JPH07212627A - Noise reduction device - Google Patents

Abstract

PURPOSE:To obtain an excellent noise reduction effect with respect to actual picture data by applying orthogonal transformation to an inter-field or inter- frame difference signal to apply nonlinear processing thereto, applying inverse orthogonal transformation to the signal and taking a difference between the signal and an input video signal. CONSTITUTION:An adder 102 subtracts a delayed video signal (signal of one preceding frame) from an input video signal (signal of current frame) from an input terminal 101 to extract an inter-frame difference signal. An orthogonal transformation signal processing section 106 gives a transformation output or transformation coefficient from an orthogonal transformation circuit 107 to an inverse orthogonal transformation circuit 109 via a nonlinear circuit 108. The output is fed to an adder 103, in which it is subtracted from an input signal of the current frame. Since nonlinear processing is applied to coefficient data of an orthogonal transformation output by using a nonlinear curve approximated to an optimum curve obtained by a histogram of transformation coefficient data, an excellent noise reduction effect is obtained from the current picture data.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a noise reducing device for improving the SN ratio of a video signal, and more particularly to a noise reducing device using orthogonal transformation such as Hadamard transformation.

[0002]

2. Description of the Related Art A noise reducing device (Noise Redu) for reducing noise contained in a video signal to improve the SN ratio.
Various cer: noise reducers) have been proposed in the past.

FIG. 1 shows an example of a basic circuit configuration of such a noise reduction device. One of such circuits is used for a composite video signal, and another circuit is used for each signal component (component) of a component video signal. In the circuit of FIG. 1, an adder 102 that acts as a subtractor
From, the difference between the current input image from the input terminal 101 and the latest processed image from the image memory 104 is obtained, and the difference output is sent to the orthogonal transform signal processing unit 106, and the Hadamard transform circuit, etc. The orthogonal transform circuit 107 performs the orthogonal transform to produce a transform coefficient output, which is processed in the transform domain.

By the way, the noise component tends to be evenly distributed to the respective conversion coefficients, and is represented as a low coefficient value. In the presence of motion, motion components tend to be unevenly distributed among the coefficients. This transform coefficient is passed through a circuit 108 having a group of non-linear curves. This non-linear curve zeros out the components of the value above a certain threshold N as a detail of the movement. Values below the threshold N are passed or attenuated according to the non-linear function. After the nonlinear processing, the output from the nonlinear circuit 108 is the inverse orthogonal transform circuit 10
To the adder 103 for subtraction from the current input signal.

The simplest non-linear curve of the non-linear circuit 108 is shown in FIG. A in FIG. 7 represents a function F (x) that is multiplied by the input conversion coefficient data x, and B in FIG. 7 represents output data with respect to the input data x, that is, xF (x). . This function F (x) is expressed by the following equation. F (x) = K; | x | ≦ N = 0; | x |> N In this equation, N is a noise threshold and K is a gain of the function. In the example of FIG. 7, N = 40, K = 1. Shows the case. This is not desirable and contains the worst curve.

[0006] Here, Ebihara et al., "Noise Reducer of Television Signal Using Hadamard Transform", Journal of Television Society, Vol.37, No.12 (1983), pp.1030-1036.
Discloses an example using Hadamard transform for the orthogonal transform. Further, since the composite NTSC signal is used as the input video signal, it is necessary to take the modulation into consideration. Sampling once or twice per subcarrier period is required to convert the signal with the same offset strength. When the sampling rate is four times the subcarrier frequency, that is, f _S = 4f _SC , sampling every 2 pixels or every 4 pixels is required, and in this example, sampling is performed every 2 pixels. As a result, each conversion coefficient gives not only the frequency but also the index of the luminance or color component of the image. That is, part of the conversion coefficient represents the presence of a strong luminance signal, and the other part represents the presence of a strong color difference signal. Here, since human visual characteristics with respect to brightness and color are different, two sets of non-linear curves for the luminance coefficient and for the color difference coefficient are used in the above example.

