JP3416355B2 - Apparatus and method for removing noise from color image - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus and method for removing noise from a color image for providing a high quality color image.

[0002]

2. Description of the Related Art Generally, in an image processing apparatus, image noise removal or sharpening processing is performed in order to provide a high quality image. This noise removal is a process of removing noise mixed in from an image input device or the like with a spatial filter.

Conventionally, as a noise removal filter,
A filter that averages the pixel data of the local pixel group centered on the pixel of interest and outputs it as pixel data of the pixel of interest, that is, a filter that removes high-frequency components by an integration operation has been used. Image noise can be removed by scanning this filter over the image and taking the local moving average.

However, when the above filter is used, high frequency components of the image are smoothed and lost, so that noise is removed and the sharpness of the contour portion in the image is also lost. Therefore, in general, a median filter is used as a method for removing noise without deteriorating information that visually gives sharpness such as a contour portion in an image. This median value filter outputs the median value of the pixel data of each pixel forming the local pixel group as the pixel data of the target pixel. Therefore, since the output pixel data is the pixel data of the pixel (including the target pixel) actually existing in the neighborhood area of the target pixel,
It is possible to remove noise while preserving the sharpness of the contour portion in the image.

[0005]

By the way, although the above method has been applied to a monochrome image,
Applying this method to color images creates new problems. That is, as is well known, a color image is composed of planes of a plurality of color elements (for example, red (R), green (G), and blue (B)). In such a color image, not only high-frequency components in each plane but also pixels in which the color is visually different from the human visual area are recognized as noise. However, in general, when denoising a color image,
Since the above median filter is scanned for each plane that makes up the image, noise can be removed and contours can be saved for each plane, but even when they are represented as a color image. It is unpredictable whether or not the noise removal and the preservation of the contour portion are properly performed. That is, although the high-frequency component noise in each plane is efficiently removed, if the pixels when each plane is represented as a color image have a visually large color difference compared to its neighboring area, The pixel may be recognized as noise. It is impossible to remove such noise by the above method.

When the filter is scanned for each plane as described above, in order to independently determine the pixel mapping value (pixel data after filtering) in each plane, the mapping values are combined to produce a color image. When expressing, there was a problem in color stability, such as unnatural coloration occurring locally.

Therefore, in view of the above problems, it is an object of the present invention to provide a noise removing apparatus and method capable of accurately removing noise even in a color image.

It is another object of the present invention to provide a noise removing apparatus and method which are free from unnatural coloring after noise removal and have high color stability.

[0009]

According to a first aspect of the present invention, there is provided a method of scanning a color image with a filter composed of a plurality of pixels to remove noise from the color image. An average position calculation step of obtaining an average position in the color space of the distribution by obtaining a distribution in which the pixel data of the color image of each pixel scanned by the filter at that time is mapped in the color space, b) The magnitude of the variance value of the distance between the position of each pixel data mapped in the color space and the average position.
A pixel selection step of selecting one of the respective pixel data as selected pixel data according to , (c) pixel data for the selected pixel, a filter output control step of giving as the output of the filter at that time, It is characterized by being provided.

Further, in order to achieve the above object, the invention of claim 2 sets the average position as a barycentric position of the distribution,
When the integer M is a predetermined integer satisfying the condition of 1 ≦ M ≦ N with respect to the number N of pixels forming the filter, the pixel selection step is performed for each pixel data mapped in the color space. Among the positions, the pixel data of the pixel closest to the M-th position from the center of gravity position is selected as the selected pixel data.

[0011]

Further, in order to achieve the above-mentioned object, the invention of claim 3 is an apparatus for scanning a color image by a filter composed of a plurality of pixels to remove noise of the color image. An average position calculating unit that obtains a distribution in which pixel data of the color image of each pixel scanned by a filter is mapped in a color space, and obtains an average position of the distribution in the color space,
(b) A pixel that selects one of the pixel data as selected pixel data according to the magnitude of the variance value of the distance between the position of each pixel data mapped in the color space and the average position It is characterized by comprising: a selection unit; and (c) a filter output control unit that gives the pixel data of the selected pixel as an output of the filter at that time.

In order to achieve the above-mentioned object, the invention of claim 4 sets the average position as the barycentric position of the distribution, an integer M with respect to the number N of pixels constituting the filter, 1 ≦ When a predetermined integer satisfying the condition of M ≦ N is set, the pixel selecting means sets the pixel data of the pixel closest to the M-th position from the barycentric position among the positions of each pixel data mapped in the color space. It is characterized in that it is used as means for selecting as selected pixel data.

[0014]

[0015]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described in detail below with reference to the drawings. In this specification, first, an outline of an image processing apparatus including a color image noise removing apparatus according to the present invention will be described, and then a color image noise removing method according to the present invention will be described.

A. Outline of Image Processing Apparatus: FIG. 1 is a block diagram showing an electrical configuration of an image processing apparatus 100 incorporating a color image noise removing apparatus according to the present invention. The image processing apparatus 100 includes a control unit 50 and an operation unit 1.
0, display unit 20, image input unit 30, and image output unit 40
Are connected and configured. The operation unit 10 is an input unit (for example, a keyboard or a mouse) to which a user gives a command or information, and the display unit 20 is a display unit (for example, a color display) capable of displaying a color image. The user can give commands and information to the control unit 50 from the operation unit 10, and can confirm the result of the image processing and the message from the control unit 50 on the display unit 20. Image input unit 3
Reference numeral 0 is an image input means (for example, an input scanner) having a function of inputting a color original image. In addition, the image output unit 40
Is an output unit (for example, an output scanner) that outputs a color image that has been subjected to noise removal processing and other desired processing (for example, color tone adjustment) according to the present invention.

As shown in FIG. 1, the control section 50 includes a storage section 60, an image processing section 70 and a noise removal processing section 80. The storage unit 60 is a memory, and is provided with an image data storage unit 60A that stores color image data and a setting data storage unit 60B that stores setting data necessary for image processing. In addition, the image processing unit 70 receives image data from the image input unit 30 based on a user's instruction from the operation unit 10 and stores the image data in the image data storage unit 60A, and also performs image processing on the image data. It has a function of outputting the result to the display unit 20 or the image output unit 40. Further, the noise removal processing unit 80 has a function of performing noise removal processing of image data based on the setting data stored in the setting data storage unit 60B when receiving a noise removal command from the image processing unit 70. . The noise removal processing method by the noise removal processing unit 80 will be described in detail later.

B. Noise Removal Method: FIG. 2 is a flowchart showing an example of the noise removal method according to the present invention.

First, a desired color image is read by the image input unit 30 prior to the noise removal processing (step S1). The image data read here is RGB
The data is represented by, and is stored in the image data storage unit 60A and displayed as a color image on the display of the display unit 20. FIG. 3A is a conceptual diagram showing how a color image is displayed on the display unit 20. The image E and the noise N1 are displayed on the screen 21 on the display of the display unit 20. In the example considered here, both the image E and the noise N1 have colors different from the background A.

By the way, what is perceived as noise by human eyes is a pixel whose color is significantly different from that of the neighboring area. Therefore, it is necessary to remove noise based on human vision. However, the color difference evaluated in the RGB space (distance between two points in the space) is
It does not match the color difference perceived by humans. That is, RG
Even if the distance between two points is large in B space, it may seem that human eyes are not so different in color, and conversely, even if the distance in RGB space is small, humans can see large colors. It may feel different. Therefore, a process of expanding the image data (RGB data) read in the image data storage unit 60a into a Lab uniform color space which is a color space corresponding to human vision is performed. In this way, noise removal corresponding to human vision can be performed.

