JPS5925426B2 - Noise removal circuit - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION To remove noise contained in an input signal, such as a video signal obtained from a VTR, a noise removal circuit having a low-pass filter or a high-pass filter is generally used.

For example, as shown in FIG. 1, a video signal Si supplied to a terminal 1 is supplied to a low-pass filter 2 and a high-pass filter 3, respectively, and the output obtained from the high-pass filter 3 is further transmitted to a second filter.
As shown by the curve □ in the figure, the output signal is supplied to the nonlinear circuit 4 which has an input/output characteristic having a dead zone Ta to remove the noise present in the dead zone 7a, and then synthesized with the output of the low-pass filter 2 described above. 5, and an output video signal 50 containing no noise signal is obtained from a terminal 6.

This noise removal circuit aims to improve image quality by removing noise components that exist in the high frequency range, but when filter circuits such as low-pass filters and high-pass filters are used in this way, the following It has the following drawbacks.

That is, in addition to insufficient separation of the signal component and noise component in the video signal Si in this conventional circuit, the signal component of the output of the pass filter 3 is subjected to large distortion in the nonlinear circuit 4. Fine brightness changes in parts of the reproduced image where the brightness changes are small, such as the background, disappear, and as a result, a scaly pattern is generated in the image corresponding to the parts where the brightness changes are gradual.

On the other hand, if the brightness changes are steep, smear occurs and the sharpness of the image deteriorates. In order to remove noise without suffering much image degradation, that is, to effectively remove noise in signal components, it is necessary to better separate the signal component and noise component, and then perform signal manipulation to remove the noise. However, when manipulating the signal, it is only necessary to make the signal subjected to distortion less noticeable from the human visual characteristics. In order to satisfy each of these conditions, orthogonal transformation may be used.

The idea of using orthogonal transform to remove noise and improve image quality deterioration was immediately proposed by the applicant, but the present invention deals with a noise removal circuit that uses this orthogonal transform. Furthermore, as will be described later, the inversely transformed output is taken out by taking a weighted average, thereby reducing the distribution of image distortion and thereby further improving the image quality. A noise removal circuit using orthogonal transform according to the present invention will be described below with reference to the drawings, but first, a few words about orthogonal transform will be described first. Now, if the block of the video signal sequence which is the input signal is represented by X, similarly the block of the output signal sequence is represented by Y, the orthogonal transformation matrix is represented by A, and the inverse transformation matrix is similarly represented by B. Therefore, when the input signal is orthogonally transformed, Therefore, the inverse transform output is: Therefore, the transform coefficient, that is, the orthogonal transform output Y is a linear combination of the row vector and the input signal.

By the way, in the present invention, orthogonal transforms such as Warschu, Hadamard, and Barr can be used, but in this example, the characteristics of the television signal can be well extracted, and the inverse transform can be performed using the same procedure as the transform. Let us exemplify the case of using the Hadamard transformation, which has the characteristics of being able to perform the following.

FIG. 3 shows an example of a noise removal circuit using orthogonal transform that has already been proposed by the applicant, and this example shows the case of one-dimensional orthogonal transform.

That is, this is a case where signals corresponding to a plurality of consecutive unit areas (corresponding to one picture element in terms of an image) are orthogonally transformed in a video signal in a certain horizontal scanning section. . The video signal Si supplied to the terminal 11 is supplied to a parallelization circuit 12 to obtain a plurality of temporally synchronized video signals.

The parallelization circuit 12 is composed of a plurality of delay lines connected in series, and the number of delay lines to be used is determined from the matrix configuration of the orthogonal transformation circuit, which will be described later. In this example, three delay lines 12a to 12c are connected in series. Note that the delay amounts τ are both selected to be equal, and in this example are about 100 ns.

This delay amount τ is determined by the sampling frequency necessary to sample the entire band of the video signal Si. For example, if the entire band of the video signal Si is approximately 4.5 MHz, the sampling frequency is approximately 10 MHz. Therefore, in that case, the delay amount τ is selected to be 100 ns. By the way, if the video signal supplied to the terminal 11, that is, the non-delayed output, is denoted as Sil, then the plurality of delayed outputs Si2 to Si obtained from the delay lines 12a to 12c for parallelization are
4 are respectively supplied to the orthogonal transform circuit 20.

The orthogonal transform circuit 20 employs Hadamard transform as described above. For convenience of explanation, in this example, a case using a fourth-order Hadamard transformation matrix H4 will be exemplified.