In the example of Ebihara, the simulation is performed using actual data. Also, 2x2, 2x4, 4x
It simulates several different transformation matrices, including four. Nonlinear curve F used in the Ebihara example above
The shape of (x) is shown in FIG. 8, and the formula of F (x) is as follows: F (x) = K (1- | x | / N); | x | ≦ N = 0; | x |> N Defined in. Here, N is a noise threshold and K is a gain of the function. In the example of FIG. 8, the case of N = 40 and K = 1 is shown. Further, A in FIG. 8 represents the function F (x), and B represents the output data xF (x) with respect to the input data x.

Next, Nakajima et al., "A New Noise Reduction System for Video Camera", I
EEE Transactions on Consumer Electronics, Vol. 37, No. 3, August, 1991 (Nak
ajima et al. "A NEW NOISEREDUCTION SYSTEM FOR VIDE
O CAMERA ", IEEE Transactions on Consumer Electronic
s, Vol.37, No.3, August 1991), a noise reduction apparatus using component signal processing is disclosed. That is, the luminance (Y) and the two color difference signals (R-
Y, BY) are processed independently. Four different nonlinear curves are compared and the curve of F (x) chosen as the best is shown in FIG. The formula of F (x) is defined as follows.

F (x) = K; | x | ≦ N = KN / | x |; | x |> N FIG. 9 shows the case where the noise threshold N = 20 and the gain K = 1 of the function, and A is the function. F (x) represents the output data xF (x), and B represents the output data xF (x). For each simulation, all transform coefficients except the DC coefficient use the same non-linear parameters, namely N and K. The DC coefficient does not contribute to noise reduction.

[0010]

By the way, in the above-mentioned technique of Ebihara et al., The non-linear curve parameters for luminance and color are controlled independently. The frequency distribution of each coefficient is not considered when selecting the non-linear function parameter. For example, the histogram (frequency graph) of the motion image shows that the coefficient for DC (direct current) tends to have a considerably wider distribution than the coefficient for high frequency. Similarly,
If the sequence of images is a horizontal rotation of the scene, i.e. a horizontal pan, the horizontal frequencies will have a wider distribution. If the nonlinear curve parameters for all coefficients are fixed, it cannot be optimized for different types of scenes.

In the technique of Ebihara et al., A memory-mapped ROM is used to realize the non-linear curve F (x). This means that small N and K values are applied which give the user limited control over the parameter selection.

Since the results obtained by the Nakajima et al. Technique described above are determined using simulated input patterns, execution with real data can produce different results. It is difficult to test all possible situations with computer generated test images. Through the simulation using the actual data, the present inventor found that the non-linear curve (FIG. 9) selected as the optimum in the technique of Nakajima et al. Was compared with the non-linear curve used in the technique of Ebihara et al. (FIG. 8). We have found that it actually works worse. Similarly, through statistical analysis of real image data, the same non-linear curve does not perform as well as other noise reduction devices. Comparison of non-linear curves cannot be realized only by performing simulation for each curve having the same non-linear coefficient (N and K). Each non-linear curve may have the best performance at different values of N and K.

In the technique of Nakajima et al., Excluding the DC coefficient, the same values of N and K are used for all coefficients. The non-linear curve for the DC coefficient is set to zero. Therefore, the DC coefficient does not contribute to noise reduction and does not deteriorate motion details. The best achievable MSE can never be achieved because the noise of this coefficient is never reduced. For example, using statistical analysis, the best MSE, provided all coefficients can change, is 16.
It is determined to be 0. If all coefficients use the same values for N and K, the highest MSE will be 17.1.
If the DC coefficient is set to 0, the MSE can be further increased to 18.6. Although the DC coefficient contributes slightly to noise reduction, there may be situations where it should not always be set to zero.

The present invention has been made in view of the above circumstances, and it is possible to realize a good noise reduction effect for actual image data in a noise reduction device including orthogonal transformation such as Hadamard transformation. To provide a simple noise reduction device.