Returning to FIG. 2, next, when the user inputs a noise removal command from the operation unit 10, step S2
Then, the noise removal processing unit 80 develops (maps) the pixel data of the local pixel group of the color image previously read in the Lab uniform color space. FIG. 3B is a conceptual diagram of the filter F for selecting the local pixel group of the color image. In the example considered here, the filter F is 3
It is composed of a pixel matrix of × 3, and the selection of the local pixel group is performed by scanning the color image displayed on the screen 21 by the filter F. In this scanning procedure, first, the filter F is scanned in the X direction (main scanning direction) in the drawing starting from the upper left corner of the screen 21, and when the right end of the screen 21 is reached, the Y direction ( Sub scanning direction)
The scanning is performed in the X direction from the position of the left end on the line adjacent to, and the same procedure is repeated until the filter F reaches the lower right corner of the screen 21. Then, the pixel data of the local pixel group scanned by the filter F at a certain point of time is developed in the Lab uniform color space. For example, when the filter F is at the position F1 in FIG. 3A, the pixel data of 9 pixels included in the position F1 is developed in the Lab uniform color space.

In the above expansion processing, it is necessary to convert the pixel data of the local pixel group from the RGB space to the Lab uniform color space, but in the example considered here, it is stored in the setting data storage unit 60B in advance. The image processing unit 70 performs conversion processing based on a lookup table (conversion table). The converted pixel data is stored in the image data storage unit 60A.

In FIG. 4, the pixel data of the local pixel group is Lab.
It is a conceptual diagram which shows a mode that it was developed in the uniform color space. Figure 4
4A is a diagram when the filter F is at the position F1 on the screen 21, and FIG. 4B is a diagram when the filter F is at the position F2. As shown in each of FIGS. 4A and 4B, nine pixels Pa <b> 1 to 9 selected by the filter F
Pa9 and Pb1 to Pb9 are expanded in the Lab uniform color space. As shown in FIG. 3A, the position F1 includes noise N1, and the position F2 includes the contour portion E1 of the image E. Therefore, FIG.
In FIG. 4A, only two noise pixels Pa8 and Pa9 are present at positions chromatically separated from the other pixels Pa1 to Pa7, whereas in FIG.
Pb5 to Pb9 included in and other pixels Pb1 to
There are almost the same number of two pixel groups of Pb4.

Next, returning to FIG. 2, in step S3,
The noise removal processing unit 80 obtains the average position of the coordinate value group of the pixel data expanded in the Lab uniform color space. In this example, the position of the center of gravity is considered as the average position, and the position of the center of gravity G
Is

[0025]

[Equation 1]

Is calculated by In the formula 1, L
k, ak, and bk are the L-axis coordinate value, a-axis coordinate value, and b of the pixel data of 9 pixels expanded in the Lab uniform color space, respectively.
Axial coordinate values, and Lx, ax, and bx are an L-axis coordinate value, an a-axis coordinate value, and a b-axis coordinate value of the gravity center position G, respectively.

As shown in FIG. 4 (a), when the local pixel group includes noises (pixels Pa8, Pa9) that are chromatically separated, the center of gravity is located near the pixel group (Pa1 to Pa7) other than the noise. Ga is present. On the other hand, FIG.
As shown in (b), when the local pixel group includes pixels in the contour portion of the image, the two pixel groups have substantially the same number, and therefore the center-of-gravity position Gb exists near the middle of the two pixel groups.

Returning to FIG. 2, next, in step S4, the distance from the position of the center of gravity to each pixel is calculated. Since this distance is calculated in the Lab uniform color space, it corresponds to human vision as described above. The distance XPk (k = 1, 9) from the position of the center of gravity to each pixel is calculated by the following equation.

[0029]

[Equation 2]

Next, in step S5, the variance value of the distance XPk obtained as described above is obtained. The variance value σ ² is calculated by the following equation.

[0031]

[Equation 3]

Since this variance value σ ² indicates the degree of variation in the pixel data of the local pixel group in the Lab uniform color space, compared to the local pixel group including noise (FIG. 4A), the contour portion It becomes large in the local pixel group (FIG. 4B).