Therefore, the matrix structure is as shown in equation (8) below. In order to realize this fourth-order Hadamard transformation matrix H4, the structure shown in FIG. 4 may be used.

The fourth-order Hadamard transform circuit 20 has four row lines 11 to 14 and four column lines m1 to M4, respectively, as shown in the figure.
1 to 14 are further branched into two row lines in order to extract the positive and negative signals supplied thereto separately, and one of the row lines is provided with a phase inversion circuit 15, respectively.

A plurality of attenuators 1 are provided between the plurality of row lines and column lines so as to satisfy the matrix condition of equation (8).
3 are connected between the desired lines. Therefore, if the outputs Sil to Si4 obtained by the parallel circuit 12 are now supplied to the terminals 21a to 21d of the plurality of row lines 11 to 14, respectively, the desired outputs will be output from the terminals 22a to 22d of the plurality of column lines m1 to M4, respectively. orthogonal transform outputs SO1 to S-04 are obtained.

These orthogonal transformation outputs are expressed by the following equation. Next, let us examine the properties of the orthogonal transform outputs SO1 to SO4 obtained in this way.

Since the power spectrum of the video signal contains many low frequency components as shown in FIG. 8, there are many parts that change smoothly.

Furthermore, as is well known, the high frequency components of the video signal correspond to the contours of the image, and the low frequency components correspond to positions in the image where the brightness does not change much. Therefore, among the orthogonal transform outputs, the low-order transform output, that is, SO1 in this example, has a strong tendency to contain many low frequency components, whereas the high-order orthogonal transform outputs, that is, SO2 to SO4, have a strong tendency to contain many low frequency components. It tends to contain high frequency components. This tendency is strong in the case of high-order transformation matrices such as 8th and 16th orders.

In other words, by orthogonally transforming a video signal, this video signal is concentrated in the low-order orthogonal transform output, and the higher the order, the less it appears in the orthogonal transform output. It means something. On the other hand, noise contained in a video signal obtained from the output of a VTR or the like is distributed over substantially the entire band and is considered to be random noise.

Therefore, when a video signal with such noise is orthogonally transformed, the noise components will be distributed with approximately uniform amplitude in each orthogonal transform output. ~・The orthogonal transform output has a good S/N, and the noise component is dominant in the high-order orthogonal transform output, so its S/N
deteriorates. What can be said from these facts is that when the level of the orthogonal transform output is high, it can be regarded as a video signal component.

On the other hand, if the level of the orthogonal transform output is small, it may be regarded as a noise component. Moreover, as mentioned above, the S/N deteriorates in high frequency components, but human visual characteristics for these high frequency components generally deteriorate.
Even if the reproducibility in the high frequency region during reproduction is poor, it does not have much influence on the reproduced image. From the above, when the orthogonal transform output is at a low level that is approximately determined by the frequency spectrum of noise, it is configured to treat it as a noise component and remove it.

Therefore, a non-linear circuit 16 is provided at the subsequent stage of the orthogonal transform circuit 20, and by supplying the respective orthogonal transform outputs SO1 to SO4 to this non-linear circuit 16, the noise present in the dead zone 7a shown in FIG. 2 is removed. It was designed to do so.

In this example, other outputs SO1 except the orthogonal transformation output SO1
Nonlinear circuits 16a to 16c are installed on the transmission paths of 2 to SO4, respectively.
is provided to remove noise contained in each of the signals SO2 to SO4.

Note that the reason why the nonlinear circuit 16 is not interposed in the first orthogonal transformation output SO1 is to prevent waveform distortion etc. from occurring in this output SO1, which greatly affects the reproduced image. Second
A specific circuit example of the nonlinear circuit 16 having input/output characteristics as shown in the figure is shown in FIG.

Since the nonlinear circuit 16 described below is conventionally well-known, its outline will be explained as follows: The orthogonal conversion output is passed through an amplifier 21 for buffering, and a circuit 22A and a circuit 22A that apply DC bias in a desired direction, respectively. 22B. One circuit 22A biases the signal in the positive direction as shown and the other circuit 22B biases the signal in the negative direction. The outputs obtained from these circuits 22A and 22B are sent to the half-wave rectifier circuit 23A.
, 23B, and then supplied to the combiner 24.

Therefore, the synthesized output obtained from the synthesizer 24 is below the desired level, that is, from which noise components existing within the dead zone 7a in FIG. 2 are removed. Note that the nonlinear circuit applicable to the present invention is not limited to the above example, and may of course be a nonlinear circuit having input/output characteristics as shown in FIG. 6, for example.