[0015]

A noise reducing apparatus according to the present invention is configured to take a difference between a delay means for delaying an input video signal by one field or one frame and a difference between the input video signal and an output from the delay means. A first subtraction means for outputting an inter-field difference signal or an inter-frame difference signal, an orthogonal transformation means for orthogonally transforming the difference output from the first subtraction means, and a nonlinear transformation output from the orthogonal transformation means. A second non-linear circuit for performing processing, an inverse orthogonal transform means for performing an inverse orthogonal transform on the output from the non-linear circuit, a difference between the input video signal and the output from the inverse orthogonal transform means, and sending the difference to the delay means. And a non-linear curve approximating an optimum curve obtained from the histogram of the transform coefficient data output from the orthogonal transform means. By Rukoto, for solving the above problems.

The non-linear curve of the non-linear circuit may be determined by combining the histogram data obtained from the conversion coefficients of the moving image and the image with noise. Specifically, for example, the shape of the non-linear curve used for the non-DC coefficient of the transform coefficient is different from the shape of the non-linear curve used for the DC coefficient, and the non-linear curve used for the non-DC coefficient is , A curve shape whose value decreases near 0 of the conversion coefficient. Also,
It is preferable that the orthogonal transform means is a Hadamard transform circuit and the inverse orthogonal transform means is an inverse Hadamard transform circuit.

Here, the noise reduction device according to the present invention is
It optimizes one stage of a Hadamard transform noise reduction system, and what is added to this system is a non-linear transformation curve designed to separate the noise from the image data. In conventional systems, this non-linear curve is designed to classify small field-to-field or frame-to-frame differences of high order Hadamard coefficients into noise and consider large differences to be motion. However, in the present invention,
Both high order and very small interfield or interframe differences are considered image motion. 2 above
The non-linear curve separating the two takes this into account.

[0018]

The coefficient data of the orthogonal transform output is subjected to the non-linear processing by using the non-linear curve that approximates the optimum curve obtained from the histogram of the transform coefficient data, so that a good noise reduction effect can be obtained for the actual image data. be able to.

[0019]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A preferred embodiment of a noise reducing device according to the present invention will be described with reference to the drawings. FIG. 1 shows an example of a basic circuit configuration to which the noise reduction device according to the present invention is applied. One of such circuits is used for a composite video signal, and another circuit is used for each signal component (component) of a component video signal.

In FIG. 1, the input video signal supplied to the input terminal 101 is sent to the adders 102 and 103, respectively. Each of these adders 102 and 103 subtracts the other input from the one input
It operates as a subtraction means of. The output signal from the adder 103, which is the second subtraction unit, is sent to the image memory 104 using a frame memory, delayed by one frame, and taken out from the output terminal 105. Note that a field memory may be used as the image memory 104 to delay one field.

In the adder 102 which is the first subtraction means,
The delayed video signal (the signal one frame before) from the image memory 104 is subtracted from the input video signal (current frame signal) from the input terminal 101 to extract the inter-frame difference signal. This interframe difference signal is sent to the adder 103 as a subtraction signal via the orthogonal transformation signal processing unit 106. When a field memory is used as the image memory 104, the inter-field difference signal is taken out from the adder 102.

The orthogonal transform signal processing unit 106 has a structure for sending the transform output or transform coefficient from the orthogonal transform circuit 107 such as Hadamard transform circuit to the inverse orthogonal transform circuit 109 via the non-linear circuit 108. . Here, the orthogonal transform circuit 107 performs the orthogonal transform by Hadamard transform or the like.
In the transform coefficients obtained from, noise tends to be distributed approximately evenly among the coefficients and is represented as a small coefficient value. In the presence of motion, the motion components tend to be unevenly distributed among the coefficients. This transform coefficient is passed through a circuit 108 having a group of non-linear curves.

Here, the non-linear circuit 10 in the present embodiment.
The non-linear curve in 8 is a non-linear curve used for non-DC coefficients, as shown in FIG. 2A shows the function F (x), and B shows the relationship between the input and the output. The input / output characteristics of B in FIG.
It can be stored in M and used as a lookup table. ALU (arithmetic logic unit) for input data
When the output is obtained by performing the calculation by the above, the characteristic of A in FIG. 2 may be used. The non-linear curve used for the DC coefficient may be the same curve as the conventional one. Specific examples of the non-linear curve include the non-linear curves shown in FIG. 8 and FIG. 9 described above and FIG. A non-linear curve as shown in can be used.

The non-linearly processed output from the non-linear circuit 108 is sent to the inverse orthogonal transform circuit 109 to undergo an inverse transform, for example, an inverse Hadamard transform, and sent to the adder 103 to be subtracted from the input signal of the current frame.

The major differences between the respective techniques in the plurality of embodiments of such a noise reduction device are (1) the order of the matrix, (2) the shape of the non-linear function, (3) the selection of the parameters of the non-linear function, (4) Sampling frequency, and (5)
Bit precision is mentioned.

Next, the meaning of the non-linear curve as shown in FIG. 2 used in this embodiment and the noise reducing action when using it will be described below.

Optimal shaped curves were developed through statistical analysis of the histogram data of the transformed output of the image signal.
It is not possible to determine the optimum curve for every image.
However, it is possible to implement the shape of the curve using a memory-mapped ROM and to implement the linear approximation of the shape in hardware using several multipliers.

Histogram data obtained from two image sequences, ie, (1) a moving image (including motion and no noise) and (2) an image with noise (including no motion and including film noise) are combined. By this, the optimum curve is determined. Since the Hadamard transform is orthogonal, it is shown that variations in the spatial domain data can be related to variations in the transform domain coefficient data.

The histogram obtained from each of the image sequences is the coefficient distribution in the Hadamard region.

The output of the non-linear function in a moving image is added on the image by the noise reducer and represents image degradation. The sum of coefficient fluctuations represents the root mean square error of this image deterioration.

For an image with noise, the difference between the input and output of the non-linear curve represents the amount of noise reduction applied. Therefore, the sum of the fluctuations of the difference represents the root mean square error (Mean Square Error, hereinafter referred to as MSE) of the processed image.
Since noise and motion are uncorrelated signals, motion degradation MSE
An overall MSE estimate can be obtained by adding the noise reduction MSE to. The mathematical expression for this is MSE _total = MSE _move + MSE _noise . That is,

[0032]

[Equation 1]

Where σ _{k, noise} is the standard deviation of the coefficient k of the difference between the input and output of F (x) for a static and noisy image sequence, and σ _{k, move} includes noise. The standard deviation of the coefficient k at the output of F (x) for the missing image sequence, P is the transformation matrix.

The following variables are introduced to represent the total MSE in terms of actual histogram data.
That is, S _{k, move} (n) represents the total number of times the coefficient k receives the value n when the histogram is formed from the moving image. For example, when the coefficient value n is equal to 20, S
_{k, move} (20) is the total number of occurrences of the coefficient k having a value of 20. S _{k, noise} (n) represents the total number of times the coefficient k received the value n when the histogram was formed from the noisy image. F _k (n) represents a non-linear function used with the coefficient k. | Max | represents the maximum value at which n becomes equal, and -Max <n <Max is the range of histogram levels.

Using these new variables, the M of the moving image is
SE is represented by the following equation.

[0036]

[Equation 2]

Similarly, the MSE of the noise image is expressed by the following equation.

[0038]

[Equation 3]

Therefore, the total MSE is

[0040]

[Equation 4]

In order to optimize the MSE, the nonlinear curve F
It is necessary that the optimal shape of (x) be determined. This is determined by minimizing MSE _total ,

[0042]

[Equation 5]

Since minimizing the addition formula is equivalent to minimizing the argument of the addition formula, the above {MSE _total }
_min is determined by minimizing the following equation.

[0044]

[Equation 6]

FIGS. 3 to 6 show optimum curves for four sample coefficients, and show histogram data extracted from 10 frames of image data. The data for the histogram is obtained as coefficient output data that is Hadamard transformed using, for example, a 4 × 4 Hadamard matrix. Each figure corresponds to different coefficient frequencies. That is, FIG. 3 corresponds to a DC coefficient (X = 0, Y = 0), and FIGS. 4 to 6 correspond to an increased horizontal frequency (X = 1 to 3), respectively.

In FIGS. 3 to 6, the curve a shows the histogram of the moving image, the curve b shows the histogram of the noise image, and the curve c shows the optimum curve. The histograms of the moving image and the noise image are normalized to 1 at the origin. Noise samples are obtained from static images of the blue sky called "birds" and moving image samples are obtained from a noiseless image of a group of dancers called "dances". The plot shape shown is representative of the optimum curve determined by other motion / noise combinations.

A non-linear curve similar to the Gaussian curve of FIG.
Represents a DC (direct current) term of = 0 and Y = 0. However, in most of the other samples, the average DC term is zero. The DC term in FIG. 3 is not zero-mean because the “dance” sequence contains zooming. Coefficients other than the DC term have a curve of zero average as in typical samples shown in FIGS. 4 to 6, and the optimum curve has a minimum value at the origin. The two side lobes are symmetric with respect to the origin.
Since the maximum value of the histogram is normalized to 1, it is difficult to visually grasp the necessity of this minimum value when looking at FIGS. 4 to 6. If the histogram distribution can be regarded as a probability density function, the area surrounded by each distribution curve becomes 1. In order for the coefficient of the moving image to have a distribution of 1, the value in the vicinity of 0 is required to be large. This means that for these coefficients, motion is dominant in the near 0 region. Therefore, the optimum curve should consider both high coefficient values and very small coefficient values as motion.

As mentioned above, it is not practical to calculate the optimum curve for every image. However, since the curve shape is a property of some noise / motion combinations, an approximation to the optimum curve shape may improve the effectiveness of the noise reduction device. Shapes similar to those seen in FIGS. 3-6 can be stored in ROM and used as a memory map. However, even though the shapes are similar for other motion / noise combinations, the sidelobe gain and width change. Approximation to the optimal curve, including gain and threshold variables, can increase the degrees of freedom. An approximation to an optimal curve has been proposed, which is a simple approximation of the optimal shape of the non-DC coefficient. An example of this new F (x) curve is shown in FIG. 2 described above. The equation for this curve is:

[0049]

[Equation 7]

Here, N is a noise threshold and K is a gain of the function. In the example of FIG. 2, N = 45 and K = 1 are shown. 2A shows the function F (x), and B shows the output data xF (x) for the input data x. It is possible to propose another more complicated function as a function representing the approximate curve of such an optimum shape, but the function can be easily realized because the hardware configuration such as a multiplier is extremely small. Have advantages. Moreover, if an improvement is obtained by such a simple approximation,
Other more complex approximations will yield even greater improvements.

Next, FIGS. 7 to 10 show examples of various shapes of the non-linear curve for comparison with the non-linear curve of the present embodiment (FIG. 2). In these FIG. 7 to FIG.
Each A represents the function F (x), and each B represents the output data xF (x).

FIG. 7 shows a straight threshold non-linear curve, which is not desirable for noise reduction and is classified as the worst case curve. FIG. 8 shows a function used in the technique of Ebihara et al. FIG. 9 shows the technique selected by Nakajima et al. Described above as the optimum non-linear curve for use in such a noise reduction system. Figure 1
0 indicates another non-linear curve used for comparison, the function F (x) is F (x) = K; | x | ≦ N / 2 = 2K (N− | x |) / N; N ≧ | x |> N / 2 = 0; | x |> N. The example of FIG. 10 shows a case where the noise threshold N = 40 and the gain K of the function is K = 1.