Returning to FIG. 2, in step S6, the noise removal processing unit 80 determines a selected pixel from the local pixel group according to the above-mentioned variance value σ ² . At this time, FIG.
When the variance value σ ² is small (when the local pixel group includes noise) as described above, the pixel corresponding to the noise is located farther from the center of gravity Ga than the pixel groups other than the noise. Pixels near the center of gravity Ga are selected pixels.
Further, when the variance value σ ² is large as shown in FIG. 4B (when the pixels of the contour portion of the image are included in the local pixel group), in order to emphasize the sharpness of the contour portion, from the center of gravity position Gb, The far pixel is set as the selected pixel. In the example considered here, the variance value σ ² and the distance M from the barycentric position to the selected pixel are arranged in order from the smallest M (the selected pixel is the Mth pixel closest to the barycentric position, where M is 1 ≦ M It is an integer that satisfies ≦ N, and N is the correspondence relationship of the number of pixels of the local pixel group (9 in this example), and is determined by comparing the variance value σ ² with a predetermined threshold value α. That is, first, the variance value σ ² is compared with the predetermined threshold value α. Then, if σ ² <α, the selected pixel is the closest pixel (M = 1) from the position of the center of gravity, and σ ² ≧ α
If so, the selected pixel is the sixth closest pixel (M =
6).

The value of the number M in the order of the distance from the barycentric position to the selected pixel according to the threshold value α and the variance value σ ² is
It is stored in the setting data storage unit 60B in advance and can be changed as necessary so that the user can perform desired processing.

Further, as shown in FIG. 4B, when the local pixel group includes pixels in the contour portion of the image and two or more pixel groups of the same number are formed in the Lab uniform color space, the above M
Since which pixel group the second pixel belongs to is governed by an accidental element, the shape of the contour portion may be deformed on a pixel-by-pixel basis. Since the sharpness of the part is not lost, there is no problem for the entire image.

Next, in step S7, the image processing unit 70 outputs the pixel data of the selected pixel as pixel data of the central pixel of the filter F on the original image, and the original image is stored in the image data storage unit 60A. And stored on the display of the display unit 20. At this time, the pixel data output as the pixel data on the original image is the pixel data expressed in RGB.

Next, in step S8, the filter F is set to 1
Scan for pixels. The operation procedure of the filter F is as described above. Then, after the filter F finishes processing up to the pixel at the lower right corner, it is determined in step S9 that a series of processes has finished, and the noise removal process ends. If the filter F has not reached the lower right corner, it is determined that the processing has not been completed, and the process returns to step S2.
The noise removal processing is performed again. That is, the above series of processing is repeated until the filter F reaches the lower right corner.

As described above, when noise is included in the local pixel group (when the variance value σ ² is small), a pixel close to the barycentric position of the local pixel group is selected and the selected pixel Since the RGB pixel data is output as the pixel data of the central pixel of the filter, noise is efficiently removed. Further, since the selected pixel is determined in the Lab uniform color space, only the part where human vision recognizes as noise can be removed.

On the other hand, when the local pixel group includes an outline portion (when the variance value σ ² is large), a pixel far from the barycentric position of the local pixel group is selected and R of the selected pixel is selected.
Since the GB image data is output as the image data of the central pixel of the filter, the sharpness of the contour portion is surely preserved.

Further, since the pixel data to be output is pixel data of pixels actually existing in the local pixel group, a stable image of a color in which unnatural color development is not recognized can be obtained.

[0041]

[Modification] An example of the present invention has been described above, but the present invention is not limited to the above example. For example, the filter in the noise removal processing technique of this example is a 3 × 3 pixel matrix, but this may be a filter of another shape or size.

Further, in this example, the variance value of the distance from the barycentric position to each pixel in the Lab uniform color space is obtained, and the number M is selected according to the variance value, but this number M is set as a fixed value. Alternatively, the pixel located at the Mth closest distance from the center of gravity may be selected as the selected pixel. However, it is easier to satisfy both the high noise removal processing capability and the storability of the contour portion of the image in the above-described embodiment.

Further, in this example, the barycentric position is used as the average position of each pixel in the Lab uniform color space, but the median coordinate position may be used instead of the barycentric position. The median coordinate position is a position corresponding to these three coordinate values, which is obtained independently for each of the L coordinate value, a coordinate value, and b coordinate value of each pixel developed in the Lab uniform color space. Is shown. Assuming that the number of target pixels is N, the median of the coordinate values considered here is N / 2th when N is an even number and (N + 1) / 2 when N is an odd number, counting from the minimum value of the coordinates. The th coordinate value.

[0044]

As described above, according to the first and third aspects of the invention, the distribution obtained by mapping the image data of the color image of each pixel scanned by the filter at a certain time point in the color space is obtained. Obtain the average position in the color space of the distribution , select one of the pixel data as selected pixel data according to the magnitude of the variance value of the distance between the average position and each pixel, the selected pixel Since the data is given as the output of the filter at that time, it is possible to accurately remove noise even in a color image and prevent unnatural color development.

According to the second and fourth aspects of the invention, the average position is set as the barycentric position of the distribution, and the integer M is 1 ≦ with respect to the number N of pixels forming the filter.
When a predetermined integer that satisfies the condition of M ≦ N is set, the pixel data of the pixel closest to the Mth position from the barycentric position among the positions of each pixel data is selected as the selected pixel data. In the above, noise can be removed accurately, and unnatural coloring can be prevented.

[0046]

[Brief description of drawings]

FIG. 1 is a block diagram showing an electrical configuration of an image processing apparatus 100 to which a color image noise removal technique according to the present invention can be applied.

FIG. 2 is a flowchart showing an example of a noise removing method according to the embodiment of the present invention.

FIG. 3 is a conceptual diagram showing an example of a display state on the display unit and a filter.

FIG. 4 is a conceptual diagram showing how pixel data of a local pixel group is expanded in a Lab uniform color space.

[Explanation of symbols]

100 image processing device 10 Operation part 20 Display 30 Image input section 40 Image output section 60 storage 70 Image processing unit 80 Noise removal processing unit F filter

─────────────────────────────────────────────────── --Continued from the front page (58) Fields surveyed (Int.Cl. ⁷ , DB name) G06T 1/00-7/60 H04N 1/40

Claims (4)

(57) [Claims] 1. A method for scanning a color image with a filter composed of a plurality of pixels to remove noise from the color image, comprising: (a) the color for each pixel scanned by the filter at that time. Obtaining a distribution in which the pixel data of the image is mapped in a color space, an average position calculating step of obtaining an average position of the distribution in the color space, and (b) a position of each pixel data mapped in the color space and the A pixel selection step of selecting one of the pixel data as selected pixel data according to the magnitude of the variance value of the distance from the average position; and (c) the pixel data for the selected pixel, A filter output control step of giving the output of the filter at the time point, and a noise removal method for a color image, comprising: 2. The average position is a barycentric position of the distribution, and the integer M is a predetermined integer satisfying the condition of 1 ≦ M ≦ N with respect to the number N of pixels forming the filter. At this time, the pixel selection step is a step of selecting, as the selected pixel data, pixel data of a pixel closest to the M-th position from the barycentric position among the positions of each pixel data mapped in the color space. The method for removing noise from a color image according to claim 1, wherein 3. A filter comprising a plurality of pixels
The color image to remove noise from the color image
A device, (a) for each pixel being scanned by the filter at that time
Map the pixel data of the above color image to the color space.
The distribution in the color space
An average position calculating means for obtaining a uniform position, and (b) the position of each pixel data mapped in the color space
According to the magnitude of the variance of the distance between
Select one of the above pixel data as selected pixel data
And (c) pixel data about the selected pixel at that time.
Filter output control hand to give as the output of the filter
Denoising instrumentation of a color image, characterized in that it comprises a stage, a
Place 4. The apparatus according to claim 3, wherein the average position is a barycentric position of the distribution, and an integer M is set to a number N of pixels forming the filter.
And a predetermined integer that satisfies the condition 1 ≦ M ≦ N
The pixel selection means determines the position of each pixel data mapped in the color space.
Then, the pixel data of the pixel closest to the center of gravity is
It should be a means for selecting as the selected pixel data.
A color image noise removing device characterized by: