion vector of the image sequence based on the histogram. 16. The system of claim 15, wherein the global motion estimator And use the histogram to select a motion vector having a maximum count value in the histogram as a global motion vector. 17. The system of claim 16, wherein the global motion estimator The global motion vector selected from the histogram

Count value of

And the count value of the entire motion vector

Is called a parameter

To

And calculate the

The larger is, the more pixels follow the global motion and the global motion vector

Image processing means that is more reliable. 18. The apparatus of claim 17, wherein the noise level adjuster Parameter to reduce motion blur

And perform a noise level adjustment on the basis of. 19. The apparatus of claim 18, wherein the noise level adjuster The parameter

Determine if a global motion vector is reliable, and if so, adjust the estimated noise level using an adjustment parameter to reduce motion blur by reducing the overall gain of temporal filtering. 20. The apparatus of claim 19, wherein the noise level adjuster If the global motion vector

If 0, it means that there is no camera movement and therefore no adjustment is performed. 21. The apparatus of claim 20, wherein the noise level adjuster

Is the standard deviation of the adjusted estimated noise,

Is an adjustment parameter, the global motion vector

If is not zero, the standard deviation of the noise

To

The image processing system, characterized in that for adjusting. The method of claim 21,

and

end

When the threshold is satisfied, the noise level adjuster if

If, then the detected global motion vector indicates unreliable, so that no adjustment is performed.

Set to 1; if

Then the detected global motion vector represents very reliable,

To

Predefined

Set to; if

If,

Using the simple reduction function of

To adjust the adjustment parameters

Determining an image processing system. In the video noise reduction method of the video frame sequence, Estimating a noise level in the image sequence; Motion estimation is performed on the noisy current image frame and the previous noise attenuated frame to obtain a motion vector representing the relative motion between the pixels of the current noisy frame and the corresponding pixels of the previous noisy frame. Making a step; Estimating a global motion vector of an image sequence using the motion vector and calculating a reliability of a global motion vector; Adjusting a noise level based on the reliability of the global motion vector; Performing motion compensation time filtering using the adjusted noise level to produce a noise attenuated frame. The method of claim 23, wherein

and

end

When the threshold satisfies the noise level adjustment step, if

If it is, the detected camera movement is unreliable, and since there is no global movement of the screen, no adjustment is performed.

Set to 1; if

Then the detected global motion vector represents very reliable,

To

Predefined

Set to; if

If,

Using the simple reduction function of

An adjustment parameter comprising interpolating

Determining the image noise reduction method. The method of claim 20,

and

end

When the threshold is satisfied, the noise level adjuster if

If it is, the detected camera movement is unreliable, and since there is no global movement of the screen, no adjustment is performed.

Set to 1; if

Then the detected global motion vector represents very reliable,

Is

Predefined to satisfy

Set to; if

If,

Using the simple reduction function of

To adjust the adjustment parameters

Determining an image processing system.

Images