JPH0213069A - Noise removing device - Google Patents

Abstract

PURPOSE:To obtain a still picture in which noise is effectively removed by executing noise removal through a time low-pass filter when a change quantity in a time base direction is equal to or below a normal value V1, through a space low-pass filter when it is equal to or above a normal value V2, and in mixing and fetching outputs from both filters when a change quantity V is between the V1 and V2. CONSTITUTION:By a movement detecting part 2, a movement quantity in each picture element unit is detected. When the movement quantity V of an image is smaller than the lower limit V1, only the output signal of a time low-pass filter 3 is sent and stored to a frame memory 9. To the contrary, only the output signal of a space low-pass filter 4, in which the movement quantity V of the image is larger than a rise V2, is sent and stored to the frame memory 9. When the movement quantity V is in a medium degree, the output, in which both outputs of the time low-pass filter 3 and space low-pass filter 4 are multiplied with coefficients K and 1-K, is stored in to the frame memory 9. Thus, to any image, the still picture can be obtained in which the noise is effectively removed from a video signal without omitting a signal.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a noise removal device, and more particularly to a noise removal device suitable for obtaining a still image from which noise has been removed from a video signal.

(Conventional technology) In video signals, adjacent pixels and pixels between frames have a fairly strong correlation.On the other hand, random noise has no correlation, so by averaging the pixels with a strong correlation, However, horizontal and vertical averaging processing is equivalent to applying a two-dimensional low-pass filter, and low-frequency noise components superimposed on nearby pixels are reduced due to correlation. Since it cannot be removed, other distortions occur, such as abnormal noise patterns and image blurring.On the other hand, averaging processing between frames can yield good results.

FIG. 4 shows a circuit configuration for performing conventional inter-frame averaging processing, and is generally called a digital noise reducer. This is a cyclic filter using a one-frame delay circuit 11. Although it is possible to reduce noise using an acyclic filter, a cyclic filter is used because it requires a large number of expensive frame delay circuits and the problem of group delay can be practically ignored in interframe processing. . In Figure 4, the phase of the C signal of chroma inverter 12 and NTSC A8 is inverted every frame, so
This is to make them consistent.

FIG. 5 shows the frequency characteristics of the noise reducer shown in FIG. 4. In the case of a still image, there are video signal components at the peaks of the comb, and only noise components at the valleys. By attenuating this valley portion, noise is reduced. The degree of noise improvement obtained with this configuration is as follows: SN ratio improvement rate = 101og ((14K) / (1-K
)) [Uko0<K<1. Figure 6 shows the relationship between K and the SN non-improvement rate. In the case of a moving image, a video signal component exists in the valley portion of FIG. 5, and if this component is removed, an afterimage will occur (blurring of the image due to the low-pass filter on the time axis). Therefore, as shown in FIG. 4, a motion detection section 13 is provided to perform processing to reduce K when motion is detected.

In this way, conventional noise removal devices use a so-called temporal low-pass filter that accumulates signals in the time axis direction, and also provides feedback according to the difference between the moving image part and the still image part of the input video signal. Noise was removed by changing the constants.

(Problems to be Solved by the Invention) However, such conventional devices have the following drawbacks.

If noise is removed using a 0-hour low-pass filter, the resolution of the video portion will drop.

■If the feedback constant of the filter is changed depending on the moving image portion and the still image portion of the video signal to maintain the resolution of the moving image portion, the noise removal ability for the moving image portion will be degraded.

As described above, the conventional apparatus does not effectively remove noise from both the moving image portion and the still image portion.

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to eliminate the above-mentioned drawbacks and to be able to obtain a still image with noise effectively removed from the input video signal, regardless of whether the input video signal is a moving image portion or a still image portion. It is an object of the present invention to provide a noise removal device that can be used as a signal source for a printer or the like to obtain high-quality hard copies.

[Means to solve the problem]

In order to achieve such an object, the present invention includes a motion detection unit that detects the amount of change V in the time axis direction for a certain pixel of interest in a video signal, and a temporal low-pass filter that removes noise in the video signal based on correlation between frames. , if the output V of the motion detection section and the spatial low-pass filter that removes noise in the video signal through spatial filtering processing is less than the lower limit specified value ■1, only the output of the temporal low-pass filter is passed through, and the output V
is the upper limit specified value■ If it is 2 or more, only the output of the spatial low-pass filter is passed through, and when the output V is between v1 and v2, the output of the temporal low-pass filter and the output of the spatial low-pass filter are mixed and passed through 0. It is characterized by having the following.

