JPS63260277A - Movement adaptive type picture quality improvement device - Google Patents

Abstract

PURPOSE:To improve the picture quality by detecting the movement at each picture element from preceding and succeeding frames, using a time area filter at a still region so as to suppress noise, using a spatial area filter so as to compensate the contour and preventing the fog due to movement at the moving region. CONSTITUTION:A value being the subtraction (2) of a present frame from a value of one preceding frame for each picture element and the present frame value are given to a D1 delay 6 in the time area filter. A difference signal between frames through the D1 delay 6 is sent to a nonlinear circuit 7, in which the gain is the unity in case of the still region and the limiter type is adopted in case of the moving region. An output of the circuit 7 and the value of the present frame are added (8). The added result (8) is fed to an LPF 11 and a D2 delay 12 of the spatial area filter, a high frequency component by subtracting (13) both outputs is given to a nonlinear circuit 14, the result is added (15) to the output of the D2 delay 12 and the result is outputted (16). Thus, the noise is suppressed in the still picture region and the contour is compensated, and the fog due to movement is prevented in the moving picture region to improve the picture quality.

Description

Detailed Description of the Invention (Field of Industrial Application) The present invention provides motion adaptive image quality for suppressing noise components included in image signals and improving sharpness in equipment that handles images such as television receivers. This relates to an improvement device.

(Prior art) Spatio-temporal smoothing has traditionally been used as an effective noise reducer for moving images such as television signals, changing the characteristics of noise reduction in the spatial domain and noise reduction in the time domain by adapting them to temporal changes in the image. (Communications Society Journal Vol. 1. J70-8-1: Hejima, Sawada RH
Intra-frame/inter-frame adaptive extrapolation predictive coding of DTV signals").

The operation of this filter is to compare the signal of the previous frame processed by the time domain filter and the current frame signal, and if there is no difference, the signal is filtered by a spatial domain low pass filter (LPF).
), and if there is a difference, apply LPF to a certain extent of amplitude change, thereby reducing noise without causing motion blur in moving images.

(Problems to be Solved by the Invention) However, the above-described conventional spatio-temporal filters have been used mainly as pre-processing in signal encoding to reduce noise. Therefore, contour compensation in television reception lines and the like is not considered. In this process, the image may become blurred, but its sharpness will not increase.

In addition, adaptive processing is performed independently for each pixel based on the time change of each pixel, so if there is a lot of noise even in a still image, especially pulse noise, it may be detected as motion and smoothed by a spatial filter. become. Conversely, even in a moving area of an image, depending on the shape of the image, some pixels may become stationary, or even if the contrast of the shape change is low, the pixels do not change much over time and appear to be stationary. Put it away.

From the above points, conventional spatio-temporal filters tend to perform inappropriate filter operations, and the resolution tends to decrease. In particular, when there is a lot of noise, the time domain filter does not work effectively, which poses a problem.

SUMMARY OF THE INVENTION Therefore, it is an object of the present invention to provide a motion adaptive image quality improvement device that solves the problems of the conventional techniques described above.

(Means for Solving the Problems) In order to achieve the above object, the present invention converts the difference between the previous frame value and the current frame value into an absolute value, and passes the difference between the previous frame value and the current frame value through a first low-pass filter with a narrow band. a motion detection means for obtaining a detected value; a subtraction circuit for subtracting a current frame value from a previous frame value processed in a time domain; and a subtraction circuit for subtracting a current frame value from a previous frame value processed in a time domain; A first delay device for compensation, a delayed subtraction result obtained by passing the subtraction result of the subtraction circuit through the first delay device, and the motion detection value are respectively inputted, and an output gain with respect to this input delay subtraction result is inputted. a first nonlinear circuit that changes and outputs according to both input values, and adds the output of this first nonlinear circuit and a delayed current frame value obtained by delaying the current frame value through the first delay device. a first processing means that performs processing in the time domain using an adder circuit that obtains an output; and a second low-pass filter that compensates for the delay of the second low-pass filter in the spatial domain with respect to the processing output of the first processing means in the time domain. a subtraction circuit that subtracts the processing output in the time domain passed through the second delay device from the value passed through the second low-pass filter; a second non-linear circuit which inputs the motion detection values passed through the second delay device and outputs the input subtraction results by changing the output gain according to both input values;
and a second processing means for performing processing in the spatial domain using an addition circuit that obtains an output by adding the output of the nonlinear circuit and the processing output in the time domain through the second delay device. The present invention provides a motion-adaptive image quality improvement device having the following characteristics.

(Example of implementation) The motion adaptive image quality improvement device according to the present invention appropriately detects motion in a moving image, sufficiently reduces noise with a time domain filter in a still region, and performs contour compensation with a spatial domain filter. The sharpness of the image is increased, and in the motion domain, the noise reduction of the temporal domain filter is minimized to prevent motion blur, and the low-pass filter (L) is used in the spatial domain filter.
PF) is applied to suppress noise.

FIG. 1 is a block diagram showing an embodiment of a motion adaptive image quality improvement device according to the present invention.

In the figure, an input terminal 1 receives a two-way signal such as a television signal. 2.13 is a subtracter, 3 is a limiter, 4 is an absolute value (ABS) circuit, and 5.11 is a low pass filter (LPF).

The D1 delay 6 is a delay device with a delay amount D1 that compensates for the delay of the LPF 5. 7.14 is a nonlinear circuit, 8.1
5 is an adder.

(IF-Dl) Delay 9 is the delay device D from 1 frame.
This is a delay device with a delay amount smaller by 1.

D2 delay 10.12 is a delay device with delay lD2 that compensates for the delay of LPFII. 16 is an output terminal.

The first half of the device (the left half of the figure) constitutes a time domain filter, which is basically a cyclic filter with a delay of one frame. Further, the latter half of the device (the right half of the figure) constitutes a spatial filter, has a one-dimensional or two-dimensional LPFll, and depending on the characteristics of the nonlinear circuit 14, can be used as a contour compensation filter or an LPF. Also, D1 delay 6 and D2 delay 10°12 are each LPF5. This is to compensate for each delay of the LPF II and maintain the correct phase relationship of the LPF.

Next, the operation of each part in the above configuration will be explained.

First, it is a time domain filter, but 1 is cyclically added.
A subtracter 2 subtracts the value of the current frame from the value of the previous frame for each pixel to obtain a difference. This difference component is passed through 01 delay 6 in order to compensate for the delay of LPF 5 in the motion detection circuit, which will be explained later. Also, in order to adjust the timing in the later addition, the current frame value is also D
Pass through 1 delay 6.

The interframe difference signal that has passed through the D1 delay 6 is processed by the nonlinear circuit 7 as follows. First, in a stationary region, in order to strongly enhance the effect of the noise reducer, the linear output is set to a state close to the gain "1". As a result, the time-domain filter forms an LPF with a fairly narrow band.
Random noise is greatly reduced. On the other hand, in the motion area, a normally used limit-type characteristic is used, and as the input value increases, the gain decreases rapidly above a certain value to prevent blurring due to motion. An example of the input/output characteristics of this nonlinear circuit 7 is shown in FIG. Here, the negative side is symmetrical. Moreover, this nonlinear circuit 7 has an internal limiter, and when the input signal level is about 0 to 255n, the limiter is applied at about 64, and the same processing is performed for higher levels.Here, , Usually, when the difference between frames is large, it is a motion region, and in that case, the characteristics of a nonlinear circuit are such that all outputs are “0” above a certain input value.
'', so there is no problem even if the limiter described above is applied.

By adding the output of the nonlinear circuit 7 obtained in this manner and the current frame value in an adder 8, a noise-reduced output is obtained.

Next is the spatial area filter, where LPFll is a method that changes the two-dimensional characteristics of the filter, such as operating two types of filters, vertical and horizontal, separately, depending on the spatial shape of the image. However, in this case, there is no two-dimensional change in characteristics, and the change in characteristics is assumed to be one-dimensional.

The configuration of the spatial band filter is as follows: First, the LPF is applied to the input signal.
From the value multiplied by ll, the input signal value passed through the D2 delay 12 that compensates for the delay of PFll is subtracted by the subtracter 1.
3 will reduce your dignity. Here, this LPFll may have a one-dimensional configuration or a two-dimensional configuration, and more ideal characteristics can be obtained in the case of a two-dimensional configuration. A simple example is shown in FIG. In FIG. 5, 1H delay is a delay for one line, ■ is a delay for one dot, and e is an adder.

In addition, FIG. 7 shows the pixel coefficients of the secondary LPF in FIG. 1 each
Can be multiplied by /16.

As mentioned above, from the output of L, PFll, its LPF11
Since the unmultiplied value is subtracted by the subtracter 13, a difference is also obtained in the frequency range, and the high frequency component of the signal is obtained. This high frequency component is passed through the nonlinear circuit 14, but if this nonlinear circuit 14 is linear and the gain is
1", the value of only the delay will be canceled out by adding the value of only the delay later, and the output of LPFll will be output as is. On the other hand, the output of the nonlinear circuit 14 will be 0". If there is, the value of only the delay will be output as is, and no processing will be performed.

