JPH07123299A - Picture signal noise elimination filter - Google Patents

Abstract

PURPOSE:To eliminate noise efficiently without loss of smooth motion by preserving a detailed part for a pre-processing in high efficiency coding of a picture signal such as a television signal. CONSTITUTION:A picture signal is divided into small blocks and a motion vector 14 is detected from each of the divided small blocks, and a picture element block 13 of a filtering object field is merged with a picture element block 12 of a preceding or succeeding field referenced by the detected motion vector 14 to generate a frame block 15. Then a small window whose variance in picture element values is minimized is selected by a selection section 18 among plural in-field small windows 16 and plural inter-field small windows 17 and a picture element value of an object picture element is replaced with a mean value of picture elements included in the selected small window.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to image signal noise for improving the coding efficiency by efficiently removing random noise contained in the image signal as preprocessing in high efficiency coding of a television signal or the like. It relates to a removal filter.

[0002]

2. Description of the Related Art When removing noise such as random noise included in an image signal of a television or the like, an average value of pixel values in a small window composed of a pixel to be filtered and pixels in the vicinity of the pixel to be filtered A low-pass filter that replaces the values is often used. However, the low-pass filter has a drawback that the resolution of a fine portion is lowered because it cuts off high-frequency components. In order to solve this, a method has been conventionally considered that can remove noise while preserving high-definition parts such as edges.

FIG. 6 is a diagram showing an example of a window used in a conventional filter for removing noise while maintaining an edge fine portion. When the pixel value of the digitized image was f _ij, N pieces of the small window W _k including f _ij, k =
Consider 1, 2, ..., N. For each of the small windows W _k , the average value m _k of the pixel values in the window and the variance σ _k ² are obtained. When the window that gives the minimum value of the obtained variances is W _min , the filter output g _ij is m _min . The reason why an edge is blurred is that a small window is formed across the edges, so such a method reduces the possibility that the edge is included in the small window used for performing filtering. As a result, it is possible to remove noise while maintaining the fine portion.

FIG. 7 is a typical block diagram for implementing the conventional filter described above. First, for the filtering target pixel f _ij with an image input from the input terminal 701, N window forming units 702
In, N small windows are constructed. Next, the variance detector 703 calculates variances σ ₁ ² , σ ₂ ² , ..., σ _N ² of the pixel values in each small window. Further, the average value calculation unit 704 calculates the average values m ₁ , m ₂ , ..., _{M N} of the pixel values in each small window. The N variances σ ₁ ² , σ ₂ ² , ..., σ _N ² are compared in a comparator 705, and a parameter 708 representing a window giving the minimum value is output. On the other hand, of the N average values m ₁ , m ₂ , ..., _{M N} , the parameter 708
The average value m _{min of the} small windows determined by
Selected at 06, and m as the filtering output g _ij
min is available at output terminal 707.

[0005]

The conventional method is applied to an interlaced signal such as a television signal.
Consider the case of applying to a field unit which is a still image as shown in FIG. In the case of an interlaced image, lines are temporally interleaved in the vertical direction. Therefore, for a still region, a pixel value at a position spatially separated by two lines in the vertical direction is used as a pixel value for calculating an average value. Will be used. Therefore, there is a problem that the probability that the fine portion and the edge are included in the small window increases, and the fine portion is blurred.

The present invention aims to solve this problem, and preserves the fine portion without impairing the smoothness of movement.
The purpose is to enable efficient removal of noise.

[0007]

FIG. 1 is a diagram for explaining the principle of the present invention. In the figure, 10 is a reference field, 11
Is a filtering target field, 12 is a pixel block of a reference field, 13 is a pixel block of a filtering target field, 14 is a motion vector, 15 is a merged frame block, 16 is a small window within a field, 17 is a small window between fields, Reference numeral 18 denotes a selection unit that selects a window and determines a pixel value.

In order to solve the above problems, the present invention uses the following means. First, the image signal is divided into small blocks, the motion vector 14 is detected from the past or future field (reference field 10) for each pixel signal of each divided small block 13, and the block 13 is detected for each small block. And the block 12 of the past or future field referred to by the detected motion vector 14 are merged to form a frame block 15.