In this case, the configuration is as shown in FIG. In the figure, 25A is an amplifier for the buffer, 26 is a double-wave rectifier circuit, and 27 is a comparison circuit, in which the rectified output is compared with a reference value. 27a is a reference value (voltage) supply terminal.

When the rectified output is larger than the reference value, the switch circuit 28 connected in parallel to the signal transmission path is opened (
OFF), otherwise it is ON. Therefore, the input/output characteristics shown in FIG. 6 can be obtained, so that the desired noise removal can be achieved. In addition, 25B
indicates a buffer amplifier. In this way, if the orthogonal transform outputs SO2 to SO4 are supplied, respectively, with the desired nonlinear circuits 16a to 16c interposed in the respective transmission paths,
Outputs obtained by removing noise components contained in high-order orthogonal transform outputs are obtained from each nonlinear circuit. These outputs and the above-mentioned orthogonal transform output SO1 are respectively supplied to the inverse transform circuit 30. In the case of the Hadamard transform, the inverse transform can also be performed using the same procedure as the transform circuit, so the inverse transform circuit 30 also has exactly the same configuration as the orthogonal transform circuit 20 described above.

Therefore, the above-mentioned orthogonal transformation output S is applied to the terminals 21a to 21d.
If Ol~SO4 are supplied, the inverse conversion outputs Si/~Si4' from which noise has been removed are output from the terminals 22a~22d, respectively.
will be obtained. Therefore, this inverse transformation output Si/~
By supplying Si4' to the serialization circuit 31 consisting of a delay line, an output signal SO from which noise components have been removed can be obtained from the terminal 32. Serialization circuit 3
1 is composed of three series-connected delay lines 31a to 31c similar to the parallel circuit 12, and the delay amount of each is selected to be 100 ns as described above.

Then, by supplying the above-mentioned inversely converted outputs to desired terminals of these delay lines 31a to 31c, the desired output signal SO, which is a continuous video signal corresponding to a unit area, can be obtained. In this way, if orthogonal transformation and its inverse transformation are used as the signal transformation method, the signal relationship between the orthogonal transformation and the inverse transformation will be exactly the same.

Therefore, it is possible to very easily remove the deterioration of the reproduced image due to the mismatch in amplitude, phase characteristics, etc. between the respective filters, which occurs when a low-pass filter or a high-pass filter is used. In addition, as mentioned above, the frequency band from which noise is removed is a high-frequency component that does not particularly affect human visual characteristics, so even if the noise is removed, the effect on the reproduced image will be small. few. However, this does not mean that there is no impact. This effect appears when orthogonally transforming contour parts, etc., and the image may be distorted accordingly. For convenience, the degree of such influence will be explained using two-dimensional orthogonal transformation. In the case of two-dimensional orthogonal transformation, video signals in two or more adjacent horizontal sections are respectively synchronized and orthogonally transformed.

For example, as shown in FIG. 9, fine patterns in the video signals SN-SN+1 in adjacent horizontal scanning sections may be orthogonally transformed. The example shown in the figure shows the case of 8th order conversion, and therefore the signals to be synchronized are from Sil to Si8.
If the time series of the input signal is used as it is and the unit area to be orthogonally transformed is expressed as a block as shown in FIG. We consider noise processing operations involving orthogonal transforms. As is well known, the 8th order Hadamard transformation matrix H8 is expressed by the following equation.

For convenience of explanation, if it is assumed that the level of signals other than the signal representing the contour 33 is zero, at the time when the contour 33 appears only in a part of the area to be orthogonally transformed as shown in FIG. 10A, the input signals Si4 and Only Si8 has a predetermined level a.

That is, the orthogonal transformation outputs SO1 to SO8 at the time point shown in A in the figure become the values shown by the equation (alternative) by using the equation (alternative).

Here, as shown in FIG. 8, the power spectrum for the signal frequency is such that the higher the frequency, the lower the power energy.

On the other hand, in the orthogonal transformation output, high-frequency components are distributed in the high-order transformation output compared to the low-order one, so the level of the high-order orthogonal transformation output is smaller than that of the low-order one. Therefore, among the orthogonal transformation outputs shown in equation (self), the higher-order outputs SO3, SO4...
The levels of . . . do not all take the same value a, but gradually decrease values. Therefore, these high-order output levels have a strong tendency to be below the maximum level of the dead zone 7a of the nonlinear characteristic, and therefore, when the orthogonal transformation outputs SO2 to SO8 are supplied to the nonlinear circuit 16, these orthogonal transformation outputs are not obtained at all. .

