JPH051178Y2 - - Google Patents

Description

[Detailed Description of the Invention] Technical Field The present invention relates to a noise removal circuit for video signals, and more particularly to a noise removal circuit configured with a cyclic filter using a frame memory to remove noise components contained in a video signal. It is something.

Background technology A video signal is a signal in which image information is repeated at a frame period, and autocorrelation between frames (hereinafter referred to as
(abbreviated as inter-frame correlation) is very strong. On the other hand, noise components included in video signals generally have almost no correlation between frames, so when video signals are averaged temporally for each frame period, the energy of the signal components hardly changes, and only the energy of the noise components decreases. I will do it. In view of this fact, there is a noise removal circuit, the basic configuration of which is shown in FIG.

In Figure 5, it is extracted from the broadcast wave,
A video signal obtained by reading from a recording medium is sent to an A/D (analog/digital) converter 1.
is digitized. This digital video signal passes through a multiplier 2 and becomes one input of an adder 3. The addition output of adder 3 is directly converted to D/A (digital/
It is converted into an analog signal by the analog) converter 4 and becomes a video output, and is also supplied to the frame memory 5 and the 1
By storing frame worth of data,
It is delayed by a period equivalent to one frame. The data read from the frame memory 5 passes through a multiplier 6, becomes another input to the adder 3, and is added to the video signal one frame later. As described above, the cyclic filter 7 using the frame memory is configured. This cyclic filter 7 has a transfer function H(z _F ) expressed as H(z _F )=(1-K)/(1-Kz _F ^-1 ), and is a low-pass filter in the time direction (between frames). It becomes a filter. Here, z _F ^-1 is a unit delay operator corresponding to a delay of one frame.

Cyclic filter 7 using this frame memory
According to , the noise has an average amplitude of √(1-K)/(1
+K) times. Therefore, the value of coefficient K is set to 1
If it approaches , it becomes a full-feedback recursive filter, which can reduce noise most efficiently.

By the way, there is movement in the video signal, and in the moving image section, a so-called blurring phenomenon occurs when image information delayed by a period equivalent to one frame is added one after another. In order to eliminate this blurring phenomenon, it is necessary to change the coefficient K that determines the feedback amount of the recursive filter 7 in the range of 1≧K≧0 according to the degree of movement of the video signal. For this purpose, a motion detection circuit 8 is provided to detect motion of the video signal. This motion detection circuit 8 uses a subtracter 9 to detect the difference (level difference) between the original video signal and the video signal delayed by a period equivalent to one frame in the frame memory 5, and converts this difference into a positive value using an absolute value circuit 10. The configuration is such that information indicating the degree of motion of the video signal is obtained by digitizing the motion detection information, and the coefficient K is set in the coefficient setting circuit 11 in accordance with this motion detection information. For example, in the still image section, K
By setting K = 1, noise can be removed, and in the video section, by setting K = 0, noise removal will be sacrificed to some extent, but it will be possible to suppress the above-mentioned blurring phenomenon and prevent deterioration of image quality. .

In the motion detection of the video signal using the difference described above,
In terms of spatial frequency, motion is detected by targeting signal components in the entire band related to motion between frames, and in terms of temporal frequency, relatively high components around 15 Hz are detected (the (See Figure 6).

Here, in the phenomenon of blur caused by movement of video signals, when considering what kind of components are conspicuous in terms of visual characteristics in the spatio-temporal frequency domain, the components are generally high spatial frequencies and slow-moving components;
In other words, one component corresponds to slow movement of the outline of the image, and the next component of interest is one that moves quickly at low spatial frequencies, that is, one that corresponds to fast movement of a large flat area on the screen. There is. Therefore, we consider the speed of movement and the distribution of spectra in the spatiotemporal frequency domain.

In Fig. 7, when looking at the spectral distribution a of vertical spatial frequency ν and temporal frequency f, if the image moves in the vertical direction, if the speed is fast, then a or c
The spectrum exists in the plane shown by , and if it is late, it is in the b or d plane. The difference between b and d or a and c is whether the direction of movement is upward or downward. The same thing can be said when looking at the spectral distribution b of the horizontal spatial frequency μ and the temporal frequency f. In other words, when the speed is fast, g and e, and when it is slow, f,
The spectrum is distributed in the plane of h. G and e or f and h are distinguished depending on whether the direction of movement is left or right. Therefore, the components that move slowly at high spatial frequencies are in the regions a and b indicated by X, and the components that move quickly at low spatial frequencies are the regions indicated by Y. however,
Since the positions on the ν and μ axes are the positions where the still image spectrum exists, we will consider excluding them. Moreover, the broken line is a region where the power of the luminance Y signal is large.

As is clear from the above explanation, in conventional devices that perform motion detection based on the magnitude of the absolute value of the difference between frames, the high spatial frequency component, that is, the slow movement of the outline of the image, actually causes blurring. In this case, the motion component cannot be detected, or a component that is not very noticeable due to the fast movement of the contour part may be detected, so the feedback amount of the recursive filter 7 is Control cannot be performed according to visual characteristics.

Summary of the invention This invention was created in view of the points mentioned above.
It is an object of the present invention to provide a video signal noise removal circuit that can efficiently remove noise using interframe correlation within a range that does not cause blurring in visual characteristics.

The noise removal circuit according to the present invention detects the level of at least one of the high spatial frequency - low temporal frequency component (excluding DC component) and the low spatial frequency - high temporal frequency component of the video signal, and detects the level of at least one of the high spatial frequency - low temporal frequency component (excluding DC component) and The structure is such that the amount of feedback of a cyclic filter using a frame memory is controlled according to the amount of feedback.

Embodiments Hereinafter, embodiments of the present invention will be described in detail based on the drawings.

FIG. 1 is a block diagram showing an embodiment of the present invention, in which parts equivalent to those in FIG. 5 are designated by the same reference numerals. In the figure, an input video signal A is converted into a digital signal by an A/D converter 1, and then input to a cyclic filter 7A. The cyclic filter 7A has a frame memory 5, (1-K) times,
K times multipliers 2 and 6, and an adder 3 that adds the respective outputs of these multipliers 2 and 6, that is, the original video signal and the video signal delayed by a period equivalent to one frame;
Y/C separation circuit 12 that separates the video signal output from this adder 3 into a luminance Y signal and a color C signal.
, a chroma inverter 13 that inverts the phase of the color signal separated by the Y/C separation circuit 12, and a chroma inverter 13 that inverts the phase of the color signal separated by the Y/C separation circuit 12;
It consists of an adder 14 that adds the luminance signal separated by the C separation circuit 12, and the output of this adder 14 is directly supplied to the frame memory 5, and the D/A converter 4 converts the analog signal is converted into video output B. Note that the chroma inverter 13 is provided in order to match the phase of the color signal between frames, since the phase of the color signal differs by 180° from frame to frame in the NTSC video signal.

The motion detection circuit 8A is configured to detect, based on the luminance signal separated by the Y/C separation circuit 12, a component that is thought to cause motion blur due to visual characteristics. That is, in the motion detection circuit 8A, a low spatial frequency component D is obtained by passing the luminance signal component through an LPF (low pass filter) 15 in a two-dimensional space, and this low spatial frequency component D is converted into a luminance signal by a subtractor 16. By subtracting from the signal component Y, a high spatial frequency component E is obtained. The low spatial frequency component D and the high spatial frequency component E correspond to the shaded portions in FIGS. 2a and 3a, respectively. In addition, in FIGS. 2 and 3, ν,
μ and f indicate vertical spatial frequency, horizontal spatial frequency, and temporal frequency, respectively.

When the high spatial frequency component E passes through a BPF (band pass filter) 17 in the time direction, the spatiotemporal component G shown in FIG. 3b and c, that is, the high spatial frequency - low temporal frequency component is obtained. .
On the other hand, the low spatial frequency component D is the HPF in the time direction.
(High pass filter) By passing through 18,
A spatiotemporal component F, ie, a low spatial frequency-high temporal frequency component, shown in FIG. 2b and c is obtained. As a result, components with noticeable motion blur are extracted from the video signal. The spatiotemporal components G and F are supplied to level detection circuits 19 and 20, and the absolute value of each level is detected. The detection outputs I and H of the level detection circuits 19 and 20 are added by an adder 21, and this addition output J becomes a control input of the coefficient setting circuit 11. The coefficient setting circuit 11 has a characteristic that the coefficient K is inversely proportional to the control input J (see Fig. 4), and when the value of the control input J is close to 0, it is regarded as a still image part and the coefficient K is set to 1. Set it to a value close to
Conversely, when the value of J is large and the component that causes motion blur is large, the coefficient K is set to a value close to 0 in order to prevent image quality deterioration due to blur.

Note that the above-described series of operations is performed for each pixel or for each certain pixel block (m lines×npels).

In this way, in the moving image section, attention is focused only on spatiotemporal frequency components where motion blur is easily noticeable due to visual characteristics, and the coefficient K that determines the feedback amount of the recursive filter 7A is controlled according to the level of the component. As a result, noise components with no inter-frame correlation can be removed efficiently and without causing visual motion blur.

In the above embodiment, the levels of both the high spatial frequency-low temporal frequency component and the low spatial frequency-high temporal frequency component of the video signal are detected,
Although the configuration is such that the amount of feedback of the cyclic filter 7A is controlled based on both levels, the level of at least one of both components is detected and the amount of feedback of the cyclic filter 7A is controlled based only on that level. The structure may be configured as follows, and corresponding effects can be obtained.

Effects of the invention As explained above, the noise removal circuit according to the invention controls the feedback amount of the recursive filter using frame memory based only on the spatiotemporal frequency components that cause motion blur due to visual characteristics in the moving image part. With this configuration, noise components with no inter-frame correlation can be removed efficiently and without causing motion blur.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention.
Figures 2 a to c are diagrams showing detected components in the low spatial frequency domain, Figures 3 a to c are diagrams showing detected components in the high spatial frequency domain, and Figure 4 is a diagram showing the coefficient setting circuit in Figure 1. Figure 5 is a block diagram showing the basic type of noise removal circuit using inter-frame correlation, Figure 6 is a diagram showing time-frequency characteristics of inter-frame difference detection, and Figures 7 a and b are time-frequency characteristics. FIG. 3 is a diagram showing the spectral distribution of image motion components in the spatial frequency domain. Explanation of the signs of the main parts, 2, 6... Multiplier, 5
...Frame memory, 7,7A...Cyclic filter, 8,8A...Motion detection circuit, 11...Coefficient setting circuit.

Claims (1)

[Scope of utility model registration request] A noise removal circuit configured with a recursive filter using a frame memory to remove noise components contained in a video signal, the circuit comprising: a high spatial frequency, a low temporal frequency component, and a low spatial frequency of the video signal; Noise removal of a video signal characterized by comprising: a detection means for detecting the level of at least one of the high time frequency components; and a control means for controlling the feedback amount of the recursive filter according to the detection level. circuit.