KR100586996B1 - Noise Level Estimate Apparatus - Google Patents

Abstract

   The present invention provides a delay unit for delaying image data, a low pass filter for passing only low frequency components of input first image data and second image data output from the delay unit, and the first image data and the second image. The present invention relates to a noise level estimating apparatus including a noise level calculator for calculating a noise level of an image based on image data. The noise level estimating apparatus of the present invention compares the windows of the first image data and the second image data output from the low pass filter, and calculates a center window difference value and a peripheral window minimum difference value and a peripheral value of each of the plurality of first windows. And a motion detector for calculating a window average difference value and outputting weights of the first windows to the noise level calculator. As a result, a noise level estimating apparatus capable of effectively and accurately estimating a noise level of an input image is provided.

Description

       Noise Level Estimate Apparatus

1 is an overall control block diagram of a noise level estimation apparatus according to the present invention;

2 is a control flowchart of a motion detector of a noise level estimating apparatus according to the present invention;

3 is an exemplary diagram of SAD (-1, -1) of a SAD function unit according to the present invention;

4 is a flowchart for calculating a weight of a weight calculator according to the present invention;

5 is a control block diagram of a noise level calculator according to the present invention.

* Explanation of symbols for the main parts of the drawings

10: delay unit 20: low pass filter (LPF)

30: motion detection unit 33: alignment circuit unit

35: weight calculation unit 40: SAD function unit

50: noise level calculator

The present invention relates to a noise level estimating apparatus, and more particularly, to a noise level estimating apparatus capable of effectively and accurately estimating a noise level of an input image.

In general, in a digital display device, image processing technology is an important core technology that determines the competitiveness of a product. The image processing technology of such a digital display device is a technique of increasing the sharpness of an image or removing noise present in an image by increasing a system of a limited bandwidth display image or a blurred display image.

However, all image processing technologies have unwanted side effects, and the image processing technology aims to obtain a desired image while minimizing such side effects.

The fundamental cause of the side effects in the image processing technology is due to noise transmitted undesirably on the air transmission or image processing circuit.

In this case, if the image processing technique is applied to an image to which unwanted noise is applied as described above, the noise component also becomes more prominent as the system becomes larger, rather than the original image without the image processing technique. You may get it.

Therefore, by determining the noise level of the input image to determine whether the image has a lot of noise components, and selectively applying the image processing technology only to the image having a small noise component, the maximum image processing effect can be obtained while minimizing the side effects as described above. Can be.

Accordingly, it is an object of the present invention to provide a noise level estimating apparatus that can effectively and accurately estimate the noise level of an input image.

The object is a delay unit for delaying the image data, a low pass filter for passing only low frequency components of the first image data input and the second image data output from the delay unit, and the first image data and the second image. A noise level estimating apparatus including a noise level calculating unit for calculating a noise level of an image based on image data, the noise level estimating unit comprising: comparing windows of first image data and second image data output from the low pass filter, And a motion detector configured to calculate a center window difference value, a peripheral window minimum difference value, and a peripheral window average difference value for the first window, and output the weights of the first windows to the noise level calculator. Achieved by a level estimator.

Here, the motion detector compares the first window of the low pass first image data and the windows of the low pass second image data, and compares the center window difference value and the peripheral window with respect to the center pixel of each first window. A SAD function unit for outputting difference values, an alignment circuit unit for calculating peripheral window difference values from the SAD function unit, and outputting a peripheral window minimum difference value and a peripheral window average difference value for each center pixel, and a preset background Receiving a center reference value and outputting each center pixel based on a center window difference value, a peripheral window minimum difference value, a peripheral window average difference value, and the background area reference value of each center pixel output from the SAD function unit and the alignment circuit unit; It is preferable to include a weight calculator for outputting a weight corresponding to the.

Here, the weight calculation unit outputs a weight of a predetermined α value when the central window difference value corresponding to the center pixel is smaller than the corresponding peripheral window minimum difference value by a predetermined size or more, and is smaller than the peripheral window minimum difference value. If it is not smaller than the size, it is determined whether the center pixel belongs to the background area, and if it does not belong to the background area, the weight of the β value smaller than the α value is output. If the center pixel belongs to the background area, the difference of the center window is different. As the value increases, it is preferable to output a predetermined weight closer to the β value in the α value.

Here, the noise level calculator calculates a noise standard deviation value by comparing the first image data with the second image data output from the delay unit, and outputs the noise standard deviation value and the respective centers output from the motion detector. It is preferable to calculate the noise level based on the weight corresponding to the pixel.