(Function) With the above configuration, the present invention removes noise via a temporal low-pass filter in the case of a still image portion where the change amount V in the time axis direction is equal to or less than a specified value V for a pixel of interest with a video signal. If the amount of change V is greater than a certain specified value v2, noise is removed via a spatial low-pass filter, and the image is an intermediate image between static and dynamic where the amount of change V is between 1 and v2. In the case of the part, noise is removed by mixing and extracting the outputs from the temporal low-pass filter and the spatial low-pass filter, so that the noise can be effectively removed from the video signal without any signal loss for any image. A still image from which noise has been removed can be obtained, and this still image can be used as a signal γ source for a printer or the like to obtain a high quality hard copy.

〔Example〕

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

FIG. 1 shows the circuit configuration of a noise removal device according to an embodiment of the present invention. In this figure, 1 is a video input terminal into which a digitized video signal is manually input, and 2 is a function that detects the amount of movement in each pixel of the manually input video signal and generates an output proportional to the amount of movement. This is the detection section. The motion detection method in the motion detection unit 2 can be realized by a method in which the hour wheel considers a difference in pixel level between frames to be motion if it is above a certain level, but the ability to discriminate between noise and motion is low. Therefore, we have developed methods such as using visual characteristics for noise, orthogonally transforming sample data within a frame to distinguish between noise and motion, or calculating and creating a motion vector based on inter-frame correlation. Use methods.

3 is a general temporal low-pass filter (temporal LPF) that removes noise from the correlation between frames. By the way, components with weak correlation between frames are noise and motion components, but when looking at the pixels in the vicinity of one frame within a frame, there are motion components with strong correlation. On the other hand, noise components are often isolated points even within a frame. Therefore, the temporal low-pass filter 3 is configured to, for example, extract only weak correlations both between frames and within frames, and remove these as noise. That is, in order to investigate the intra-frame correlation, the Brevi screen is subdivided into small regions, and each region is subjected to orthogonal transformation using, for example, Hadamard transformation j% (hadama
rd transform). Considering the fourth-order Hadamard transformation, the transformation matrix is as follows, and if aQ"-'a3 is the pixel extracted as shown in Figure 7, fo""f3 is the average brightness and the vertical direction, respectively. pattern, horizontal direction pattern, and diagonal direction pattern.Since white noise has a small two-dimensional correlation, it is not concentrated in a specific conversion output, but is dispersed to the same extent in each conversion output.

Therefore, the noise can be extracted by a circuit that passes only a small amplitude of the converted output. In the actual configuration, a 4×2 Hadamard transform with four pixels in the horizontal direction and two pixels in the vertical direction is used. First, the 8 square elements of the frame difference signal are converted into parallel signals by serial-parallel conversion, and subjected to Hadamard transformation.

Next, the converted output is passed through a circuit that extracts only small amplitude signals, then inversely converted, and the noise component is obtained by parallel-to-serial conversion.

A signal obtained by removing this noise component from the original signal is output from the temporal low-pass filter 3.

4 is a general spatial low-pass filter (spatial LPF) that reduces noise by low-pass filtering. For example, U and V are spatial frequencies, and the Fourier-transformed image F (u, v), The transfer function of the filter is defined as II (
u, v), then filtering generally becomes G(u, v)-If(u, v)F(u, v). This is inversely Fourier transformed and output from the spatial low-pass filter 4. While noise generally has a spectrum that extends to high frequencies, the spectrum of an image is concentrated in a low frequency range, so image noise can be reduced by such filtering. Furthermore, erasing a specific spatial frequency region is effective in removing periodic disturbances superimposed on an image.

Note that in order to remove stone-like noise, for each pixel, compare the density value of that pixel with the density values of surrounding pixels, and if the number of pixels with similar density is small,
The spatial low-pass filter 4 is considered to be noise and replaces the density value with the average density of surrounding pixels.The spatial low-pass filter 4 includes a filter based on the υ-divided average method, a median filter, and a filter based on selective local averaging. A smoothing filter based on averaging processing such as a filter can also be used.

Reference numeral 5 denotes a lookup table (LIIT) using ll0M (read-on memory), RAM (random access memory), etc., which reads the level of the output V of the motion detection section 2 and uses it as a read address to read the second data according to the level. The output value of 1- is output to the coefficient as shown in the figure.

6 is a multiplier that multiplies the output of the temporal low-pass filter 3 and the coefficient of the look-up table 5, and 7 is a multiplier that multiplies the output of the spatial low-pass filter 4 and the coefficient 1- of the look-up table 5.
8 is the output of multiplier 6 and multiplier 7.
This is an adder that adds the output of Reference numeral 9 denotes a frame memory (FM) in which the video signal from which the noise manually generated from the adder 8 has been removed is stored as a still image on a screen-by-screen basis. This stored signal is then output through the video output terminal 10 to an external device as a still image video output.

Next, the operation of the embodiment of the present invention will be explained.

In the above configuration, when a digitized video signal is input to the video input terminal 1, the motion detection section 2 first detects the motion (2) in each pixel unit, and the output V proportional to the detection (++'i) is detected. This output (2) is converted in the look-up table 5 as shown in FIG.

In other words, the image motion fluorescence ■ is smaller than the lower limit V1 (V〈
■, ), it is determined that it is a still image and the output signal of the temporal low-pass filter 3 is sent only to the frame memory 9.
is sent to and stored. On the other hand, the amount of movement in the image is the upper limit.
When (V2) is larger than 2, the signal becomes K-0, so only the output signal of the spatial low-pass filter 4 is sent to the frame memory 9 and stored therein.

Since the motion image of the video signal is accumulated and recorded at 1/60 seconds at the time of shooting, it is equivalent to being filtered through a kind of spatial low-pass filter. There is no change in the signal even after filter 4 is applied. When the amount of motion V is medium (V+ < V < V2), the output of both the temporal low-pass filter 3 and the spatial low-pass filter 4 is multiplied by the coefficients K and 1-K shown in FIG. 2, and the output is stored in the frame memory 9. Write ↑, set to 0.

In this way, by appropriately using a temporal low-pass filter for the still image portion, a spatial low-pass filter for the moving image portion, and the sum of the respective intermediate images for still and moving images, it is possible to It is also possible to obtain a still image from which noise has been effectively removed from the video signal even for an image of 18.

FIG. 3 shows a circuit configuration of another embodiment of the present invention. In this figure, 41.1!2 are spatial low-pass filters with different cutoff frequencies, and 71.72 are multipliers. The other components are the same as the embodiment shown in FIG.

As described above, in the embodiment shown in FIG. 3, a plurality of spatial low-pass filters 41 and 42 having different cutoff frequencies are provided to further increase the noise removal ability. That is, the cutoff frequency of the spatial low-pass filter 42 added to the above-described device is smaller than that of the spatial low-pass filter 41. Also, if the amount of movement is large, the high frequency component of No. 13 is lost, so when V<V3<V2, the coefficient input to the multiplier 6 becomes 0, and the coefficient IK input to the multiplier 72 becomes 1.
The lookup table 5 is configured so that:

By doing this, the bandwidth of the noise changes according to the hunt width of the human-powered video signal, so that the XIt sound can be efficiently removed. The still image is stored in the still image frame memory 9 from which noise has been removed through these processes.

〔Effect of the invention〕

As explained above, according to the present invention, for a pixel of interest with a video signal, in the case of a still image portion where the amount of change in the time axis direction is less than a certain specified value v1, noise is removed via a temporal low-pass filter. , its change Q V is a specified value v2
In the case of the above moving image part, noise is removed through a spatial low-pass filter, and in the case of an image part in which the amount of change ■ is between ■1 and ■2, between static and kyoku, a temporal low-pass filter and a spatial low-pass filter are used. Since noise is removed by mixing and extracting the outputs from the filters, a still image with noise effectively removed from the video signal can be obtained without any signal loss, even for any image. This has the effect that the still image can be used as a signal source for a printer or the like to obtain a high quality hard copy.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the circuit configuration of one embodiment of the present invention, FIG. 2 is an explanatory diagram showing the output value of the lookup table of FIG. 1, and FIG. 3 is a circuit diagram of another embodiment of the present invention. Figure 4 is a block diagram showing the basic configuration of the conventional device, Figure 5 is a graph showing the frequency characteristics of the conventional device, and Figure 6 shows the value of K and the degree of improvement in the SN ratio of the conventional device. FIG. 7 is an explanatory diagram showing the positions of human pixels for 2×2 Hadamard transformation. 2...Motion detection unit, 3...Temporal low-pass filter, 4.41.42...Spatial low-pass filter, 6...
Multiplier, 7.71.72... Multiplier, 8... Adder, 9... Frame memory. 1#, Gyorang's Jide-eup I column n-1 to t-block diagram No. 1
Diagram (Movement SE V) Lugga, Fidabu's name is called ``Ho τ shobi Mingpeng 2nd fig. ) K slide shows user f'
The graph shown by the graph is 1 (K) 4. The noise level of the user is 5 Nr.
Graph showing lTh improvement skin Figure 6

Claims (1)

[Scope of Claims] 1) a motion detection unit that detects the amount of change V in the time axis direction for a certain pixel of interest in a video signal; a temporal low-pass filter that removes noise in the video signal by correlation between frames; and a space. a spatial low-pass filter that removes noise from the video signal through filtering processing, and a provision that allows only the output of the temporal low-pass filter to pass if the output V of the motion detection section is below a lower limit specified value V_1, and that the output V is the upper limit. means for passing only the output of the spatial low-pass filter when the value is greater than or equal to V_2, and for passing the mixed output of the temporal low-pass filter and the output of the spatial low-pass filter when the output V is between V_1 and V_2; A noise removal device characterized by comprising: