CN101352047B - Method and device for removing grid noise - Google Patents

Abstract

A device and a method for removing grid noises are disclosed. The device for removing grid noises in accordance with an embodiment of the present invention calculates an average value of each line of an inputted Bayer pattern image, calculates an estimate value estimating an average of even number lines placed between the odd number lines of a Bayer pattern image and odd number lines placed between the even number lines of a Bayer pattern image, calculates a difference value between the estimate value and the average value, calculates a correction coefficient by using the difference value, and applies the correction coefficient to the Bayer pattern image and outputs a corrected Bayer pattern image.

Description

Be used to remove the apparatus and method of grid noise

Technical field

The present invention relates to a kind of apparatus and method that are used to remove grid noise, relate in particular to a kind of apparatus and method that are used for removing the grid noise of image processing system.

Background technology

In the Bayer format-pattern in being input to image processing system, because the capable value difference of Gr (red green) and Gb (blue-green) grid noise can occur usually during color interpolation is handled.Be explained below with reference to accompanying drawing.

Fig. 1 has represented the example of the mean value of typical B ayer format-pattern green pixel in the horizontal direction.

In the drawings as can be seen, there is essential difference between the mean value of the green pixel between even number line and the odd-numbered line in the horizontal direction.As shown in Figure 2, this difference causes grid noise by color interpolation, and Fig. 2 has represented the example of the grid noise that color interpolation produces in the conventional image processing system.

Usually use Gaussian filter or median filter to remove these noises.Although these filters can be removed grid noise, they have also damaged image detail (for example, high frequency edge or border).In other words, in fact can not when keeping image detail, remove grid noise by the method for utilizing usual filter to remove grid noise.

Summary of the invention

In order to address the above problem, the invention provides a kind of apparatus and method that are used to remove grid noise, their center pixels in mask (mask) are under the situation of green, remove grid noise and can not damage image by the difference of utilizing the mean value between even number line and the odd-numbered line.

Other purposes of the present invention will be by execution mode described below and clearer.

To achieve these goals, one aspect of the present invention is characterised in that a kind of device that is used to remove grid noise.

The device that is used to remove grid noise according to an embodiment of the invention can have: average calculation unit, and it is used to calculate mean value of each row of the Bayer format-pattern of input; Estimation unit, it is used to calculate the estimated value of the mean value that is arranged on the odd-numbered line between the even number line that is arranged on even number line between the odd-numbered line and Bayer format-pattern of estimating the Bayer format-pattern; Difference computational unit, it is used to calculate by the estimated value of this estimation unit calculating with by the difference between the mean value of this average calculation unit calculating; And correction coefficient calculation, it is used for by utilizing the difference that is generated by this difference computational unit to come the calculation correction coefficient.

Herein, this device preferably also has correcting unit, and this correcting unit is used for this correction coefficient is applied to the Bayer format-pattern that this Bayer format-pattern and output calibration are crossed.

Preferably, the correction coefficient that is obtained by this correction coefficient calculation is the value that the mean value of the difference that will be calculated by this difference computational unit obtains divided by 2.Can determine correction coefficient at each green component of Bayer format-pattern.

Correcting unit can deduct the correction coefficient of each green component from each green component of the odd-numbered line of Bayer format-pattern, and the correction coefficient of each green component can be added on each green component of even number line of Bayer format-pattern.

In order to realize above purpose, another aspect of the present invention is characterised in that a kind of method of removing grid noise.

In the method that is used for removing grid noise, can calculate each mean value of going of the Bayer format-pattern of input according to an embodiment of the invention; Can calculate the estimated value of the mean value that is arranged on the odd-numbered line between the even number line that is arranged on even number line between the odd-numbered line and Bayer format-pattern of estimating the Bayer format-pattern; Can calculate the difference between this estimated value and this mean value; Can come the calculation correction coefficient by utilizing this difference; And this correction coefficient can be applied to this Bayer format-pattern, and the Bayer format-pattern of can output calibration crossing.

Description of drawings

Fig. 1 has represented the example of the mean value of typical B ayer format-pattern green pixel in the horizontal direction;

Fig. 2 has represented in the image processing system of routine the example of the grid noise that produced by color difference;

Fig. 3 has represented to be input to the example of mask of the Bayer format-pattern of removal device of the present invention;

Fig. 4 has represented the structure of device that is used to remove grid noise according to an embodiment of the invention;

Fig. 5 illustration the correction coefficient generation unit how to generate correction coefficient;

Fig. 6 illustration correcting unit shown in Figure 4 how to carry out correction;

Fig. 7 has represented the curve chart according to the mean value of the green pixel on the horizontal direction after removing grid noise of an embodiment of the invention;

Fig. 8 has represented the enlarged drawing of the part of the dashed lines labeled among Fig. 7;

Fig. 9 has represented the example of having removed the image of grid noise according to of the present invention;

Figure 10 is the optimized example of the correction coefficient that obtained according to the present invention of illustration; And

Figure 11 has represented the example according to the correction coefficient that obtains from 6 * 6 masks of an embodiment of the invention.

Embodiment

By the description below with reference to accompanying drawing, above-mentioned purpose, feature and advantage will become more apparent.Should be noted that to parts identical in the accompanying drawing identical Reference numeral all is provided and no matter figure number how.Below, an embodiment of the invention will be described with reference to the drawings.

Fig. 3 is the example of the mask of the illustration Bayer format-pattern that is input to device of the present invention.The present invention is applied to 5 * 5 masks, and as shown in Figure 3, the center pixel of this mask is green.Although for ease of describing 5 * 5 masks are used as example, the present invention is not limited to 5 * 5 masks.