Here, the α value is 1, and the β value is preferably 0.

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings. 1 is an overall control block diagram of a noise level estimation apparatus according to the present invention. As shown in the figure, the noise level estimating apparatus according to the present invention includes a delay unit 10 for outputting second image data Y ₋₁ by delaying the input image data by one frame and a first input current. A low pass filter (LPF) 20 which removes high frequency components of the image data Y ₀ and the second image data Y ₋₁ and passes only low frequency components, and a first image output from the LPF 20. A motion detector 30 which detects the movement of the pixel and outputs the weight of the pixel by comparing the pixel information of the data LPY ₀ and the delayed second image data LPY- ₁ and the first image; The noise standard deviation value is calculated by comparing the data Y ₀ and the second image data Y ₋₁ , and based on the noise standard deviation value and the weight of the corresponding pixel output from the motion detection unit 30. Noise level calculator 50 for calculating the noise level is included.

Here, the delay unit 10, the LPF 20, and the motion detection unit 30 process the image data in pixel units and output them in pixel units, and the noise level calculator 50 processes pixel units and outputs them in frame units. do.

The delay unit 10 stores the input image data and delays one frame after comparing the data of the previous frame of the input image data with the frame data of the first input image data Y ₀ . Output image data (Y- ₁ ). Here, the second image data Y _-1 is delayed by one frame, and the frame data is the same as the image data.

The LPF 20 removes the high frequency component of the currently input first image data Y ₀ and passes only the low frequency component, and removes the high frequency component of the delayed second image data Y ₋₁ . LPF 24 which passes only low frequency components.

The motion detector 30 compares the pixel information of the current first input image LPY ₀ output from the LPF 20 with the delayed second image data LPY ₋₁ to determine whether the corresponding pixel moves. It detects and outputs the weight of the pixel.

Here, the motion detector 30 will be described in detail with reference to FIG. 2 as follows. The motion detector 30 may include pixels of each of the plurality of first windows including peripheral pixels around a predetermined number of center pixels in the first image data LPY ₀ and second image data LPY ₋₁ . A center window difference value S _C comparing pixels of each center window corresponding to each of the plurality of first windows, pixels of each of the plurality of first windows, and each peripheral pixel corresponding to each of the first windows To compare the pixels of the peripheral windows corresponding to each of the first windows among the peripheral windows having the same size as the first window, and compare the peripheral window differences corresponding to the respective center pixels (S _{-1 and -1).} SAD function unit 40 for outputting _{˜S 1,1} and peripheral window difference values S _{−1, -1} to S _1,1 corresponding to each center pixel output from SAD function unit 40. an operation to close the window the minimum difference value (S _MIN) and around the window that corresponds to the maximum of each of the center pixels This value (S _MAX) and the surrounding window average differential value (S _AVE) alignment circuit 33 which outputs the group receiving a set background area threshold (Thres _M), for each center pixel that is output from the SAD function part 40 The corresponding center window difference value (S _C ) corresponding to the minimum window difference value (S _MIN ), the peripheral window average difference value (S _AVE ), and the background area reference value (Thres _M ) of the corresponding center pixel are output from the alignment circuit unit 33. The weight calculation unit 35 calculates and outputs a weight corresponding to each center pixel.

Here, the SAD function unit 40 calculates the center window difference value S _C and the peripheral window difference values S _{-1, -1} to S _1,1 corresponding to each center pixel through Equation 1 as follows. Output

(Formula 1)

Where f ' _C (x, y) = the pixel value passing through the LPF 22 at the (x, y) position of the current frame,

f ' _P (x, y) = pixel value passed through LPF 22 at position (x, y) of the previous frame,

wx = window size in the x-axis direction

wy = window size in the y-axis direction.

In the SAD function unit 40, the center window difference value S _C and the peripheral window difference values S _{-1, -1} to S _1,1 of the first window having the center pixel (x, y). A case of outputting the (S _{-1, -1} ) value will be described with reference to FIG. 3.

3a of Figure 3 is the first central pixel in the image data _{(LPY 0) (x, y} ) , and the first window size, wx in the x-axis direction as a 3 × 3 and wy in the y-axis direction (w ₀₎ Indicates. In addition, 3b of FIG. 3 shows a center window (w ₁ ) having a center pixel of (x, y) in the second image data LPY _-1 , a size of 3 × 3, wx in the x-axis direction and wy in the y-axis direction. And a peripheral window (w _{-1, -1} ) among the peripheral windows centering on each peripheral pixel corresponding to the first window (w ₀ ). Here, the first window w ₀ and the center window w ₁ represent the same size window at the same position.

Thus, SAD function unit 40 includes a central window (w ₁₎ corresponding to the first window (w ₀₎ in the present first window (w ₀₎ and a previous frame of the frame in the SAD (0,0), (41) Outputs the center window difference (S _C ), which is a comparison value of. In addition, the peripheral pixels (x-wx / a, y-wy) corresponding to the first window (w ₀ ) of the current frame and the first window (w ₀ ) of the previous frame in the SAD (-1, -1) 42. / b) Outputs the peripheral window difference value (S _{-1, -1} ) _, which is a comparison value of the peripheral window (w _{-1, -1} ) with respect to the center. Here, in FIG. 3, only the center window w ₁ and the peripheral windows w _{-1, -1 are} shown, but the remaining peripheral window difference values of each peripheral window corresponding to the first window w ₀ are similarly. Outputs (S _{0, -1} to S _1,1 ).

Therefore, if the first window w ₀ of the center pixel (x, y) is not moved, a small value of the center window difference value S _C is output.

In this manner, the SAD function unit 40 has respective center window difference values S _C corresponding to each center pixel of the first image data LPY ₀ and the second image data LPY ₋₁ and each peripheral window. The difference values S _{-1, -1} to S _1,1 are output.

The alignment circuit unit 33 calculates peripheral window difference values S _{-1, -1} to S _{1 , 1} corresponding to each center pixel output from the SAD function unit 40, and then surrounds the peripheral window corresponding to each center pixel. Outputs the minimum difference value (S _MIN ), the maximum difference value of the surrounding window (S _MAX ), and the average difference value of the surrounding window (S _AVE ).

The weight calculator 35 receives a preset background region reference value Thr _M that distinguishes the background region from the non-background region, and the center window difference value corresponding to each center pixel output from the SAD function unit 40. (S _C ) based on the peripheral window minimum difference value S _MIN , the peripheral window average difference value S _AVE , and the background area reference value Thres _M of the corresponding center pixel outputted from the alignment circuit unit 33. It determines whether the pixel moves and outputs the weight corresponding thereto.

Here, the process of calculating the weight of each center pixel by the weight calculator 35 will be described with reference to FIG. 4.

First, the weight calculator 35 determines whether the center window difference value S _C corresponding to each center pixel is smaller than the minimum window difference value S _MIN of the corresponding center pixel by a predetermined size or more. Here, the center window difference value (S _C) is close to the window at least the difference value of the center pixel (S _MIN) to a sufficiently small not whether more in one embodiment of the determination criterion, the center window difference value (S _C) is close to the window at least If the difference value (S _MIN n) x n is smaller than the difference between the peripheral window average difference values (S _AVE ) of the corresponding center pixel (S _C <S _MIN × nS _AVE ), the first window of the corresponding center pixel After determining that there is no motion, a predetermined value of α (= 1) is output as a weight of the corresponding center pixel (S10). Here, when it is determined that there is a movement in the corresponding first window (S _C ≥ S _MIN x n-S _AVE ), it is determined whether the corresponding first window belongs to the background area and does not belong to the background area (│ S _MIN -S _{AVE |} ≥ The background area reference value Thr _M is outputted with a β (= 0) value smaller than the α (= 1) value by weight (S20). If the corresponding first window belongs to the background area (│ S _MIN -S _AVE │ <background area reference value (Thres _M )), the greater the central window difference value (S _C ) of the corresponding center pixel, the greater the risk of motion determination. Is gradually increased to output a weight close to the β (= 0) value from the α (= 1) value. That is, when S _C <S _MIN + Thres _M when the corresponding first window belongs to the background area, α (= 1) is output as a weight of the corresponding center pixel (S30), and S _C ≥S _MIN + p If x Thres _M , γ is output as a weight of the corresponding center pixel (S40), and if S _C <S _MIN + q x Thres _M , δ is output as a weight of the corresponding center pixel (S50). In addition, if S _C <S _MIN + r × Thres _M when the corresponding first window belongs to the background area, ε is output as a weight of the corresponding center pixel (S60).

Herein, the relationship between α, β, γ, δ, ε is α (= 1)> γ> δ> ε. > (beta) = 0, and the relationship of p, q, and r is 1 <p <q <r ... to be.

The noise level calculator 50 calculates a noise standard deviation value by comparing the first image data Y ₀ and the second image data Y ₋₁ , and the noise standard deviation value and the motion detection unit 30. The noise level is calculated based on the weight of the corresponding center pixel. The noise level calculator 50 will be described in detail with reference to FIG. 5 as follows.

The noise level calculator 50 accumulates the weight corresponding to each center pixel output from the weight calculator 35 in one frame unit and the plurality of weights in the first image data Y ₀ . An arithmetic unit 52 for outputting a difference value Y _diff of pixels of each window corresponding to each window from each window having a predetermined size centered on each center pixel of the second image data Y ₋₁ ; And a standard deviation calculator 53 for calculating a noise standard deviation value σ _N corresponding to each of the plurality of center pixels by using the difference value Y _diff output from the calculator 52. Here, each window of the first image data Y ₀ calculated by the calculating unit 52 may be larger than the first window, and each corresponding window corresponding to each window of the second image data Y ₋₁ may be formed as the first window. It is preferable that they are the same size at the same position as each window of one video data Y ₀ .

Here, the operation of the standard deviation calculator 53 is expressed by Equation 2 as follows.

(수식2)　

(Formula 2)

Where Y _diff (x, y) = pixel difference between the current frame and the previous frame at (x, y),

wx '= window size in the x-axis direction,

wy '= window size in the y-axis direction.

In addition, the noise level calculator 50 calculates the noise standard deviation value σ _N corresponding to each center pixel output from the standard deviation calculator 53 and the weight of the corresponding center pixel output from the weight calculator 35. A multiplier 55 multiplying by a noise accumulator 57 that accumulates the standard deviation multiplication value of the center pixels output from the multiplier 55 in a unit of frame, and a noise output from the noise accumulator 57 And a division unit 59 for dividing the accumulated value by the weight accumulation value output from the weight accumulation unit 51 and outputting a noise level value.

Through such a configuration, the apparatus for estimating noise level according to the present invention further determines the movement of each center pixel of the input image in detail, and accordingly outputs a weighted value corresponding to the movement of each center pixel in stages, The noise level can be estimated more accurately based on the weights output in stages.

As described above, according to the present invention, it is possible to provide a noise level estimating apparatus capable of more effectively estimating a noise level of an input image.

Claims (5)

A delay unit for delaying the image data, a low pass filter for passing only low frequency components of the first image data and the second image data output from the delay unit, and an image based on the first image data and the second image data. A noise level estimating apparatus comprising a noise level calculating section for calculating a noise level of Comparing the windows of the first image data and the second image data output from the low pass filter, and calculates the center window difference value, the peripheral window minimum difference value and the peripheral window average difference value for each of the plurality of first windows And a motion detector for outputting a weight of each first window to the noise level calculator. The method of claim 1, The motion detection unit, SAD function that compares each first window of the low pass first image data and the windows of the low pass second image data, and outputs the center window difference value and the peripheral window difference value for the center pixel of each first window. Wealth, An alignment circuit unit for calculating peripheral window difference values from the SAD function unit and outputting peripheral window minimum difference values and peripheral window average difference values for the respective center pixels; Receiving a predetermined background area reference value and based on the center window difference value, the peripheral window minimum difference value, the peripheral window average difference value, and the background area reference value of each of the center pixels output from the SAD function unit and the alignment circuit unit; And a weight calculator for outputting a weight corresponding to each center pixel. The method of claim 2, The weight calculation unit, If the central window difference value corresponding to the center pixel is smaller than the corresponding peripheral window minimum difference value by more than a predetermined size, a weight of a predetermined α value is output, and if the central window difference value is not smaller than the predetermined minimum window difference value by more than a predetermined size, the center It is determined whether the pixel belongs to the background area, and if it does not belong to the background area, the weight of β value smaller than the α value is output. If the center pixel belongs to the background area, the center window difference value is larger as the value of α. And outputting a predetermined weight close to the β value. The method of claim 3, The noise level calculator,    The noise standard deviation value is calculated by comparing the first image data with the second image data output from the delay unit, and based on the noise standard deviation value and weights corresponding to the respective center pixels output from the motion detector. Noise level estimator, characterized in that for calculating the noise level. The method according to claim 3 or 4, Wherein the value of α is 1 and the value of β is zero.

Images