JPH10262160A - Signal-to-noise ratio (s/n)detector and noise reduction device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize noise reduction with less image quality deterioration without mis-detection of S/N, based on a local characteristic of an image. SOLUTION: A local S/N detection section 200 obtains 2-dimension S/N and 3-dimension S/N for each block of an input image subject to block processing by a block processing circuit 100 and uses the S/N with less noise in both the S/N ratios for a local S/N of a concerned block. A global S/N detection section 400 collects the local S/N as to an entire input image and uses a most frequent value as the global S/N. A threshold setting section 500 sets a threshold in response to an image of each block according to the S/N with less noise in the local S/N and the global S/N. A noise reduction section 600 replaces a pixel value of a processing object pixel with a mean pixel value of a pixel whose difference from a pixel value of the processing object pixel is a threshold or below set to a block to which the processing object pixel belongs among pixels in the vicinity of the processing object pixel in an input image.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an SN ratio detecting device and a noise reducing device, and more particularly, to an SN ratio detecting device capable of realizing noise reduction with little image quality deterioration without erroneously detecting an SN ratio due to local features of an image. The present invention relates to a noise reduction device.

[0002]

2. Description of the Related Art A conventional general image noise reduction apparatus is described in "Multi-Dimensional Signal Processing of TV Images", pp. 223-181. 1
89-190 and "Digital Television Technology" edited by Japan Broadcasting Corporation Broadcasting Research Institute pp. 268-270. These are methods of reducing noise by a LPF in the time direction using the fact that the video component in the input image has a frame correlation while the noise changes with time. However, if this method is performed, the image quality is degraded due to motion blur, and thus such noise reduction is not usually performed.
On the other hand, a method of separating a video component and a noise component by using non-linear signal processing and reducing the noise by means such as averaging or level compression has been studied and put to practical use.

For example, Harashima et al., “Ε-separated nonlinear digital filter” (Research IEICE '82 / 4 Vol. J65-A).
No. 4 pp. 297-304), Applicants'"Method of Reducing Noise in Television Video Signals"
-214192), “Noise reduction circuit” (Japanese Unexamined Patent Application Publication No.
-250264), "Noise reduction device" (Japanese Patent Laid-Open No.
-102871) and its improved type (Japanese Patent Laid-Open No.
No. 172368, Japanese Patent Application No. 7-322733, Japanese Patent Application No. 8-36281, Japanese Patent Application No. 8-36397, Japanese Patent Application No. 8
No. 36520, Japanese Patent Application No. 8-237093, Japanese Patent Application No. 8
No. 237,094, Japanese Patent Application No. 8-128472) and the like. Each of these methods has its own characteristics. Basically, a threshold value is externally provided, and noise (noise) and video components are separated according to the threshold value.

[0004] The present applicants have proposed a "SN value detection method for television video signals" (Japanese Patent Laid-Open Publication No. Hei 8-201464), and threshold control using a statistical test method (chi-square test method). Reported that can be optimized considerably.

[0005]

However, in the conventional noise reduction device, a person sets a threshold,
Or, the screen was set uniformly. For this reason, local image quality degradation may occur due to noise reduction due to partial pattern differences or movements. Furthermore, in the statistical SN value detection method for detecting the SN ratio from the screen, the operation speed is not enough to perform a local statistical test for each block at present, and real-time processing is difficult.

Therefore, the present invention has been made in view of the above-mentioned points, and has been developed by a simple method.
It is an object to provide a ratio detection device and a noise reduction device.

[0007]

In order to achieve the above object, the apparatus of the present invention divides an input image into a plurality of blocks and obtains a two-dimensional SN ratio and a three-dimensional SN ratio for each of the blocks. A local SN ratio detecting means for setting a value of a smaller noise among the SN ratios to a local SN ratio of the corresponding block; a global SN ratio for collecting the local SN ratio for the entire input image and setting a mode value to a global SN ratio
Ratio detection means, the local SN ratio and the global S
And an SN ratio selecting means for selecting a value of the N ratio having less noise for each block.

Here, the local SN ratio detecting means obtains the two-dimensional SN ratio for each block using the variance of the pixel values in the block as noise, and calculates the mean square of the time difference of each pixel in the block as noise. The three-dimensional SN ratio can be obtained as follows.

According to another aspect of the present invention, an input image is divided into a plurality of blocks, and a two-dimensional SN ratio and a three-dimensional SN ratio are obtained for each of the blocks.
A local S / N ratio detecting means for setting a value of the ratio having less noise to a local S / N ratio of the corresponding block; a global S / N ratio for summing up the local S / N ratio for the entire input image and setting a mode value to a global S / N ratio Detecting means;
Threshold setting means for setting a threshold corresponding to the image of each block in accordance with the value of the local SN ratio and the global SN ratio that has less noise; Averaging means for replacing a pixel value of the processing target pixel with an average pixel value of a pixel whose difference from the pixel value of the target pixel is equal to or less than a threshold value set in a block to which the processing target pixel belongs. It is characterized by.

[0010] Here, there is provided means for calculating the luminance of the image in the block for each of the divided blocks, and the threshold value setting means sets the threshold value to be larger when the luminance of the image in the block is lower than when it is high. It can also be corrected to a value.

Here, the averaging means includes means for detecting a two-dimensional inclination of an image including the processing target pixel, and calculating a pixel value of the processing target pixel and the inclination value to calculate the processing target pixel. Means for calculating a pixel value of which inclination has been corrected, and a difference between the pixel value and the inclination-corrected pixel value of the neighboring pixels is a pixel value equal to or less than a threshold value set in a block to which the processing target pixel belongs. The tilt-corrected pixel value may be replaced with the average pixel value of the pixel.

Here, the average pixel value is obtained from three-dimensional neighboring pixels of the pixel to be processed, and the neighboring pixels are at least a frame to which the pixel to be processed belongs and a frame which belongs to a frame preceding the frame. You can also.

A comparing unit that compares the replaced pixel value to be processed with the pixel value to be processed in the input image; and a pixel to be processed in which the pixel value to be processed in the input image is replaced by the comparing unit. If the value is larger than the value, a value obtained by subtracting the threshold value from the processing target pixel value of the input image is output.If the processing target pixel value of the input image is smaller than the replaced processing target pixel value, the threshold value is set. Selecting to output a value added to the processing target pixel value of the input image, and to output the processing target pixel value of the input image when the processing target pixel value of the input image is equal to the processing target pixel value of the output image; An addition / subtraction unit may be provided.

According to another aspect of the present invention, an input image is divided into a plurality of blocks, and a two-dimensional SN ratio and a three-dimensional SN ratio are obtained for each of the blocks.
A local S / N ratio detecting means for setting a value of the ratio having less noise to a local S / N ratio of the corresponding block; a global S / N ratio for summing up the local S / N ratio for the entire input image and setting a mode value to a global S / N ratio Detecting means;
Threshold setting means for setting a threshold corresponding to an image of each block in accordance with the value of the local SN ratio and the global SN ratio that has less noise, and noise reduction means for globally reducing and outputting the input image Comparing means for comparing the output image of the noise reducing means with the processing target pixel value of the input image; and the comparing means, wherein the processing target pixel value of the input image is larger than the processing target pixel value of the output image. Outputs a value obtained by subtracting the threshold from the processing target pixel value of the input image. If the processing target pixel value of the input image is smaller than the processing target pixel value of the output image, the threshold is set to the processing of the input image. A value added to the target pixel value is output. If the processing target pixel value of the input image is equal to the processing target pixel value of the output image, the processing target pixel value of the input image is output. Characterized by comprising a selection subtraction means.

Further, in order to achieve the above object, the apparatus of the present invention divides an input image into a plurality of blocks, obtains a two-dimensional SN ratio and a three-dimensional SN ratio for each of the blocks,
A local SN ratio detecting means for setting a value of the ratio having less noise as a local SN ratio of the corresponding block; and detecting a global SN ratio by comparing noise in the input image with a noise distribution according to a statistical distribution rule. Global S / N ratio detection means, threshold value setting means for setting a threshold value according to the image of each block in accordance with a value of the local S / N ratio and the global S / N ratio having less noise, and a threshold value of a processing target pixel in the input image. Among the neighboring pixels, the difference from the pixel value of the processing target pixel is an average pixel value of pixels having a pixel value equal to or less than a threshold value set in a block to which the processing target pixel belongs,
Averaging means for replacing a pixel value of the processing target pixel.

[0016]

Embodiments of the present invention will be described below in detail with reference to the drawings.

(First Embodiment) FIG. 1 is a block diagram showing a first embodiment of a noise reduction apparatus to which the present invention is applied.

The apparatus of the present invention generally includes a block processing circuit 100 for dividing an input image, a local SN ratio detecting unit 200 using two-dimensional variance of pixel values of the blocked image and an inter-frame difference, A global S / N ratio detection unit 400 for summing up the local S / N ratio to obtain a mode value, a threshold setting unit 500 for setting a threshold for each pixel for each block using the local S / N ratio and the global S / N ratio, and a neighborhood below the threshold value It comprises a noise reduction unit 600 that performs noise reduction by replacing a pixel to be processed with an average value of pixels.

First, the block processing circuit 100 divides an input image into blocks having a fixed size on the screen and forms blocks. Here, the input image is provided, for example, as a baseband television video signal, and one block has a size of 8 pixels × 8 pixels.

Next, the local SN ratio detecting section 200 calculates a two-dimensional SN ratio and a three-dimensional SN ratio for each block.

First, the two-dimensional SN ratio is obtained by calculating the arithmetic mean of the luminance (pixel value) of all pixels for each block, and calculating the square mean or variance of the difference from the pixel value of each pixel in the block.
It can be calculated by calculating the ratio between the maximum amplitude of the luminance signal and this root mean square.

