KR20070035800A - Removing apparatus and method for video noise using the slope of adjacent pixel - Google Patents

Abstract

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus and method for removing image noise using an inclination of adjacent pixels in order to interpolate noise using inclination of brightness information between neighboring pixels. .

In accordance with another aspect of the present invention, an apparatus for removing noise of an image using a slope of an adjacent pixel includes: region-specific slope detection means for detecting a slope of a first region of pixels adjacent to an input pixel; Comparing means for comparing the detected first slope for each area with a threshold value; Comparison result of the comparison means, when the gradient value for each first region is larger than a previously stored threshold value, the gradient detection means for each adjacent pixel region using the absolute difference value of the adjacent pixels to detect the gradient for each second region; And pixel extracting means for extracting and applying the input pixel value by using the adjacent pixel value different from the detected inclination value of the second region.

Image, Pixel, Adjacent Pixel, Noise

Description

Reducing apparatus and method for video noise using the slope of adjacent pixel}

1 is a configuration diagram showing a first example of a conventional video noise reduction device.

2 is a configuration diagram showing a second example of a conventional video noise reduction device.

3 is a configuration diagram showing a third example of the conventional video capture removal device;

4 is a diagram illustrating a conventional method of compensating for image noise.

5 is a diagram illustrating an example of conventional image noise compensation.

Figure 6 is a block diagram showing an image noise removing device using the slope of the adjacent pixel in accordance with the present invention.

7 is a view showing an initial image and a noise image according to the present invention.

8 is a view showing a state for calculating the slope for each area in the present invention.

9 to 11 are diagrams showing examples of calculating tilt using adjacent pixels in the present invention.

12 illustrates an interpolation example of horizontal noise pixels according to the present invention.

13 is a flowchart illustrating a method of removing noise using an adjacent pixel slope according to the present invention.

<Explanation of symbols for the main parts of the drawings>

101 ... area gradient detector 102 ... threshold storage unit

103 ... Tilt comparison unit 104 ... Adjacent pixel tilt detection unit

105.Pixel Extractor

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus and method for removing image noise using an inclination of adjacent pixels in order to interpolate noise using inclination of brightness information between neighboring pixels. .

Due to the rapid development of information technology, new video systems are rapidly developed and commercialized to stimulate consumer's desire. In particular, new products such as liquid crystal displays (LCDs) and plasma display panels (PDPs) are in the process of releasing new products after the development stage of the 90s. These display devices are gradually increasing in size, and the size of an image is also increasing in size.

For example, the size of LCD and PDP products is being developed and marketed at high speeds such as 40 inches and 50 inches beyond 30 inches. In addition, with the rapid development of hardware (H / W), image data is also becoming larger in size such as VGA: 640 x 480, XGA: 1024 x 768, SXGA: 1280 x 1024, UXGA: 1600 x 1200.

However, as the enlargement proceeds, some problems arise. The most important among them is the lack of video contents to fully utilize the hardware functions. In other words, while hardware is rapidly being enlarged, the image to support it cannot follow the hardware. It is hard to find high definition (HD) or higher (XGA or higher) video, which is the most common video content. Therefore, most of the video used in the large-scale video system currently displays VGA-level (640 x 480) images inside the set-top box and displays them on a large LCD or PDP screen.

Therefore, to use a small image on the screen of a large display device as it is today, a function called a scaler is a function. However, most small images used include noise on the screen. If noise is included in the image data, not only the visual quality of the human being is degraded, but also the functional degradation of the image processing system (still image compression, video compression, image information transmission, etc.) is generated. Also, compared to the image without noise, the image with noisy noise has a significantly lower compression ratio depending on the type and intensity of the noise under the same compression conditions, and since the noise has a high frequency component, DCT (Discrete Cosine Transform) transformation This reduces the efficiency of image information processing.

In addition, recently, in order to display a video signal such as NTSC / PAL / SECAM on a display device capable of displaying a larger resolution, noise reduction included in an existing video signal (NTSC / PAL / SECEM) signal has been removed. It is important to have a pretreatment function to perform. If the received image signal is directly transmitted to the display device without such a preprocessing function, many difficulties will occur visually.

1 is a block diagram for removing noise of a conventional video signal.