That is, in this example, only the lowest-order orthogonal transform output SO1 is supplied to the inverse transform circuit 30. When SOl=a and SO2-...=SO8=0, the inverse transformation outputs Sil' to Si8' are expressed by the following equations.

If the contents of the equation (self) are illustrated using a block similar to that shown in Fig. 10, it will become as shown in Fig. 11, and the inverse transformation output Si/~Si8' will be distributed over all unit areas. Become. In other words, by inversely transforming
The contour 33 shown in Figure A cannot be reproduced, resulting in image distortion.

This phenomenon is the same in the case of FIG. 10C. However, in the case of B in the same figure, among the orthogonal transformation outputs, only the first and fifth order outputs SO1 and SO5 are output, and in this case, the fifth order output SO
5 has a higher level than in the case described above, so there is an output even through the nonlinear circuit 16, so the inversely converted output is as follows.

If the contents of the formula (self) are shown in a diagram, it becomes as shown in FIG. 12, which essentially reproduces the pattern shown in FIG. 10B. As is clear from the above description, pattern distortion is particularly noticeable at the time of orthogonal transformation as shown in FIGS. 10A and 10C.

Therefore, the present invention aims to reduce the occurrence of such distortion and further improve the reproduced image.
By attenuating the output level of the inverse transform output corresponding to a unit area that is susceptible to distortion than the level of other inverse transform outputs and then combining them, that is, by performing weighted averaging, the target noise is removed from the output. It is designed to obtain signals.

For example, when removing noise using a two-dimensional and eight-order Hadamard transform as described above, the inverse Conversion output Si/, Si4', Si
5' and Sif levels Ll, L4, L5, and L8 are attenuated.

For example, choose a value that is half of the inverse transform output. In the case of a one-dimensional, eighth-order Hadamard transform, the level is selected such that its level gradually increases toward the central region, as shown in FIG.

Next, an example of the noise removal circuit according to the present invention will be explained, but in detail, the case will be based on a two-dimensional 8th-order Hadamard transform.
Figure 5 is a system diagram of the main parts.

Parts corresponding to those in the previous figure are given the same reference numerals and their explanations are omitted, but in this example, in order to achieve two-dimensional orthogonal transformation, two parallelization circuits 12 are provided before the orthogonal transformation circuit 20.
A, 12B are provided, and one parallel circuit 12
A 1H delay circuit 40 is interposed before the circuit B. The parallel circuit 12B is also constructed similarly to the other circuit 12A, so that 12d to 12f represent delay circuits selected for 100 ns, respectively. In this way, a pair of parallel circuits 12A, 12B
By providing these, the signals Sil to Si8 corresponding to minute unit areas in the video signals SN-SN+1 in adjacent horizontal scanning sections are synchronized and supplied to the orthogonal transformation circuit 20.

It goes without saying that the orthogonal transformation circuit 20 is constructed to satisfy the matrix of equation (2), and its construction will therefore be omitted. The plurality of orthogonal transform outputs SO1 to SO8 are supplied to the inverse transform circuit 30 via corresponding nonlinear circuits 16a to 16g. Here, the width Δw of the dead zone 7a of the nonlinear circuits 16a to 16g can be made constant, but
Of course, it may be varied depending on the amount of noise components contained in i.

The embodiment deals with the latter case, and the explanation will be given later. The plurality of inverse conversion outputs Sil' to Si8'' are supplied to corresponding level adjustment circuits 36A to 36G, respectively, and adjusted to a desired level relationship as shown in FIG.

Subsequently, these inverse conversion outputs Si/~Sif with different loads are combined via serialization circuits 31A and 31B, but the serialization circuit 31A has a 1H signal in order to serialize the signals as well as to synchronize the signals. A delay circuit 37 is provided. 38 indicates a synthesizer.

Note that it is of course possible to omit the level adjustment circuits 36B, 36C, 36F, and 36G provided on the transmission path of the inverse conversion output whose level does not need to be attenuated. The serialized outputs are fed to an attenuator 39 for averaging.

The amount of attenuation is the reciprocal of the sum ΣLn of the levels L1, . . ., L8 in the level adjustment circuits 36A to 36H. In this way, when the weight of the inverse transformation output Si/~Si8' is changed and then averaged, the inverse transformation output S
The spread of distortion is that much smaller than when simply averaging i/~Si8'', and image deterioration can be prevented.

In other words, in the case of simple averaging, the distortion is uniformly distributed over the entire unit area to be orthogonally transformed, and in this case, the distortion in the area where the signal component exists is the same as the distortion in the area where the signal component does not exist. appear more prominently. In other words, the distortion in the signal regions Si4 and Si8 in FIG. 10A is noticeable. On the other hand, in the present invention, the level of the inverse transform output corresponding to the region where distortion is conspicuous is made small in advance, so even if the distortion is distributed throughout, the distortion in the signal region appears to be small, resulting in image deterioration. can be prevented.

By the way, in the system diagram shown in FIG.
In order to vary the width Δw of the dead zone 7 m1 in No. 6, the following configuration may be used.

In this example, a noise component existing at a black level during a no-signal period of a video signal, for example, a vertical retrace period, is detected, and gate circuits 41a to 41g are provided on the transmission path of the orthogonal transform output. and orthogonal transform outputs are supplied to each of them.

On the other hand, 42 indicates a gate signal forming circuit, and the gate signal formed based on the input video signal Si is supplied to the plurality of gate circuits 41a to 41g described above, and a black level noise component in the vertical retrace period is extracted. . The extracted noise components are supplied to subsequent circuits 43a to 43g consisting of a rectifier circuit and a low-pass filter, and the power of the noise components is determined.

The signal representing this power is supplied as a width control signal of the dead zone 7a in the corresponding nonlinear circuit 16. For example, when there are many noise components, this power output becomes large. Control may be performed by changing the bias of the nonlinear element to widen the width Δw of the dead zone 7a, and by doing so, the noise component in the video signal can be effectively removed. It should be noted that it is not necessary to particularly use electric power as the width control signal, and it is sufficient to use a circuit configured to obtain an output according to fluctuations in the noise component.

As explained above, in the present invention, noise is removed by orthogonal transformation, so in addition to being able to remove noise more effectively than in the past, the present invention eliminates orthogonal transformation, which is particularly susceptible to image distortion. Weight the output corresponding to the area of
That is, since the final output is obtained by weight averaging, it has the characteristic that distortion in the signal region is small and image deterioration can be prevented. In the above-described embodiment, noise is removed using the eight-order Hadamard transform in two dimensions, but the selection and combination of dimensions and orders are arbitrary.

When a Hadamard transform of a higher order is used, the separation of signal components and noise components becomes more complete. Needless to say, in that case, the method of applying the load should be selected so that it gradually increases toward the central conversion region.

[Brief explanation of drawings]

Fig. 1 is a system diagram showing an example of a conventional noise removal circuit used to explain the present invention, Figs. 2 and 6 are input/output characteristic curve diagrams used in the Fufu nonlinear circuit, and Fig. 3 is the invention of the present invention. 4 is a circuit diagram showing an example of an orthogonal transform circuit, and FIGS. 5 and 7 are connection diagrams showing an example of a nonlinear circuit. Figure 8 is a curve diagram showing the power spectrum of the video signal, and Figure 9 is a curve diagram showing the power spectrum of the video signal.
A waveform diagram of a video signal used in dimensional orthogonal transformation,
FIGS. 10 to 12 are diagrams for explaining image distortion caused by orthogonal transformation, FIGS. 13 and 14 are diagrams showing an example of weighted average, and FIG. 15 is a diagram showing an example of the noise removal circuit according to the present invention. FIG. 12, 12A, 12B are parallelization circuits, 20 is an orthogonal transformation circuit, 16 is a nonlinear circuit, 30 is an inverse transformation circuit, 31, 31
A and 31B are serialization circuits, Si is a video signal that is an input signal, SO is a video signal from which noise has been removed, 36A to 36H are level adjustment circuits for loading, 37 and 40 are 1H delay circuits, and 39 is an attenuation circuit. , Sil to Si8 are synchronized video signals, and Si/ to Si8' are inverse conversion outputs.

Claims (1)

[Claims] 1. The input signal is supplied to an orthogonal transform circuit to perform orthogonal transformation, and the multiple orthogonally transformed transform outputs are supplied to a nonlinear circuit, and then supplied to an inverse transform circuit to remove noise and perform the inverse transform. An output signal from which noise has been removed is obtained by attenuating the level of a desired inverse conversion output among a plurality of outputs as desired and composing the outputs with different weights. Features a noise removal circuit.