In FIG. 1, the block average calculation circuit 21 calculates the arithmetic average of the pixel values of 8 pixels × 8 pixels = 64 pixels for each block, and the time required for the processing by the block average calculation circuit 21 by the block delay circuit 22. The difference between the delayed signal and the output signal of the block average calculation circuit 21 is obtained by the subtractor 23, and the difference is squared by the squaring circuit 24. And since the maximum amplitude of the luminance signal is constant,
The averaging circuit 25 calculates the sum of the squares of the difference (variance) for all the pixels in the block, multiplies it by (1/64) and averages it.
Only the dimensional noise is calculated. Here, the processing for obtaining the arithmetic mean for each block may be processing using a two-dimensional LPF.

The above-described two-dimensional SN ratio calculation processing can be represented by the following equations.

The arithmetic mean of each block is ave2d, and the pixel value (luminance signal level) of each pixel in the block is X (i,
j) (0 ≦ i <8, 0 ≦ j <8),

[0025]

(Equation 1)

From this, the sum of squares diff2d of the difference from the two-dimensional average for each pixel value is:

[0027]

(Equation 2)

From this, the two-dimensional SN ratio SN2d obtained from the two-dimensional average of the block is

[0029]

(Equation 3)

## EQU1 ## Here, Spp represents the maximum amplitude of the luminance signal.

Next, the three-dimensional SN ratio is calculated by calculating the square mean of the inter-frame difference of the pixel value of each of 8 pixels × 8 pixels = 64 pixels for each block between the current frame and the previous frame.
It can be calculated by calculating the ratio between the maximum amplitude of the luminance signal and this root mean square.

In FIG. 1, the output signal (current frame signal) of the block delay circuit 22 used for the calculation of the two-dimensional SN ratio is divided into two, and one is input to a frame delay circuit (FDL) 26 to be delayed by one frame ( (Previous frame signal), current frame signal and previous frame signal. The difference between the two signals is obtained by a subtractor 27, and this difference is squared by a squaring circuit 28. Also in this case, since the maximum amplitude of the luminance signal is constant, the averaging circuit 29 calculates the sum of the squares of the difference (variance) for all the pixels in the block, multiplies the result by (1/64), and averages it. Is calculated.

The above three-dimensional SN ratio calculation processing can be expressed by the following equations.

The sum of squares of the difference from the pixel value one frame before d
if3d is

[0035]

(Equation 4)

Here, X ^-1 (i, j) represents the pixel value of each pixel in the block one frame before X (i, j).

The three-dimensional SN ratio S obtained from the difference between frames
N3d is

[0038]

(Equation 5)

## EQU1 ##

Local SN for each block
The ratio (hereinafter referred to as the local SN ratio) is the maximum value (the value with less noise) of the two-dimensional SN ratio and the three-dimensional SN ratio.
And By definition, it is sufficient to output the minimum value of the two-dimensional noise level and the three-dimensional noise level. Therefore, in FIG. 1, the minimum value selection circuit 30 compares the noise levels and selects the minimum value of the root mean square. An example in which output is performed as local noise has been described.

Here, the local SN ratio SN of each block
As the block, the maximum value of SN2d and SN3d (whichever is greater, or any value if they are the same) is selected.

[0042]

## EQU00006 ## SNblock = MAX (SN2d, SN3d) Here, the reason for taking the maximum value of both as the local SN ratio is that the two-dimensional SN ratio SN2d is erroneously detected because the edge of the image is included in the block In the edge portion of the image, the SN ratio decreases, while the three-dimensional SN ratio SN
This is because 3d has an erroneous detection in the moving part of the image, and the SN ratio decreases in the moving part. Therefore, the two-dimensional SN ratio and 3
A correct S / N ratio cannot be obtained in a portion where both the dimensional S / N ratios are erroneously detected, but usually a correct S / N ratio can be expected by selecting the maximum value. Further, even in the case where both are erroneously detected, by selecting the maximum value and setting the SN ratio with the smaller degree of erroneous detection as the local SN ratio, it is possible to set a threshold with less image quality deterioration than in the case of only one.

Next, the local SN ratio SNblock for each block is totaled by the global SN ratio detection unit 400 over the entire screen, and the mode is defined as the overall SN ratio. This is called a global SN ratio. When noise reduction processing is performed based on a threshold value set using the global SN ratio obtained by the majority logic, a malfunction based on erroneous detection of a partial picture difference or movement of an image as noise is removed. In FIG. 1, attention is paid to the noise level, so that the intra-frame mode extraction circuit 40 finds the mode of local noise by majority logic to obtain global noise.
By determining the ratio between the maximum amplitude of the luminance signal and the global noise, the global SN ratio is determined.

Next, a specific threshold setting method using the detected global SN ratio and the previously detected local SN ratio will be described.

As a condition for setting the threshold value, it is necessary that the threshold value is large enough to reduce the noise (noise) in the pixel, but it is necessary that the threshold value is not made larger than necessary so as not to damage the video component. Therefore, in order to prevent image quality degradation and reduce noise, it is necessary to change the threshold value according to the local features of the image. Emphasis should be placed.