Here, in order to evaluate the performance of the new non-linear curve (FIG. 2) of this embodiment, that is, the curve approximate to the optimum curve, the four curves shown in FIGS. A statistical comparison was made with the curve shown in. Table 1 shows a summary of the results of this comparison.

[0054]

[Table 1]

In Table 1, two types of "birds and faces" and "birds and dances" are used as image data sources, "birds" are noise images, and "faces" and "dances" are movements. Shows an image with. As the non-linear curve, the curves of FIGS. 7 to 10, the curve of FIG. 2 of the present embodiment, and the optimum curve are used, and the noise threshold N and the gain K of the function are fixed for all coefficients. And MSE values for the variable case are shown.

As is clear from Table 1, among the remaining five methods excluding the case of the optimum curve, the method of this embodiment is the best for both "bird and face" and "bird and dance". M
It can be seen that SE is generated.

Next, FIGS. 11 and 12 show the above-mentioned FIGS.
2 shows the root mean square error with respect to changes in the noise threshold N and the gain K of the function, so-called MSE, when using each of the nonlinear curves shown in FIG. That is, FIG. 11 shows the MS with respect to the change of the noise threshold N in the state where K that is the minimum MSE is selected.
12 shows E, and FIG. 12 shows the MSE with respect to the change of the gain K when N, which is the minimum MSE, is selected. 11 and 12, the curve a is the case of using the non-linear curve of FIG. 7, the curve b is the case of Ebihara et al. In which the non-linear curve of FIG. 8 is used, and the curve c is the case of FIG. In the case of the technique of Nakajima et al. In which a non-linear curve is used, the curve d shows the case of using the non-linear curve of FIG. 10 and the curve e shows the case of the present embodiment using the non-linear curve of FIG. This is the minimum MSE plus K
This is because it is necessary to consider the sensitivity of F (x) with respect to the changes of N and N.

From FIG. 11, the curve b representing the technology of Ebihara et al. And the curve e representing the present embodiment have low sensitivity to changes in the noise threshold N, and the curve c representing the technology of Nakajima et al. Has the highest sensitivity. I understand. Therefore, if the user makes a mistake in the target minimum MSE, sufficient noise reduction effect can be obtained in the above-described Ebihara et al. Technology and this embodiment, whereas the noise reduction effect is substantially obtained in the other technology. Fall to.

From FIG. 12, it can be seen that the difference in sensitivity for K near the minimum MSE does not appear clearly. Therefore, there is no clear best curve when considering the sensitivity of MSE for K.

Therefore, the noise reduction effect of the Hadamard transform noise reduction apparatus can be improved by using the new nonlinear curve (FIG. 2) of this embodiment. However, the curve used in this example is only a simple approximation of the optimum curve. By being able to improve performance with such a simple approximation,
Further refinements can be made using other, more precise approximations.

It is recognized that the use of the non-linear curve approximated to the above-mentioned optimum curve can be improved in that the value in the region very close to 0 is controlled not by the noise but by the motion detail. It depends.

It can also be seen that the noise reduction capability is expanded by independently selecting the nonlinear curve of each coefficient.

The present invention is not limited to the above embodiment, and for example, the orthogonal transform is not limited to the Hadamard transform, and various orthogonal transforms such as discrete cosine transform can be used. Further, the non-linear curve is not limited to the example of FIG. 2, and it goes without saying that various non-linear curves that approximate the optimum curves of FIGS. 3 to 6 can be used.

[0064]

As is apparent from the above description, according to the noise reducing apparatus of the present invention, the delay means for delaying the input video signal by one field or one frame, the input video signal and the delay means. First subtracting means for outputting an inter-field difference signal or an inter-frame difference signal by taking the difference from the output, an orthogonal transforming means for orthogonally transforming the difference output from the first subtracting means, and this orthogonal transforming means A non-linear circuit for performing non-linear processing on the output of the orthogonal transform from the above, an inverse orthogonal transform means for performing an inverse orthogonal transform on the output from the non-linear circuit, and a difference between the input video signal and the output from the inverse orthogonal transform means. A second subtraction means for sending to the delay means, and the non-linear circuit produces an optimum curve obtained from a histogram of the transform coefficient data output from the orthogonal transform means. By having a similar nonlinear curve, it is possible to obtain a good noise reduction effect with respect to the real image data.