In the frame block 15, the lines of the first field and the second field are regarded as adjacent lines, and the pixel to be filtered is spread over both the first field and the second field constituting the frame. 17 small windows between N ₁ fields
Make up. Further, N ₂ small windows 16 are formed only in the field to which the pixel to be filtered belongs. The selection unit 18 determines the average value m ₁ , m ₂ , of the values of the pixels in each of the configured N = N ₁ + N ₂ small windows.
, _{M N} and variances σ ₁ ² , σ ₂ ² , ..., σ _N ² are obtained, and the filtering target is the average value m _min of the small window W _min that gives the minimum value of the obtained N variances. Replace pixels.

[0010]

The smoothness of the motion is impaired if the fields are simply merged with respect to the still area or the moving area and the processing is performed in the small inter-field window. However, the present invention uses the motion vector to merge the frames. The pixels of the inter-field small window 17 configured in the block 15 can be regarded as the neighboring pixels of the pixel to be filtered, and the probability that an edge or the like is included in the small window to be filtered is reduced. Therefore, the noise removal filtering can be executed while maintaining the fine portion.

Further, for an area in which an appropriate motion vector cannot be detected because of a large amount of motion, a small window formed only in the field is also included in the selection target, so that the small window in the field is set in the moving area. The filtering is performed at 16, and the smoothness of movement is not impaired.

As a result, noise can be efficiently removed while preserving the fine detail and the smoothness of movement.
By using this filter as preprocessing for high-efficiency encoding, encoding efficiency can be improved.

[0013]

2 is a block diagram showing an embodiment of the present invention. In this example, consider a case where the reference field is the previous field and five small windows in the field and five small windows between the fields are formed.

The interlaced image signal is input from the image input terminal 201, stored in the field memory 202, and the motion vector detection unit 203 divides the filtering target field signal 206 into small blocks. A motion vector with half-pixel accuracy is detected, a motion compensation unit 204 generates a small block of a reference field obtained by the filtering target field signal and the detected motion vector, and a field merge unit 205 generates a small block of the filtering target field. And a small block of the reference field are merged to form a frame block 208.

For each pixel to be filtered of the constructed frame block 208, in the inter-field window construction sections 214 to 218, there are five small windows W ₆ and W spanning the first field and the second field.
₇ , W ₈ , W ₉ , W ₁₀ are configured.

In addition, from the filtering target field signal 206, the in-field window forming sections 209-2
In FIG. 13, five small windows W ₁ consisting of pixels of only the field to which the pixel to be filtered belongs,
W ₂ , W ₃ , W ₄ , and W ₅ are configured.

Each of the 10 windows W ₁ , W configured
_2, ..., with respect to W _10, variance sigma ₁ ² of the pixels in each window in the variance calculation unit 219, sigma ₂ ^2, ..., sigma ₁₀ ² are calculated. Comparator 2 for the 10 variances obtained
In FIG. 21, the one having the smallest variance is obtained, and the parameter (mi representing the window number giving the smallest variance is given.
n) 224 is output.

Meanwhile, the average value calculating unit 220 at the 10 window W _1, W _2, ..., the average value m _1, m ₂ of the pixel values in the window for each W _10, ..., m ₁₀ is After being calculated, it is input to the selection circuit 222. In the selection circuit 222, the average value m _{min of the} window specified by the parameter (min) 224 is selected and output as the filtering output g _ij to the output terminal 223.

FIG. 3 is a diagram showing a method of merging frame blocks in the embodiment of the present invention. In particular, Figure 3
(A) and (b) are cases where the vertical component of the motion vector detected by the motion vector detection unit 203 indicates the position of a half pixel, and (b) is the filtering target field is the first field An example in the case where the field is the second field of the previous frame, (b) shows an example in which the filtering target field is the second field and the reference field is the first field. In the case of a small block in which the motion vector indicates the position of half a pixel, the reference field can be merged by translating by the amount indicated by the motion vector.

On the other hand, FIGS. 3C and 3D show the case where the vertical component of the motion vector detected by the motion vector detection unit 203 indicates the position of an integer pixel,
(C) is an example when the filtering target field is the first field and the reference field is the second field of the previous frame, and (D) is an example when the filtering target field is the second field and the reference field is the first field Indicates. In the case of a small block in which the motion vector indicates the position of an integer pixel, the parallel movement corresponding to the motion vector overlaps the pixel of the filtering target field. Therefore, Fig. 3 (c) and (d)
As shown in, an offset is added to the vertical component of the motion vector and merged. Here, as the offset, the offset (vertical direction) difference between the filtering target field and the reference field is added.

4 and 5 are merged by the method described above.
Small window used for a closed frame block
In the example of c, FIG. 4 shows the first pixel to be filtered.
A small window, configured if it belongs to a field, figure
5 indicates that the pixel to be filtered belongs to the second field
The small window comprised in the case is shown. these
W out of small windows₁, W₂, W₃, W_Four, W_FiveIs
Only in the field to which the pixel to be filtered belongs
It is a small window made up of W₆, W₇, W ₈，
W₉, W_TenIs the field to which the pixel to be filtered belongs
A small widget that spans the command and reference fields.
It is a window.

With this structure, the stationary area and
Small window W between fields in moving area₆~ W_TenIs chosen
And configure a small window in a flatter area to filter
Since the ring can be executed, the storability of the fine portion is improved. Well
Also, the amount of motion is very large and an appropriate motion vector can be detected.
In areas that cannot be covered, a small window W between fields₆~ W _Ten
The variance of the pixel values in the
Small window W in the field₁~ W_FiveIs selected in the field
Things like filtering across
Absent. Therefore, the filter can be filtered without impairing the smoothness of movement.
Can be executed.

[0023] In the embodiment described above, five field in the small window W ₁ to _W-5, before between five fields that field and the reference field small window W ₆ ~
Although the case of configuring W ₁₀ is targeted, the number and shape of windows, the size and shape of blocks, the offset added to the motion vector, and the reference field can be set arbitrarily.

[0024]

As described above, according to the present invention,
When an interlaced image is input, pixels closer to the still area and the moving area can be used for filtering, and noise removal can be executed while maintaining the fine portion. Further, in a region in which an appropriate motion vector having a very large amount of motion cannot be detected, pixels in the same field are used for filtering, so smoothness of motion is not impaired. as a result,
If it is used as preprocessing for high-efficiency coding or the like, it becomes possible to improve coding efficiency.

[Brief description of drawings]

FIG. 1 is a diagram illustrating the principle of the present invention.

FIG. 2 is a block diagram showing an embodiment of the present invention.

FIG. 3 is a diagram showing a frame / block merging method according to an embodiment of the present invention.

FIG. 4 is a diagram showing a configuration of a small window according to the embodiment of the present invention.

FIG. 5 is a diagram showing a configuration of a small window according to the embodiment of the present invention.

FIG. 6 is a diagram showing an example of a window used for a conventional edge-preserving filter.

FIG. 7 is a diagram showing an example of a block configuration of a conventional edge-preserving filter.

FIG. 8 is a diagram showing how to apply filtering when a conventional edge-preserving filter is applied to an interlaced image.

[Explanation of symbols]

 10 reference field 11 filtering target field 12 reference field pixel block 13 filtering target field pixel block 14 motion vector 15 merged frame block 16 small field in field 17 small window between fields 18 selection unit 201 image input terminal 202 field memory 203 motion vector detecting unit 204 motion compensating unit 205 field merging unit 206 filtering target field signal 207 reference field signal 208 merged frame block 209 to 213 intra-field window forming unit 214 to 218 inter-field window forming unit 219 dispersion of each small window Calculation unit 220 Average value calculation unit for each small window 221 Comparator 222 Selection circuit 223 Output terminal Parameter indicating a window giving the 24 minimum variance

Claims (1)

[Claims] 1. An image signal denoising filter using an average value filter that replaces a target pixel with an average value of pixel values included in a small window including the target pixel with respect to an image signal having an interlaced structure. Means for dividing into blocks, means for detecting a motion vector from the pixel signal and past or future fields for each of the divided small blocks, and past or future fields deviated by the detected motion vector for each small block Of merging blocks of the above to make one frame block, a means of constructing a plurality of small windows for filtering target pixels in the frame block, and a field to which the filtering target pixel belongs A means for configuring multiple small windows only within the The means for obtaining the average value and the variance of the pixel values in each of the small windows, and the means for selecting the small window that minimizes the obtained variance are selected, and the pixels to be filtered are selected. An image signal denoising filter characterized by replacement with an average value of pixels included in a small window.