JP2000106630A - Noise eliminating method for digital image - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a superior noise eliminating method for digital images that has high noise-eliminating capability at a flat part of an image, completely stores an edge in a processing window and the line with the width of one pixel, and is suitable for high-speed processing which utilizes software programs. SOLUTION: In this noise-eliminating method, absolute difference among a processing object pixel and 8 surrounding pixels adjacent to the processing object pixel is calculated, and all surrounding pixels whose absolute differences is more than a predetermined threshold are replaced with the values of the processing object pixel to calculate a mean value related to the 8 surrounding pixels, and the processing object pixel is replaced with this mean value.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for removing noise from digital images used for videophones, videoconferencing, and the like.

[0002]

2. Description of the Related Art In videophones and videoconferencing, etc., in order to reduce the amount of information of an image and efficiently transmit the image, an image encoding method and a decoding method are used. Most use the discrete cosine transform. The discrete cosine transform can greatly reduce the amount of information in the image, but a unique Gaussian noise called mosquito noise appears in the decoded image,
There is a problem of impairing subjective image quality. Gaussian noise represented by mosquito noise can be removed by applying a spatial filter that removes high-frequency components.However, since high-frequency components included in feature points of an image such as edges are removed, image noise is reduced. There is a problem that the sharpness of the image is lost.

[0003] Therefore, an adaptive noise elimination method for eliminating noise without impairing the sharpness of an image has been announced.
See, for example, IEE Transactions on Systems Man and Cybernetics, SMC-8 (1978), pp. 705-710.
Page (IEEE Transactions on Systems, Man, and Cyberne
tics, Vol. SMC-8 (1978) PP 705-710), the k-nearest neighbor method.
Hereinafter, the method will be described with reference to FIG.

[0006] As shown in FIG. 6, in the k-nearest neighbor method, a processing window including a pixel to be processed p and its surrounding eight pixels is formed, and noise is removed from the pixel to be processed. The specific processing of the noise removal is as follows. First, among the eight peripheral pixels, k pixels are selected in the order in which the value is closer to the value of the processing target pixel p. Then, an average value of the selected k peripheral pixels is calculated, and the pixel p to be processed is replaced with the average value. For good noise removal while preserving image edges, k =
5 is good, and FIG. 6 also shows processing when k = 5.

The principle of the k-nearest neighbor method for removing noise while preserving image edges will be described with reference to FIG. In FIG. 6, a vertical edge exists in the processing window, and the pixels in the processing window are divided into two groups a and b. Here, the correlation between pixels belonging to the same group is clearly high, and the correlation between pixels belonging to different groups is low. Now, if the noise removal processing is performed by the k-nearest neighbor method assuming that the pixel to be processed belongs to the group b, the noise is removed because the pixel to be processed is replaced with the average value of the pixels belonging to the same group b. Attributes to group b are not lost. Therefore, it is possible to remove noise while preserving edges.

[0006]

However, in the k-nearest neighbor method, k <8 in order to preserve the edge, and all pixels around the pixel to be processed are not used for noise removal processing.
This is the same even when no edge exists in the processing window. For this reason, there is a problem that the ability to remove noise is low in a flat portion of an image. In addition, when a line having a width of one pixel exists in the processing window and a pixel to be processed exists on the line, the line is not completely saved. Furthermore, since k = 5 for good noise removal, it is necessary to calculate an average value for five pixels around the pixel to be processed, and there is a problem that high-speed processing by software is difficult.

The present invention solves the above-mentioned conventional problems, and has a high noise removal capability in a flat portion of an image.
It is an object of the present invention to provide an excellent digital image denoising method which completely preserves edges and lines of one pixel width in a processing window and is suitable for high-speed processing by software.

[0008]

A digital image noise elimination method according to the present invention calculates absolute difference values between a pixel to be processed and eight neighboring pixels adjacent thereto, and the absolute difference values are predetermined. All peripheral pixels larger than the threshold value are replaced with the value of the processing target pixel to calculate an average value of eight peripheral pixels, and the average value replaces the processing target pixel. As a result, an excellent digital image denoising method which has high noise elimination ability in a flat part of an image, completely preserves edges and lines of one pixel width in a processing window, and is suitable for high-speed processing by software. Can be provided.

[0009]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The first aspect of the present invention is a digital image noise elimination method for eliminating noise of a pixel to be processed by using N neighboring pixels adjacent to the pixel to be processed. Calculating the absolute difference value between the pixel to be processed and N peripheral pixels adjacent to the pixel to be processed, comparing the absolute difference value with a predetermined threshold, and dividing the N peripheral pixels into two groups. Of the peripheral pixels belonging to the group whose absolute difference value is larger than the threshold value, all the values are replaced with the values of the processing target pixels to calculate an average value for the N peripheral pixels, and the processing target pixels are replaced with the average value. In this way, as the number of pixels having low correlation with the processing target pixel among the N peripheral pixels increases, the processing target pixel has an effect of storing the original value.

According to a second aspect of the present invention, N is set to 8 in the first aspect, and in a flat portion of an image, all pixels around a pixel to be processed are subjected to noise reduction. In addition to the effect of being used, the use of the shift operation has the effect of reducing the amount of calculation processing of the average value for peripheral pixels.

According to a third aspect of the present invention, in the first aspect of the present invention, the threshold value is made variable according to the size of a quantization step used in a digital image encoding method or digital image decoding method. When the quantization step is small and the coding noise is small, only a small spatial change in the pixel value is removed as noise. Conversely, when the quantization step is large and the coding noise is large, the pixel value is small. This has the effect of removing even large spatial changes of as noise.

(Embodiment) An embodiment of the present invention will be described below with reference to the drawings. As shown in FIG. 1, in the digital image noise elimination method of the present invention, a pixel to be processed p and its surrounding eight pixels n (1), n (2), n (3),. A processing window is formed, and noise of the processing target pixel is removed.

FIG. 2 is a flowchart showing a specific process of noise removal. First, a variable for calculating the average value of surrounding pixels
Add is cleared with 0 (ST1). Next, the peripheral pixels
Regarding n (1), an absolute difference value diff from the pixel p to be processed is calculated (ST2). This absolute difference value diff is compared with a predetermined threshold value th (ST3). If the absolute difference value diff is larger than the threshold value th, the pixel value of the processing target pixel p is added to the variable add (ST3).
4) If not, the pixel value of the peripheral pixel n (1) is added to the variable add (ST5). After performing the above processing for the peripheral pixel n (1) for the remaining peripheral pixels n (2) to n (8) (ST6, ST7), the variable add is divided by 8, and the average value m per peripheral pixel is calculated. It is calculated (ST8). Here, since this division process is realized by shifting right by 3 bits, implementation by software is also easy. Finally, the calculated average value m is replaced with the processing target pixel p (ST9), and the noise removal processing for one pixel is completed.

FIG. 3 shows an arithmetic expression in the case where the absolute difference values for the pixel to be processed p and the eight pixels surrounding it are all equal to or smaller than the threshold th. In this case, the processing target pixel p
Is replaced by the average value of the eight surrounding pixels, and the effect of noise removal is enhanced.

FIG. 4 shows a pixel p to be processed and a peripheral pixel n
(2) shows an arithmetic expression in the case where the absolute difference values for n (3), n (5), n (7) and n (8) are equal to or smaller than the threshold th. In this case, the pixel value of the processing target pixel p is substantially preserved. This example illustrates that edges present in the processing window are preserved.

FIG. 5 shows a pixel p to be processed and peripheral pixels.
An arithmetic expression when only the absolute difference values regarding n (2) and n (7) are equal to or smaller than the threshold th is shown. Also in this case, the pixel value of the processing target pixel p is almost preserved. This example illustrates that one pixel wide lines present in the processing window are saved.

In the digital image noise elimination method of the present invention, the pixel p to be processed is replaced by the average value m obtained from the eight peripheral pixels. After calculating the average value, the pixel p to be processed may be replaced.

The threshold value th may be determined according to the size of a quantization step used in a digital image encoding method or digital image decoding method.

[0019]

As described above, according to the present invention, a high noise removal capability is obtained in a flat portion of an image, and an edge or a line of one pixel width in a processing window is completely preserved.
In addition, there is an advantageous effect that it is suitable for high-speed processing by software.

[Brief description of the drawings]

FIG. 1 is a schematic diagram showing a method for removing noise from a digital image according to an embodiment of the present invention.

FIG. 2 is a flowchart showing a process for removing noise from a digital image according to an embodiment of the present invention;

FIG. 3 is a schematic diagram showing a method for removing noise from a digital image according to an embodiment of the present invention in a case where absolute difference values regarding a processing target pixel and eight surrounding pixels are all equal to or smaller than a threshold value;

FIG. 4 is a schematic diagram illustrating a method for removing noise from a digital image according to an embodiment of the present invention when a processing window includes an edge;

FIG. 5 is a schematic diagram showing a method for removing noise from a digital image according to an embodiment of the present invention when a processing window includes a line having a width of one pixel;

FIG. 6 is a schematic diagram showing a conventional digital image noise elimination method.

Claims (3)

[Claims] 1. A method for removing noise from a pixel to be processed using N neighboring pixels adjacent to the pixel to be processed, comprising: a pixel to be processed; and N pixels adjacent to the pixel to be processed. Calculating the absolute difference value with respect to the peripheral pixels, and comparing the absolute difference value with a predetermined threshold to divide the N peripheral pixels into two groups, wherein the absolute difference value is larger than the threshold. A digital processing unit that calculates an average value of the N peripheral pixels by replacing all the values of only the peripheral pixels belonging to the group with the value of the processing target pixel, and replaces the processing target pixel with the average value. Image noise removal method. 2. The method according to claim 1, wherein N is 8.
2. A method for removing noise from a digital image as described above. 3. The method for removing noise from a digital image according to claim 1, wherein the threshold value is made variable in accordance with the size of a quantization step used in a method for encoding or decoding the digital image.