Conversely, if the gain of the nonlinear circuit 14 is negative,
The high-frequency components of the signal are added in positive phase, and contrary to the case where the gain is positive and the high-frequency components are reversed, the high-frequency components of the output signal increase and contour compensation is performed. It will be done.

Therefore, if you want to apply L P F 11, you can make the input/output polarity of the nonlinear circuit 140 positive, and if you want to perform contour compensation, you can make it negative. When the nonlinear circuit 14 is used, such processing changes can be set according to the level of the high frequency component, and the nonlinear characteristics are determined thereby.

Here, how to perform this processing is to apply LPFll to reduce noise in places where the image is spatially flat and have little change in high frequencies, and to reduce noise in places where there are changes in the edges of the image etc. Contour compensation is applied to areas where there are many area components to increase the sharpness of the image. However, if too much contour compensation is performed, the image will become unnatural, so a limiter is applied to the value that changes the image due to contour compensation.

Next, regarding the adaptive processing between the still region and the moving region, in the still region the noise component is sufficiently suppressed by the time domain filter, so LPFll is used only when the high frequency component is quite small, and in other cases In this case, contour compensation is applied relatively strongly.

Since the noise is sufficiently small, contour compensation makes it possible to obtain an image with high sharpness without noticeable noise. On the other hand, in the motion domain, noise reduction by the time-domain filter is not sufficient and a considerable amount of noise is included, so LPFll should be applied to the level of high-frequency components with relatively large amplitudes, and contour compensation should not be performed too much. . In this case, the image becomes slightly blurred and the sharpness decreases, but since the image is moving, there are not many high-frequency components in the first place, and it is difficult to detect visually, so this is not a big problem.

By performing such motion-based adaptive processing in multiple stages and making smooth changes, image quality can be improved without looking unnatural. FIG. 3 shows an example of the input/output characteristics of the nonlinear circuit 14 in the above spatial domain. Here, the negative direction of the input is symmetrical, the absolute value of the input is limited to "32", and the output beyond that is the same. This is because it does not cause a large change in the signal, and because it processes nonlinear characteristics. This is done for the sake of ease.

Note that these nonlinear characteristics are stored in a ROM (Read 0nl
table lookup (Ta
bleLook up), but if the motion adaptation is set to 4 levels, the ROM capacity will be reduced to 4 by the nonlinear circuit 7.
This means that the nonlinear circuit 14 may have a small capacity of about 1 bit.

Next, a description will be given of detection of a moving region or a still region, which serves as a reference for adaptive processing in the time domain and sky@ domain as described above.

The basic differences between the device of the present invention and the conventional device are as follows:
The point is that by applying a narrow band LPF 5 to the motion detection output, a fairly strong correlation is created for each pixel.

Normally, motion in an image does not change significantly from pixel to pixel and has a fairly strong correlation. Therefore, by creating a strong correlation with the motion output, we can significantly improve false detections due to noise components and false detections when an image with high-frequency components moves in the spatial domain compared to when detecting each pixel. It is something.

Its operation is first to detect the difference signal between frames, and since this is obtained in the time domain filter as the output of the subtracter 2, it can be used as is. Therefore, by applying a simple limiter to the output of the subtracter 2 using a limiter 3 and converting it into an absolute value using an absolute value (ABS) circuit 4, the motion of each pixel can be obtained. LPF 5 which narrows the band considerably for the output of this ABS circuit 4
A strong correlation is created by passing through the pixels, so that even if the movement of each pixel changes, the final information as to whether it is an active area or a static area does not change much from pixel to pixel.

Here, the limiter 3 is used to reduce the possibility that a still area is mistakenly detected as a moving area due to strong pulse noise, and to reduce the difference in the motion detection level due to the difference in the degree of signal change (contrast ratio). used. The reason for the latter is that linear motion detection would detect the spatial changes in proportion to the contrast ratio of the image, even though the motion areas are similar. This is to make it smaller than the ratio.

On the other hand, by installing such a limiter 3, the ABS
This is also useful for reducing the number of processing bits in absolute value conversion by the circuit 4 and digital processing by the LPF 5. An example of the characteristics of this limiter 3 is shown in FIG.

The problem here is the characteristics of the LPF 5, but considering the nature of the image, it is desired that the LPF 5 has a narrower band than the LPF 1 used in the spatial domain.

FIG. 6 is a diagram showing the configuration of the LPF 5. In the same figure, 1 day delay is a 1-line delay, ■ is a 1-dot delay, 9T is a 9-dot delay, and Φ is an adder. It is.

In the example of LPF5 shown in Fig. 6, the vertical direction is the same as LPFll in the spatial domain, but the horizontal direction is of the same coefficient type, and by adding, delaying, and subtracting, the band can be changed using a simple circuit. It's narrow.

Further, FIG. 8 shows the pixel coefficients of the secondary LPF in FIG. 6 in correspondence with the secondary pixel arrangement, and each pixel coefficient ("1") shown in the diagram is multiplied by 1/27.

Note that LPF 5 here does not directly change the image quality, unlike LPF 1 in the spatial domain, so it is not very strict, and considering the variety of image movements, what is the best one? It is difficult to determine if there is.

(Effects of the Invention) As described above, with the motion adaptive image quality improvement device of the present invention, noise is not significantly reduced even in still images using a time domain filter, whereas in the past, when there are relatively many noise components, Noise is sufficiently reduced by determining that a certain area is a stationary area. Furthermore, pulse noise is similarly reduced by the time domain filter.

In such a case, the spatial filter is not used much and the PF does not act, so there is little deterioration in sharpness. Further, in the apparatus of the present invention, when it is determined that the area is a stationary area, contour compensation is applied, and it is possible not only to simply reduce noise but also to improve the sharpness of the image. In this case, since the signal with reduced noise is processed, the noise does not stand out due to contour compensation, and overall excellent image quality can be obtained. on the other hand,
If it is determined that the area is a moving area, the time domain filter operates in the same manner as in the past, and even if a still area is erroneously determined to be a moving area, there will be no major problem. Furthermore, in the motion domain, noise reduction is mainly performed in the spatial domain, so that the final image does not contain much noise.

Furthermore, each process changes continuously and smoothly, and no unnaturalness occurs due to switching.

These adaptive processes are less likely to malfunction even in response to pulse noise and the like, and are less likely to misjudge moving areas and stationary areas depending on the spatial shape or contrast ratio of the image. Furthermore, since this processing changes continuously for each pixel, there is no unnaturalness caused by large differences in processing between adjacent pixels or changes in processing at block boundaries as in block-by-block detection.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the motion adaptive image quality improvement device of the present invention, and FIGS. 2 and 3 are diagrams showing input/output characteristics of a nonlinear circuit constituting an embodiment of the device of the present invention. ,
FIG. 4 is a diagram showing the characteristics of a limiter constituting an embodiment of the device of the present invention, FIGS. 5 and 6 are diagrams showing the configuration of a PF constituting an embodiment of the device of the present invention, and FIG. and FIG. 8 are diagrams showing pixel coefficients of the second-order LPF in FIGS. 5 and 6, respectively. 1... Input terminal, 2.13... Subtractor, 3... Limiter, 4... Absolute value (ABS) circuit, 5.11...
・Low pass filter (LPF), 6...D1 delay, 7.14...Nonlinear circuit, 8.15...Adder,
9...(1F-D+) delay, 10, 12...D
2 delay, 16...output terminal.

Claims (1)

[Scope of Claims] Motion detection means for converting the pixel-by-pixel difference between the previous frame value and the current frame value into an absolute value and obtaining a motion detection value through a first low-pass filter with a narrow band; a subtraction circuit that subtracts the current frame value from the frame value; a first delay device that compensates for the delay in the first low-pass filter in the motion detection means with respect to the current frame value;
The subtraction result of the subtraction circuit is passed through the first delay device, and the delayed subtraction result and the motion detection value are respectively input, and the gain of the output with respect to the input delay subtraction result is changed according to both input values. a first nonlinear circuit that outputs
A first process that performs processing in the time domain by an adding circuit that obtains an output by adding the output of the first nonlinear circuit and a delayed current frame value obtained by delaying the current frame value through the first delay device. means, a second delay device for compensating the delay of the second low-pass filter in the spatial domain for the processing output of the first processing means in the time domain, and a value from the value passed through the second low-pass filter. A subtraction circuit that subtracts the processing output in the time domain passed through the second delay device, and the subtraction result of this subtraction circuit and the motion detection value passed through the second delay device are respectively inputted. a second nonlinear circuit that changes the output gain for the subtraction result depending on both input values, and outputs the output of the second nonlinear circuit and the processing output in the time domain through the second delay device. 1. A motion adaptive image quality improvement device comprising: an addition circuit that performs addition to obtain an output; and second processing means that performs processing in a spatial domain.