As shown in FIG. 1, an input signal including noise is filtered by using an intermediate filter (3 * 3 median filter) 10 for noise removal to output an output signal from which noise is removed. The intermediate filter 10 performs noise reduction through intermediate filtering of 3 * 3 pixels around the center pixel.

In general, in performing noise reduction, pure signals have correlations between adjacent signals, but noises do not have correlations with each other, so signals and noises can be distinguished from each other.

By distinguishing such a signal and noise, the 3x3 intermediate filter 10 may input an input current pixel (5) value and eight pixel values (1, 2, 3, 4, 6, adjacent to the position). A total of nine pixel values including 7,8,9) are arranged in order of magnitude, and the intermediate values are output as newly substituted pixel values. The problem with this method is that the current pixel value is changed to a middle value between its own value and the remaining eight pixel values surrounding it. That is, blurring of the image occurs.

FIG. 2 is an improved invention of FIG. 1, in which an input stage of an intermediate filter 12 does not go through a noise pre-judge 11 and a pixel or intermediate filter 12 that pass through the intermediate filter 12. It is a structure including the selection part 13 which selects a signal.

In operation thereof, the noise determiner 11 determines whether the pixel currently input includes noise, and outputs a selection control signal according to the determined result. Pixels passing through the noise determiner 11 are output through the intermediate filter 12 or are directly input to the selector 13.

The selector 13 selects a med. Pixel from which the noise is removed through the intermediate filter if noise is generated by the selection control signal of the noise determiner 11, and if there is no noise, the selector 13 11) an original pixel (org. Pixel) input directly through the selection is output.

Accordingly, when no noise is present, the original pixel may be used as it is, thereby preventing the image from being crushed from occurring.

However, the conventional method using the selective intermediate filter 12 has the following problems. First, the basic assumption of the selective intermediate filter 12 is that the pixel surrounding the input pixel (Pixel) should not contain noise. If noise is included in the pixel surrounding the input pixel Pixel, there is a problem in that the noise is not removed using the selective intermediate filter 12 as described above.

3 is another example of a conventional image noise removing apparatus, and a threshold value and absolute value comparison method used to improve the function of FIG. 2.

The pixel value difference calculator 21 determines an absolute difference between the input pixel value and adjacent pixels thereof. The comparison unit 23 compares the absolute difference value with the threshold value stored in the threshold storage unit 22.

The noise pixel removing unit 24 deletes the pixel including the noise during the noise removing process when the absolute difference is greater than the threshold. Subsequently, the pixel from which the noise is removed is output through the filtering of the filter unit 25.

In the same manner as in FIG. 3, the value of the pixel Pixel newly extracted by the filter unit 25 to be used is limited to the average value of the surrounding pixels. In this case, noise reduction is possible, but it is difficult to express smooth colors. In other words, the maximum effect is achieved when all the colors around the center pixel have the same value. However, when the median filter is used as in the edge, a significant flattening phenomenon occurs at the edge. This may cause deterioration in image quality.

4 is a diagram illustrating (a) an initial image, (b) a noise image, and (c) a compensation image according to the present invention. (A) of FIG. 4 compensates the pixel by using the average value of adjacent pixels in the image including the noise as shown in (b), so as to generate the initial image, it becomes a compensated image as shown in (c). Here, as shown in FIG. 5, the compensation pixel takes an intermediate value between a pixel value of an initial image and a pixel value of an adjacent image. However, by outputting the compensation image as shown in FIG. 4C, it is difficult to return the original image at the edge portion.

A first object of the present invention is to improve noise cancellation at the edge part by using the area-specific gradient of adjacent pixels.

A second object of the present invention is to provide an apparatus and a method for removing noise using adjacent pixels, which determine whether a pixel in which a noise is generated is an edge portion, and calculate a slope for each region and a pixel for each region of a selected pixel.

In order to achieve the above object, an image noise removing apparatus using an inclination of an adjacent pixel according to the present invention is provided.

Area-specific inclination detection means for detecting inclination for each first area of pixels adjacent to the input pixel;

Comparing means for comparing the detected first slope for each area with a threshold value;

Comparison result of the comparison means, when the gradient value for each first region is larger than a previously stored threshold value, the gradient detection means for each adjacent pixel region using the absolute difference value of the adjacent pixels to detect the gradient for each second region;

And pixel extracting means for extracting and applying the input pixel value by using the adjacent pixel value different from the detected inclination value of the second region.

And a means for interpolating pixel values by applying an average value of neighboring pixels when the value of each region is smaller than a threshold as a result of the comparison by the comparing means.

Preferably, the inclination detection means for each adjacent pixel region groups peripheral pixels of the input pixel into at least two pixel units to obtain absolute absolute values, and obtains an absolute difference value based on the largest absolute value and the adjacent absolute value of the difference absolute values. It is characterized by detecting the inclination by two areas.

Preferably, the absolute difference value of adjacent pixels grouped by the at least two pixel units is obtained in the horizontal, vertical, and diagonal directions of adjacent pixels of the input pixel, respectively.

Preferably, the second region-specific inclination value is a slope corresponding to a moving direction of a pixel value obtained by using absolute difference values in a horizontal direction, a vertical direction, and a diagonal direction.

Preferably, the pixel extracting means extracts an input pixel value by multiplying the detected inclination value for each of the second regions and a variation amount of the pixel by at least one pixel value among the corresponding neighboring pixels.

In addition, according to another embodiment of the present invention, a method of removing noise of an image using a slope of an adjacent pixel is provided.

Detecting a slope for each first region of the input pixel; Determining whether an image is edged with respect to the input pixel by comparing the detected inclination value of each first region with a threshold value; If it is determined that the edge of the image is determined, obtaining absolute absolute values of adjacent pixels of the input pixel, and extracting a largest value among absolute absolute values; Calculating a slope for each second region between the extracted absolute maximum difference value and an absolute difference value of an adjacent region; And extracting the input pixel value by using the calculated inclination value for each second region and the value of at least one pixel among the adjacent pixels.

Preferably, the absolute difference value of the adjacent pixels is detected in a horizontal direction, a vertical direction, and a diagonal direction with respect to neighboring pixels of the input pixel, respectively.

Preferably, the absolute difference value of the adjacent pixel is an absolute difference value between at least two pixel values.

Preferably, the input pixel is a center pixel, and includes nine pixels including a central pixel and a peripheral pixel.

Preferably, the input pixel value is obtained by multiplying an average value of a plurality of pixel values among adjacent pixels with a slope value for each second region.

Hereinafter, with reference to the accompanying drawings as follows.

6 is a block diagram illustrating an image noise removing apparatus using adjacent pixels according to the present invention.

Referring to FIG. 6, an inclination detection unit 101 for detecting an inclination for an area of an input pixel, a threshold storage unit 102 for storing a threshold for comparing the inclination for each area, and a per area; If there is a gradient comparison unit 103 that compares the area-specific inclination value detected by the inclination detection unit 101 with a threshold value, and the area inclination value exceeding the threshold value by the inclination comparison unit 103, adjacent pixels are present. And a pixel extracting unit 105 which extracts and outputs a pixel value including noise by using the extracted inclination of each region of the adjacent pixels. do.

Referring to the accompanying drawings, an image noise removing apparatus using adjacent pixels according to an exemplary embodiment of the present invention configured as described above is as follows.

Referring to FIG. 6, the inclination detection unit 101 for each region determines an inclination for each domain of an input pixel. For example, the area-specific slope of the pixel is performed to determine whether the edge portion of the image is interpolated when the noisy pixel is interpolated. That is, the present invention is to improve the weak noise reduction in the edge portion of the image.

The slope comparison unit 103 compares the detected slope value of each region with a threshold stored in the threshold storage unit 102. Here, the threshold 102 has a different value for each image sensor, and may be a brightness value obtained by each image sensor under a predetermined illuminance. The threshold storage unit 102 stores the threshold value obtained as described above from the host.

The slope comparison unit 103 compares the detected slope for each region with a threshold so that the present invention can be applied when the slope is greater than the threshold. That is, if the value is larger than the threshold value, the image is determined as an edge part. However, if the gradient value of each region in the gradient comparison unit 103 is less than or equal to the threshold, the noise pixel may be interpolated using an intermediate filter (that is, a 3 * 3 median filter, etc.). Alternatively, when the gradient value of each region is smaller than the threshold value, the pixel value may be interpolated by applying the average value of the neighboring pixels using the threshold comparison and extraction means.

In this case, the adjacent pixel slope detector 104 bundles at least two pixel areas adjacent to the noise pixel, which is an image edge portion, and detects the slope between the relative adjacent pixels for each direction by using the absolute difference value.

When detecting an inclination of the adjacent pixel, the adjacent pixel includes at least two pixel areas, and the image of the search area (for example, 3 * 3) is positioned horizontally or vertically or diagonally around the noise pixel. Can be detected respectively. In this case, when detecting an adjacent pixel slope of the search area, neighboring pixels are used except for the noisy pixel.

When the inclination value of each region exceeds a threshold, the neighboring pixel inclination detector 104 outputs the largest inclination value among the inclinations of adjacent pixels centered on a random noise pixel to the pixel extraction unit 105. As another example, the inclination of the adjacent pixel may detect the inclination of the area and the adjacent pixel area in the inclination detection unit 101.

The pixel extractor 105 uses the inclinations detected by the adjacent pixel inclination detector 104 as information for determining in which direction the brightness information of the searched image proceeds. That is, the pixel extracting unit 104 detects an advancing direction of the brightness level by using the gradient difference by applying one or more pixels or the average value of the pixels among the adjacent pixel inclination values and the adjacent pixels to produce a smooth image of the noise pixel. Interpolate. Accordingly, the pixel extractor 105 erases the pixel including the noise in the noise removing process by using the adjacent pixel slope.

Specifically, the pixel input to the inclination detection unit 101 for each region is composed of nine pixels including itself as shown in FIG. 7. FIG. 7A illustrates an initial image without noise, and FIG. 7B illustrates an image including a noise pixel P13 at the center. Here, in FIG. 7A, brightness values of 1, 4, and 7 which are the pixel areas P11 on the left side are similar, and 2, 3, 4, 5, 8, and 9 which are the pixel areas P12 on the right side. The brightness values will be similar. However, as shown in FIG. 7B, the center pixel P13 including noise has a value whose pixel value has nothing to do with other adjacent pixels. That is, the signal corresponding to every pixel has the correlation between signals, but since the noise has the characteristic of not showing the correlation between the noise, the noise does not have the correlation and can be distinguished and detected.

As shown in FIGS. 8 to 11, when the fifth pixel is a noise pixel currently input, the area-specific tilt detection unit 101 has an average value D1, D2, and D3 of each domain as shown in FIG. 8. ) Is calculated. Here, the pixel 5 is excluded from the D2 average value calculation.

The difference between the absolute values of the calculated average values is calculated, and the absolute values (| D1-D2 |) of the average values D1 and D2 and the absolute values (| D2-D3 |) of D2 and D3 are obtained, respectively. Calculate the difference between the absolute values of the two values. At this time, the slope comparison unit 103 assumes that the pixel of this portion is an edge portion when the calculated difference between absolute values is greater than or equal to a certain threshold. In general, the edge of the image informs information about the position, shape, size, surface, and pattern of an object. The edge is the part where the brightness of the image changes from a low value to a high value or from a high value to a low value. The edge is where there is a significant difference in brightness in the image, which corresponds to the contour of the object, so it exists at either the discontinuity point of the pixel value or the discontinuity point of the pixel differential value.

When the input pixel 5 is determined to be the edge of the image, the adjacent pixel tilt detection unit 104 determines an absolute difference value in the horizontal direction, the diagonal direction, and the vertical direction to determine the inclination of each pixel of the adjacent pixel of the pixel 5. Each will be detected.

9 is an example of detecting the tilt between adjacent pixels in the horizontal direction. The absolute difference value D4 of the pixel 1,2 areas and the absolute difference value D5 of the pixels 2 and 3 are compared with each other to detect the slope D5 / D4 of two adjacent pixels in the horizontal direction (→, ←). do. In addition, in order to detect adjacent pixels in the horizontal direction, the absolute differences between the pixels 7,8 and 8,9 located below are compared and obtained. One pixel is included in another adjacent pixel to obtain an absolute difference value of the adjacent pixels.

10 is an example of tilt detection between adjacent pixels in a diagonal direction. Using absolute difference value D5 of pixels 4 and 2, absolute difference value D6 of pixels 2 and 3, absolute difference value D7 of pixels 3 and 6, and absolute difference value D8 of pixels 8 and 6 The formation of the slope can be detected with respect to the diagonal direction ↗.

11 is an example of tilt detection between adjacent pixels in different diagonal directions. By comparing the absolute difference value D9 of pixels 2 and 6, the absolute difference value D10 of pixels 6 and 9, the absolute difference value D11 of pixels 8 and 9, and the absolute difference value D12 of pixels 4 and 8, The diagonal direction (↖, ↘) is formed.

9 to 11, it can be determined from which part the edge is formed. The largest direction among the absolute differences D4 to D12 is first examined.

Here, when D4 is the largest of the absolute difference values, it can be seen that the amount of change of the current pixel is progressing in the horizontal direction. In this case, the pixel extracting unit 105 calculates the value of the pixel 5 from the gradient D5 / D4 obtained by the pixel areas D5 and D4 in the horizontal direction as [pixel number * gradient (D5 / D4)]. Here, after determining the slope D5 / D4 using at least two adjacent pixels, the noise pixel value is obtained by multiplying one pixel (eg, pixel 6) at a relative position among adjacent pixels of the noise pixel. Here, the one (or the plurality of) pixels in the relative position may apply any one of the fourth pixel, the average value of the fourth and sixth pixels, or the average value of the adjacent pixels of the noise pixel.

In addition, after calculating and storing the adjacent pixel slope in advance as shown in FIG. 12, the value corresponding to the adjacent pixel slope is determined, and thus the value of the noise pixel may be extracted. The said 0/6-6/6 is a value calculated | required by D5 / D4.

In the above case, when the variation amount of the absolute difference value D5 is the greatest, the pixel movement is performed diagonally. Therefore, the slope of D6 / D5 is extracted and applied to the calculation of pixel value 5. Here, the detection of the generated pixel value 5 can be obtained as follows. That is, the fifth pixel value = 4 * (| 2-3 |) / (| 4-2 |). Here, the pixel 2 or the 8 located in the edge formation direction may be applied to the pixel 5 without applying the pixel 4 to be considered. In addition, the present invention may be applied as a combination ((4 * 2) / 3) with other adjacent pixel (s) without considering only one pixel.

If the absolute difference value is the largest D8, the pixel value 5 can be obtained as D7 / D8, and the pixel value = 8 * (| 3-6 |) / (| 8-6 |). Become. Since D8 proceeds in a similar direction to D5, it can be obtained in the same manner for D5 and D8.

In this manner, the fifth pixel value may be detected using the slope and the adjacent pixel values in the advancing direction of FIG. 11.

The present invention obtains the absolute difference value of the adjacent pixels and detects the slope of one adjacent pixel region based on the largest difference absolute value among the absolute differences. As another example, the adjacent pixel is determined using the absolute difference value of the adjacent pixels. After the respective area inclinations are obtained, one of the largest area inclinations obtained may be obtained.

13 is a flowchart illustrating a method of removing noise using a slope of an adjacent pixel according to the present invention.

Referring to FIG. 13, slopes of respective pixel areas are obtained from the input pixels (S101). After checking whether the obtained area gradient is greater than a threshold value (S103), if the threshold value is less than or equal to the threshold value, the noise pixel value is extracted using an intermediate filter as in the conventional method (S104). Alternatively, when the inclination value of each region is less than or equal to a threshold value, a pixel value may be interpolated by applying an average value of surrounding pixels.

When the inclination value of each region exceeds a threshold, it is determined as an edge part, and the inclination of adjacent pixel regions is compared by comparing the inclination of the pixel region obtained above (S105). In this case, relative inclinations between adjacent pixel areas are compared, and the largest inclination value is calculated among the relative inclinations (S107). Here, each inclination is a value obtained by dividing the absolute difference between at least two adjacent pixels, and the inclination of each adjacent pixel area is obtained, and the inclination of the current pixel with the largest variation is detected.

When the slope having the largest gradient value is obtained as described above, the noise pixel value is extracted in consideration of this (S109). Here, the noise pixel value may be obtained by multiplying the largest gradient value by an average value of one pixel or more pixels among adjacent pixels of the input pixel.

In the present invention, as described above, the absolute difference value between adjacent pixel regions is obtained without detecting all the slopes of adjacent pixels, and the adjacent pixel is obtained by using the absolute difference value and the absolute difference value having the largest value among them. Region-specific tilt may also be detected.

So far, the present invention has been described with reference to the preferred embodiments, and those skilled in the art to which the present invention pertains to the detailed description of the present invention and other forms of embodiments within the essential technical scope of the present invention. Could be implemented. Here, the essential technical scope of the present invention is shown in the claims, and all differences within the equivalent range will be construed as being included in the present invention.

According to an apparatus and method for removing noise using adjacent pixels according to the present invention, by interpolating a noise pixel by using an inclination of an adjacent pixel to determine whether a pixel area including noise is an edge portion, a noise image is reduced, thereby reducing the TV image ( NTSC / PAL / SECAM) can effectively remove noise that is indispensably included in the video transmitted from NTSC / PAL / SECAM. In addition, it can be applied to various applications with filters of multimedia video services such as video communication, video communication in mobile networks, and video application services on the Internet.

In addition, the present invention extracts an improved noise pixel value by using an inclination of an adjacent pixel area by determining and detecting whether an edge is formed in a horizontal, vertical, or diagonal direction by using an inclination of an adjacent pixel. It can work.

Claims (11)

Area-specific inclination detection means for detecting inclination for each first area of pixels adjacent to the input pixel; Comparing means for comparing the detected first slope for each area with a threshold value; Comparison result of the comparison means, when the gradient value for each first region is larger than a previously stored threshold value, the gradient detection means for each adjacent pixel region using the absolute difference value of the adjacent pixels to detect the gradient for each second region; And pixel extracting means for extracting and applying the input pixel value by using the adjacent pixel value different from the detected second region-specific gradient value. The method of claim 1, And a means for interpolating pixel values by applying an average value of neighboring pixels when the value of each region is smaller than a threshold as a result of the comparison by the comparing means. The method of claim 1, The inclination detection means for each adjacent pixel region groups peripheral pixels of the input pixel into at least two pixel units to obtain absolute absolute values, and calculates the second absolute value by the largest absolute value and the adjacent absolute value among the absolute absolute values. The image noise canceling device using the slope of the adjacent pixel, characterized in that for detecting the slope. The method of claim 3, wherein The absolute value of the difference between adjacent pixels grouped in units of at least two pixels is determined in the horizontal, vertical, and diagonal directions of adjacent pixels of the input pixel, respectively. The method of claim 3, wherein The inclination value of each of the second regions is an inclination of an adjacent pixel, wherein the inclination corresponds to a moving direction of a pixel value obtained by using absolute difference values in a horizontal direction, a vertical direction, and a diagonal direction. The method of claim 1, The pixel extracting means extracts an input pixel value by multiplying the detected inclination value for each second region by at least one pixel value among the corresponding neighboring pixels by the variation amount of the pixel. Noise Canceling Device. Detecting an inclination for each first region with respect to adjacent pixels of the input pixel; Determining whether an image is edged with respect to the input pixel by comparing the detected inclination value of each first region with a threshold value; If it is determined that the edge of the image is determined, obtaining absolute absolute values of adjacent pixels of the input pixel, and extracting a largest value among absolute absolute values; Calculating a slope for each second region between the extracted absolute maximum difference value and an absolute difference value of an adjacent region; And extracting the input pixel value by using the calculated second region-specific gradient value and the value of at least one pixel from among the adjacent pixels.  The method of claim 7, wherein The absolute value difference between the adjacent pixels is detected in the horizontal, vertical and diagonal directions with respect to the peripheral pixels of the input pixel, respectively. The method according to claim 7 or 8, And the absolute difference value of the adjacent pixel is an absolute difference value between at least two pixel values. The method of claim 7, wherein The input pixel is a center pixel, and the image noise removing method using the slope of the adjacent pixel, characterized in that the pixel consists of nine pixels including the central pixel and the peripheral pixels. The method of claim 7, wherein And the input pixel value is obtained by multiplying an inclination value of the second region by an average value of a plurality of pixel values among adjacent pixels.

Images