Therefore, the threshold value setting unit 500 sets the global SN ratio from the global SN ratio detecting unit 400 and the local SN ratio which is delayed and time-aligned by the frame delay circuit (FDL) 46 to the value with the smaller noise, That is, the maximum value (or the minimum value of the noise levels)
Is selected by the minimum noise selection circuit 50 and is used as a criterion for threshold setting by the threshold setting circuit 52. It is predicted that a local SN ratio will be erroneously detected on a screen locally having features such as edges and motions, but the global SN ratio will be larger than this local SN ratio (the global noise is smaller than the local noise). Therefore, by performing the threshold control in this way, it is possible to suppress a malfunction based on erroneously detecting the local SN ratio using this feature (partial pattern difference or movement such as an edge) as noise.

The threshold value setting circuit 52 is, for example, a value that is inversely proportional to the maximum value of the two SN ratios, that is, a value that is proportional to the minimum value (the smaller value) selected by the minimum noise selection circuit 50 of the two noises. , A threshold is set for each block. As described above, even for a screen in which setting the threshold value using only the local SN ratio increases the threshold value and blurs edges and the like due to erroneous detection,
When used in combination with the global SN ratio as described above, the threshold value does not increase, the edges can be preserved, motion blur can be prevented, and image quality degradation is small.

Conversely, when the local SN ratio is larger than the global SN ratio, that is, in a screen where the SN ratio is partially high, there are blocks that have a local SN ratio larger than the expected global SN ratio. Therefore, by selecting the local SN ratio by the minimum noise selection circuit 50, the threshold value is set to be smaller than the value expected for the global SN ratio. As a result, it is possible to avoid an unnecessary increase in the threshold value on a screen having a block with a small amount of noise, and the threshold value is increased only by setting the threshold value using only the global S / N ratio. This allows noise reduction with little degradation in image quality while preserving details with shallow modulation, which are buried in noise and blurred by averaging.

Thus, the local SN ratio and the global S
Only by setting the threshold value based on the maximum value of the N ratio, it is possible to prevent image quality deterioration while effectively reducing noise.
If the threshold does not increase with the increase in the selected SN ratio, a non-linear relationship may be established between the two.

Next, noise reduction based on the threshold set by the threshold setting unit 500 will be described.

In this embodiment, if the current frame to which the pixel to be processed of the input image belongs is the n-th frame, the (n
-1), n-th and (n + 1) th frames are used. However, without using the (n + 1) th frame, the (n−1) th,
Even if only the n-th frame is used, or if the (n-2) -th, (n-1) -th and n-th three frames are used, noise reduction is similarly performed from the current frame and the previous frame. Can be.

In this embodiment, in order to reduce the noise of the pixel to be processed, as shown in FIG. 2, the neighborhood of the pixel to be processed in the n-th frame (a three-dimensional neighborhood including not only a planar neighborhood but also a temporal neighborhood) The pixel replacement for noise reduction is performed using 27 pixels centered on the pixel to be processed within the predetermined range of (2). That is, the (n-1) -th, n-th,
Of the pixels of the (n + 1) th frame, the pixel to be processed of the nth frame and a total of 9 pixels adjacent thereto, 9 pixels at the same position of the (n-1) th frame + 9 pixels at the same position of the (n + 1) th frame Pixel = A total of 27 pixels, and among the neighboring pixels of the processing target pixel, the difference from the pixel value of the processing target pixel becomes a pixel value equal to or smaller than a threshold (th) set for the block to which the processing target pixel belongs. The noise is reduced by selecting the pixels, adding the pixel values, and dividing by the total number of the added pixels to replace the pixel value of the pixel to be processed with the average pixel value. Pixels whose difference exceeds the threshold value (th) have low correlation with the processing target pixel and may be an image edge or a moving image, and thus are not selected in the average value calculation for replacement and are regarded as invalid data. This is to prevent the characteristics of the local image from being impaired.

Returning to FIG. 1, the noise reduction unit 60
0, the input image is converted to a frame delay circuit (FDL) 6
Processing target pixel value extraction unit 6 from the signal delayed and output by 1
2, the central pixel value of 27 pixels is extracted. Further, the delay circuits 65 and 69 are delay circuits for time adjustment for delaying the time required for processing in the processing target pixel value extraction unit 62, and the frame delay circuit (FDL) 66 is a delay circuit for delaying one frame. From the input image, the (n-
1) The images of the nth and (n + 1) th frames are created.

Then, the central pixel value of the n-th frame, which is the extracted pixel to be processed, and the n-th, (n−1), and (n)
+1) Subtractor 6 subtracts the difference from each pixel value of the frame.
3, 67, 70. Based on this, the below-th pixel selection and addition circuit 64 selects and adds only pixel values whose difference from the pixel value to be processed in the pixels of the n-th frame is equal to or less than the threshold th from the threshold setting circuit 52 as valid data, and the addition result And the selected number of pixels are output. Similarly, in the pixel selection addition circuits 68 and 71 below th, the (n-1) -th and (n +)-th
1) Only the pixel value whose difference from the pixel value to be processed in the pixels of the frame is equal to or smaller than the threshold th from the threshold setting circuit 52 is selected and added as valid data, and the addition result and the number of selected pixels are output. The valid data is selected based on the idea that pixels having a difference equal to or less than the threshold th contain noise but have a correlation with the processing target pixel as an image, and are effective for noise reduction using the correlation.

Then, in the Σ / M processing circuit 72, each th
Hereinafter, an average value is calculated and output by dividing the total sum of the added outputs of the pixel selection and addition circuits by the total number M of selected pixels. That is, the calculated average value is replaced with the pixel value to be processed and output. By performing such pixel value replacement over the entire screen, averaging according to the characteristics of an image local to the processing target pixel, that is, low-pass filter processing is realized, and noise reduction with little image quality deterioration is realized. be able to.

(Second Embodiment) By simply performing a threshold process on the difference between the pixel to be processed and each neighboring pixel,
Correct processing cannot be performed on an inclined image with gradation. Reference (Harashima et al., “Ε-separated nonlinear digital filter”
(Research in IEICE '82 / 4 Vol. J65-A No. 4p
p. 297-304)), an inclination correction method using a least square error is proposed. However, it is difficult to obtain an approximate plane for each pixel in real time.

In the present embodiment, an image inclination correction method using a two-dimensional LPF output is realized as a simpler inclination correction method. First, taking a one-dimensional case as an example, LPF
The principle of the tilt correction method using is described with reference to FIG.

When a difference between a pixel to be processed and a neighboring pixel is simply obtained for a one-dimensional tilt image including noise as shown in FIG. 3A, the difference output includes a tilt component. I will. Therefore, the tilt image is converted to LP as shown in FIG.
F processing, and the level of the LPF output is shifted so that the difference between the pixel value to be processed and the LPF output becomes zero.
If the difference between the tilt image and the level-shifted LPF output is obtained as in (C), a tilt-corrected difference output can be obtained.

FIG. 4 is an overall block diagram of a second embodiment of a noise reduction device that has performed two-dimensional tilt correction.

In FIG. 4, the components other than the tilt correction unit 80 are the first
The delay circuits 65 and 69 are circuits for delaying the time required for the processing in the inclination correction unit 80 to adjust the time.

FIG. 5 is a detailed block diagram of the inclination correction unit 80. In FIG. 5, a processing target pixel value extraction unit 85 is the same as the processing target pixel value extraction unit 62, and a delay circuit 82 is a circuit for adjusting the time required for processing by the two-dimensional LPF 81.

Here, the (n−1) th, nth, (n +)
In the three frames of 1), it is considered that the degree of inclination of the image is the same, and in FIG. 5, the signal of the n-th frame is input to a two-dimensional LPF 81 composed of 3 pixels × 3 pixels, and the low frequency component of the image is detected. This low-frequency component output indicates the inclination of the image on a two-dimensional plane, and outputs for correcting inclination of these three frames are obtained. Then, the two-dimensional LPF 81 is calculated by the subtracter 83.
The difference between the output of the image and the delayed n-th frame image is 3 pixels ×
Output every three pixels.

The level shift circuit 84 focuses on the difference at the coordinate position of, for example, the center pixel, which is the pixel to be processed in the input image, and level shifts the output of the two-dimensional LPF 81 as a whole so that the difference becomes zero. I do. Then, the output pixel value of the processing target pixel value extraction unit 85 and the output value after the two-dimensional LPF processing that has been level-shifted by the level shift circuit 84 are calculated by the adder 86. Here, when the difference between the processing target pixel output from the inclination correction unit 80 and its neighboring pixels is obtained by the subtracters 63, 67, and 70 (see FIG. 4), the difference between the coordinates of each neighboring pixel of the processing target pixel is determined. The output of the tilt correction unit 80 obtained by adding the corresponding level shift output value and the value of the processing target pixel is subjected to a difference operation with each of the neighboring pixels, whereby the tilt included in the difference can be corrected.

As described above, according to the present embodiment, the threshold value is set for each block by the threshold value setting unit 500 based on the maximum values of the local SN ratio and the global SN ratio.
By replacing the pixel value after correcting the inclination of the gradation image with 0, the noise reduction unit 610 can effectively reduce noise even in the gradation image and prevent image quality deterioration.

(Third Embodiment) By the way, the easiness of visually recognizing noise also changes with the luminance (brightness) of an image. In general, noise is easier to visually recognize in darker parts than in bright parts, so if the threshold of dark parts with low luminance is set larger, the threshold of bright parts with higher luminance is corrected to be smaller, and more effective according to visual characteristics Noise reduction can be expected. Therefore, in the present embodiment, in addition to the maximum values of the local SN ratio and the global SN ratio, the local brightness (luminance) of the image is also set as a threshold condition. It was configured to have a large value.

FIG. 6 is an overall block diagram of a third embodiment of the noise reduction device having such a configuration.

The threshold value setting section 510 in FIG. 6 uses the signal obtained by delaying the output of the two-dimensional LPF 31 for calculating the block average by the frame delay circuit (FDL) 56 as a signal of local brightness, and corrects the threshold value. Is going. The two-dimensional LPF is used if the pixel values are used as they are.
Since the noise included in the pixel value affects the threshold correction,
This is to prevent this by averaging. Threshold setting circuit 5
For example, when the local brightness is equal to or higher than a predetermined level, the threshold value 3 is output after correcting the threshold value to a value smaller than the value set only by the maximum value of the SN ratio. If it is lower than the predetermined level, the output is corrected to a value larger than the value set only by the maximum value of the SN ratio. The correction may be continuously performed according to the output of the two-dimensional LPF 31.

As described above, according to the present embodiment, in addition to the maximum value of the S / N ratio, the threshold value is set by the threshold value setting unit 510 as a setting condition, and the gradation correction unit 80 sets the gradation image. By replacing the pixel value after correcting the inclination of the image, it is possible to prevent the image quality from deteriorating while effectively reducing noise even in the gradation image according to the local brightness by the noise reduction unit 610. .

(Fourth Embodiment) FIG. 7 shows a global S
FIG. 11 is an overall block diagram of a fourth embodiment of a noise reduction device configured to use a statistical method for N ratio detection.

The two-dimensional SN ratio and the three-dimensional SN ratio are detected by the local SN ratio detecting section 220 of FIG. 7 by statistical judgment on the difference output from the subtracters 23 and 27, and the global SN ratio is detected based on the detection result. The unit 410 detects a global SN ratio. Here, in the local SN ratio detection, it is difficult to perform the real-time processing of the statistical judgment in the block-by-block detection as in the above embodiment, but since the global SN ratio is almost constant in the sequence, it takes time. Even if the detection processing is performed, a slight deviation in time can be allowed, so that, for example, a chi-square test which is a statistical judgment can be used.

In the chi-square test circuit 32, the distribution of the difference output from the subtractor 23 is compared with the noise distribution of a statistical distribution rule prepared for SN ratio detection in advance to determine the significance (this noise distribution is a normal distribution And the chi-square test is used as a test for the normal distribution), and the SN ratio based on the difference between the blocks determined to be significant is adopted, and such examination is sequentially continued for a plurality of blocks corresponding to the entire screen, and the entire screen is displayed. The frequency of occurrence of the two-dimensional SN ratio adopted in the above is obtained, and an effective two-dimensional SN ratio is detected from the distribution of the frequency of occurrence. In this manner, erroneous detection is eliminated in a portion where a difference occurs due to motion in a block and in a portion where a difference occurs due to a two-dimensionally strongly correlated video component, and highly accurate two-dimensional SN ratio detection is performed. be able to.

The chi-square test circuit 33 similarly performs three-dimensional SN ratio detection on the difference output in the time direction. In the intra-frame mode extraction circuit 41, the mode of the two-dimensional SN ratio and the three-dimensional SN ratio is defined as the global SN ratio.

As in the present embodiment, even if it takes a long time only to calculate the global SN ratio, if statistical detection is introduced and highly accurate detection is performed, it is possible to prevent image quality deterioration and reduce noise. More effective.

(Fifth Embodiment) The local brightness (level) obtained as one of the threshold setting conditions as in the third embodiment is determined according to the characteristics of the local image. Change. Therefore, when the threshold value obtained in this way is used for controlling the shift amount of the dc shift type NR proposed by the present applicants, image quality degradation can be reduced as compared with the case of uniform shift control over the entire screen.

FIG. 8 is an overall block diagram of a fifth embodiment of the noise reduction apparatus in which the above-described concept of the threshold setting condition is applied to the dc shift type NR. In FIG. 8, only the noise reduction unit 620 is different from the third embodiment. Reference numeral 73 denotes a conventional noise reducer (for example, one that performs global noise reduction such as low-pass filter processing or temporal filter processing), and 74 denotes a delay circuit for time adjustment.
Reference numeral 75 denotes a dc shift circuit having the same configuration as that disclosed in Japanese Patent Application Laid-Open No. 7-250264 by the present applicants.

According to this publication, the dc shift circuit 75 is
"Comparison means for comparing the voltage value of the signal input to the first input terminal with the voltage value of the signal input to the second input terminal, and the voltage of the signal input to the first input terminal by the comparison means is When it is determined that the voltage is higher than the voltage of the signal input to the second input terminal, a signal obtained by subtracting the DC voltage input to the third input terminal from the signal input to the first input terminal is obtained. As an output signal, when it is determined that the voltage of the signal input to the first input terminal is smaller than the voltage of the signal input to the second input terminal, the signal is input to the third input terminal. A signal obtained by adding a DC voltage to a signal input to the first input terminal is set as an output signal,
When it is determined that the voltage value of the signal input to the input terminal of the second input terminal is equal to the voltage value of the signal input to the second input terminal, the addition and subtraction of the signal input to the first input terminal as the output signal And a means ”.

In the noise reduction section 620 of FIG. 8, the output signal of the delay circuit 74 for adjusting the time by delaying the time required for processing in the conventional noise reducer 73 is input to the first input terminal (original signal). Signal), the output signal of the conventional noise reducer 73 corresponds to the signal (reference signal) input to the second input terminal, and the output signal from the threshold setting circuit 53 is input to the third input terminal. DC voltage. Further, the voltage comparison circuit 76 corresponds to comparison means, and the selection addition / subtraction circuit 77 corresponds to selection addition / subtraction means.

The noise reduction by the dc shift circuit 75 is based on the following operation. First, a signal whose noise is globally reduced by the conventional noise reducer 73 of FIG. 8 is generated for a signal whose noise is to be reduced, and this signal is used as a reference signal of the dc shift circuit 75. The dc shift circuit 75 compares the original signal whose time is to be reduced with noise and this reference signal, and shifts the dc level of the signal whose noise is to be reduced for each pixel to approach the reference level, thereby obtaining a noise reduction effect. . Even if the reference signal is deteriorated in image quality due to the influence of the conventional noise reduction processing (for example, blurring of an edge in a low-pass filter, occurrence of an unnecessary edge in a median filter, or trailing motion in a temporal filter), the reference signal Since the signal is used only for the dc shift code control, the image quality deterioration of the noise reduction signal due to the influence of the reference signal image quality deterioration can be suppressed to about the dc shift amount.

Therefore, according to the present embodiment, the dc shift amount is controlled by the optimum threshold value for each block obtained by the present invention (the shift amount is small when the SN ratio is high, and the shift amount is low when the SN ratio is low). Is increased), the noise can be more effectively reduced by the dc shift depending on the SN value of the local image.

(Sixth Embodiment) The above dc shift type N in which the threshold control of the present invention is applied to a conventional noise reducer
R functions as a sufficient noise reduction device, but it is also possible to configure a dc shift type NR using the noise reduction processed image of the third embodiment of FIG. 6 obtained by the present invention as a reference signal. .

FIG. 9 is an overall block diagram of a sixth embodiment having the above configuration.

FIG. 9 is similar to the third embodiment in that the noise reduction unit 630 uses the noise reduction output from the Σ / M processing circuit 72 as the reference signal of the dc shift circuit 75. The rest is the same as the fifth embodiment in FIG. As a result, noise reduction with less image quality degradation can be performed using the noise reduction output of the present invention with less image quality degradation as a reference signal.

[0083]

According to the SN ratio detecting device of the present invention, a two-dimensional SN ratio and a three-dimensional SN ratio are obtained for each block of an input image.
The statistical method is used by taking the value of the smaller noise between the two SN ratios as the local SN ratio of the corresponding block, summing up the local SN ratio for the entire input image, and setting the mode as the global SN ratio. Thus, it is possible to realize detection of a global SN ratio and a local SN ratio suitable for real-time processing in which the SN ratio detection accuracy does not adversely affect the image quality.

Further, according to the noise reduction apparatus of the present invention, a threshold value corresponding to the image of each block is set in accordance with the smaller value of the local SN ratio and the global SN ratio obtained in the same manner as described above, and The pixel value of the pixel to be processed is the average pixel value of the pixels whose difference from the pixel value of the pixel to be processed has a pixel value equal to or less than the threshold value set in the block to which the pixel to be processed belongs, among the neighboring pixels of the pixel to be processed. , Noise reduction using local adaptive processing is realized, and image quality degradation due to noise reduction can be reduced.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a noise reduction device according to a first embodiment of the present invention.

FIG. 2 is an explanatory diagram illustrating an example of pixels near a processing target pixel.

FIG. 3 is an explanatory diagram illustrating the principle of a tilt correction method using an LPF.

FIG. 4 is a block diagram illustrating a configuration of a noise reduction device according to a second embodiment of the present invention.

FIG. 5 is a detailed block diagram of a tilt correction unit.

FIG. 6 is a block diagram illustrating a configuration of a noise reduction device according to a third embodiment of the present invention.

FIG. 7 is a block diagram illustrating a configuration of a noise reduction device according to a fourth embodiment of the present invention.

FIG. 8 is a block diagram illustrating a configuration of a noise reduction device according to a fifth embodiment of the present invention.

FIG. 9 is a block diagram illustrating a configuration of a noise reduction device according to a sixth embodiment of the present invention.

[Explanation of symbols]

 22 Block delay circuit 24, 28 Square circuit 25, 29 Averaging circuit 26, 46, 56, 61, 66 Frame delay circuit 31, 81 Two-dimensional LPF 64, 68, 71th or less pixel selection and addition circuit 65, 69, 74, 78 delay circuit 72 Σ / M processing circuit 75 dc shift circuit 80 inclination correction unit 100 blocking processing circuit 200, 210, 220 local SN ratio detection unit 400, 410 global SN ratio detection unit 500, 510 threshold setting unit 600, 610, 620,630 Noise reduction unit

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Hajinobu Mizutani 1-10-11 Kinuta, Setagaya-ku, Tokyo Japan Broadcasting Corporation Broadcasting Research Institute

Claims (9)

[Claims] 1. An input image is divided into a plurality of blocks, and a two-dimensional SN ratio and a three-dimensional SN ratio are obtained for each block. A local S / N ratio detecting unit that collects the local S / N ratio for the entire input image and sets a mode to a global S / N ratio; and a noise among the local S / N ratio and the global S / N ratio. An SN ratio detecting device comprising: an SN ratio selecting unit that selects a smaller value for each of the blocks. 2. The local SN ratio detecting means obtains, for each block, the two-dimensional SN ratio using variance of pixel values in the block as noise, and calculates a mean square of a time difference of each pixel in the block as noise. The SN ratio detecting device according to claim 1, wherein the three-dimensional SN ratio is obtained. 3. An input image is divided into a plurality of blocks, and a two-dimensional SN ratio and a three-dimensional SN ratio are obtained for each block. A local S / N ratio detecting unit that collects the local S / N ratio for the entire input image and sets a mode to a global S / N ratio; and a noise among the local S / N ratio and the global S / N ratio. Threshold value setting means for setting a threshold value according to the image of each block according to the smaller value; and among the neighboring pixels of the processing target pixel in the input image, a difference between the pixel value of the processing target pixel and the processing target pixel is determined. And averaging means for replacing the pixel value of the processing target pixel with an average pixel value of a pixel having a pixel value equal to or less than a threshold value set in a block to which the pixel belongs. Noise reduction device. 4. A means for obtaining the brightness of an image in a block for each of the divided blocks, wherein the threshold value setting means sets the threshold value to a larger value when the brightness of the image in the block is lower than when the brightness of the image in the block is higher. The noise reduction apparatus according to claim 3, wherein the correction is performed so that 5. The averaging means includes means for detecting a two-dimensional inclination of an image including the processing target pixel, and calculating a pixel value of the processing target pixel and the inclination value to calculate the inclination of the processing target pixel. Means for calculating a tilt-corrected pixel value, wherein, among the neighboring pixels, a pixel whose difference from the tilt-corrected pixel value is a pixel value equal to or less than a threshold value set in a block to which the processing target pixel belongs 4. The pixel value corrected for inclination is replaced with an average pixel value of
Or the noise reduction device according to 4. 6. The average pixel value is 3 of the pixel to be processed.
The noise reduction apparatus according to claim 5, wherein the noise reduction apparatus is obtained from neighboring pixels in a dimension, and the neighboring pixels belong to at least a frame to which the processing target pixel belongs and a frame before the frame. 7. A comparing means for comparing the replaced pixel value to be processed with a pixel value to be processed in the input image, and a pixel value to be processed in which the pixel value to be processed in the input image is replaced by the comparing means. If greater than, the threshold is subtracted from the processing target pixel value of the input image, and a value is output.If the processing target pixel value of the input image is smaller than the replaced processing target pixel value, the threshold is set to the threshold. Output the value added to the processing target pixel value of the input image,
5. The image processing apparatus according to claim 3, further comprising: a selection addition / subtraction unit that outputs a processing target pixel value of the input image when the processing target pixel value of the input image is equal to the processing target pixel value of the output image. 2. The noise reduction device according to 1. 8. An input image is divided into a plurality of blocks, and a two-dimensional SN ratio and a three-dimensional SN ratio are obtained for each block. A local S / N ratio detecting unit that collects the local S / N ratio for the entire input image and sets a mode to a global S / N ratio; and a noise among the local S / N ratio and the global S / N ratio. Threshold setting means for setting a threshold according to the image of each block according to the smaller value; noise reduction means for globally reducing the noise of the input image and outputting; output image of the noise reduction means and the input image Comparing means for comparing the pixel values to be processed, and if the pixel value to be processed of the input image is larger than the pixel value to be processed of the output image by the comparing means, A value obtained by subtracting a threshold value from a processing target pixel value of the input image is output. If the processing target pixel value of the input image is smaller than the processing target pixel value of the output image, the threshold is set to the processing target pixel value of the input image. A selection addition / subtraction unit that outputs a value obtained by adding the value to the value, and outputs a processing target pixel value of the input image when the processing target pixel value of the input image is equal to the processing target pixel value of the output image. A noise reduction device characterized by the above-mentioned. 9. An input image is divided into a plurality of blocks, a two-dimensional S / N ratio and a three-dimensional S / N ratio are obtained for each of the blocks, and a value of the two S / N ratios having less noise is used as a local S / N ratio of the block A local S / N ratio detecting unit, a global S / N ratio detecting unit that detects a global S / N ratio by comparing noise in the input image with a noise distribution according to a statistical distribution rule, Threshold setting means for setting a threshold corresponding to the image of each block according to the value of the lesser noise, and a difference between the pixel value of the processing target pixel among the pixels in the vicinity of the processing target pixel in the input image, Averaging means for replacing a pixel value of the processing target pixel with an average pixel value of a pixel having a pixel value equal to or less than a threshold value set in a block to which the processing target pixel belongs. Noise reduction device.