The non-linear curve of the non-linear circuit may be determined by combining the histogram data obtained from the conversion coefficients of the moving image and the image with noise. Specifically, the shape of the non-linear curve used for the non-DC coefficient of the transform coefficient is different from the shape of the non-linear curve used for the DC coefficient, and the non-linear curve used for the non-DC coefficient is The noise reduction capability is increased by setting the curve shape such that the value decreases in the vicinity of 0 of the coefficient.

[Brief description of drawings]

FIG. 1 is a block circuit diagram showing a basic circuit configuration to which an embodiment of a noise reduction device according to the present invention is applied.

FIG. 2 is a diagram showing a non-linear curve of a non-linear circuit used in the noise reduction device of the present embodiment.

FIG. 3 is a DC obtained by Hadamard transform of image data.
It is a figure which shows an example of the histogram of coefficient data.

FIG. 4 is a diagram showing an example of a histogram of coefficient data of horizontal frequency X = 1 obtained by Hadamard transform of image data.

FIG. 5 is a diagram showing an example of a histogram of coefficient data of horizontal frequency X = 2 obtained by Hadamard transform of image data.

FIG. 6 is a diagram showing an example of a histogram of coefficient data of horizontal frequency X = 3 obtained by Hadamard transform of image data.

FIG. 7 is a diagram showing an example of a conventional non-linear curve of a non-linear circuit used in a noise reduction device.

FIG. 8 is a diagram showing another example of a conventional non-linear curve of a non-linear circuit used in a noise reduction device.

FIG. 9 is a diagram showing still another example of a conventional nonlinear curve of a nonlinear circuit used in a noise reduction device.

FIG. 10 is a diagram showing an example for comparison of non-linear curves of a non-linear circuit used in a noise reduction device.

FIG. 11 is a diagram showing a root mean square error MSE when the noise thresholds for a plurality of nonlinear curves are changed.

FIG. 12 is a diagram showing a root mean square error MSE when the gain of a function for a plurality of nonlinear curves is changed.

[Explanation of symbols]

 102, 103 adder 104 image memory 107 orthogonal transformation circuit 108 nonlinear circuit 109 inverse orthogonal transformation circuit

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Office reference number FI Technical display location H04N 7/30 H04N 7/133 Z

Claims (4)

[Claims] 1. An inter-field difference signal or an inter-frame difference signal is output by taking the difference between the input video signal and the output from the delay means, and delay means for delaying the input video signal by one field or one frame. First subtracting means, orthogonal transforming means for orthogonally transforming the difference output from the first subtracting means, non-linear circuit for performing non-linear processing on the orthogonal transform output from the orthogonal transforming means, and output from the non-linear circuit And a second subtracting means for taking the difference between the input video signal and the output from the inverse orthogonal transforming means and sending it to the delay means. A noise reduction device having a non-linear curve that approximates an optimum curve obtained from a histogram of transform coefficient data output from the orthogonal transform means. 2. The noise reducing apparatus according to claim 1, wherein the non-linear curve of the non-linear circuit is determined by combining histogram data obtained from transformation coefficients of a moving image and an image with noise. 3. The non-linear curve of the non-linear circuit is configured such that the shape of the non-linear curve used for the non-DC coefficient of the conversion coefficient is different from the shape of the non-linear curve used for the DC coefficient. The noise reduction device according to claim 1, wherein the non-linear curve used for the curve has a curve shape whose value decreases near 0 of the conversion coefficient. 4. The noise reduction device according to claim 1, wherein the orthogonal transforming means is a Hadamard transforming circuit, and the inverse orthogonal transforming means is an inverse Hadamard transforming circuit.