Fig. 4 is the structure of device that is used to remove grid noise according to an embodiment of the invention.

As shown in the figure, the device that is used to remove grid noise according to the present invention comprises capable average calculation unit 410, row estimation unit 420, difference computational unit 430, correction coefficient generation unit 440 and correcting unit 450.

Row average calculation unit 410 is calculated each mean value of going of the Bayer format-pattern of input as shown in Figure 3.Calculate the mean value of each row by one of following formula:

[formula 1]

$\frac{\mathrm{<figure-callout\; id="1"\; label="G"\; filenames="S200680049994XE00111.png"\; state="\{\{state\}\}">G</figure-callout>}1+G2+G3}{3}$

[formula 2]

$\frac{\mathrm{<figure-callout\; id="4"\; label="G"\; filenames="S200680049994XE00111.png"\; state="\{\{state\}\}">G</figure-callout>}4+G5}{2}$

[formula 3]

$\frac{\mathrm{<figure-callout\; id="6"\; label="G"\; filenames="S200680049994XE00051.png"\; state="\{\{state\}\}">G</figure-callout>}6+G7+G8}{3}$

[formula 4]

$\frac{\mathrm{<figure-callout\; id="9"\; label="G"\; filenames="S200680049994XE00051.png"\; state="\{\{state\}\}">G</figure-callout>}9+G10}{2}$

[formula 5]

$\frac{G11+G12+G13}{3}$

Row estimation unit 420 utilizes the mean value of each row, estimates the mean value between the row.That is, utilize the mean value of the pixel of first row and the third line to come second row is estimated, utilize the mean value of the pixel of second row and fourth line to come the third line is estimated, utilize the mean value of the pixel of the third line and fifth line to come fourth line is estimated.Because have difference in the mean value, as shown in Figure 1, this will estimate to be provided with the row between being expert at, since even number line (or odd-numbered line) does not have the information about odd-numbered line (or even number line).

Difference between the mean value that calculates by the estimated value that will obtain by above-mentioned steps with by row average calculation unit 410 averages, and with averaged result divided by 2, will obtain corresponding to the even number line among Fig. 1 and half value at interval between the odd-numbered line.Will further describe this point after a while.

The estimated value of second row, the third line and the fourth line of being estimated by row estimation unit 420 is illustrated among the following formula 6-8.Carry out the estimation of capable estimation unit 420 by the mean value that calculates two row.

[formula 6]

$\frac{\mathrm{<figure-callout\; id="1"\; label="G"\; filenames="S200680049994XE00111.png"\; state="\{\{state\}\}">G</figure-callout>}1+G2+G3-G6+G7+G8}{6}$

[formula 7]

$\frac{\mathrm{<figure-callout\; id="4"\; label="G"\; filenames="S200680049994XE00111.png"\; state="\{\{state\}\}">G</figure-callout>}4+G5+G9+G10}{4}$

[formula 8]

$\frac{\mathrm{<figure-callout\; id="6"\; label="G"\; filenames="S200680049994XE00051.png"\; state="\{\{state\}\}">G</figure-callout>}6+G7+G8+G11+G12+G13}{6}$

Follow with reference to figure 4, difference computational unit 430 is calculated by the estimated value of row estimation unit 420 acquisitions with by the difference between the mean value of row average calculation unit 410 acquisitions.In other words, obtain as shown in Figure 1 the even number line curve and the difference between the odd-numbered line curve.Therefore, as follows by the difference of difference computational unit 430 acquisitions:

[formula 9]

$\frac{\mathrm{<figure-callout\; id="1"\; label="G"\; filenames="S200680049994XE00111.png"\; state="\{\{state\}\}">G</figure-callout>}1+G2+G3+G6+G7+G8}{6}-\frac{G4+G5}{2}$

$=\frac{2\&times;(\mathrm{<figure-callout\; id="1"\; label="G"\; filenames="S200680049994XE00111.png"\; state="\{\{state\}\}">G</figure-callout>}1+\mathrm{<figure-callout\; id="2"\; label="G"\; filenames="S200680049994XE00051.png"\; state="\{\{state\}\}">G</figure-callout>}2+\mathrm{<figure-callout\; id="3"\; label="G"\; filenames="S200680049994XE00111.png"\; state="\{\{state\}\}">G</figure-callout>}3+\mathrm{<figure-callout\; id="6"\; label="G"\; filenames="S200680049994XE00051.png"\; state="\{\{state\}\}">G</figure-callout>}6+\mathrm{<figure-callout\; id="7"\; label="G"\; filenames="S200680049994XE00101.png"\; state="\{\{state\}\}">G</figure-callout>}7+G8)-6\&times;(\mathrm{<figure-callout\; id="4"\; label="G"\; filenames="S200680049994XE00111.png"\; state="\{\{state\}\}">G</figure-callout>}4+G5)}{12}$

[formula 10]

$\frac{\mathrm{<figure-callout\; id="6"\; label="G"\; filenames="S200680049994XE00051.png"\; state="\{\{state\}\}">G</figure-callout>}6+G7+G8}{3}-\frac{G4+G5+G9+G10}{4}$

$=\frac{4\&times;(\mathrm{<figure-callout\; id="6"\; label="G"\; filenames="S200680049994XE00051.png"\; state="\{\{state\}\}">G</figure-callout>}6+\mathrm{<figure-callout\; id="7"\; label="G"\; filenames="S200680049994XE00101.png"\; state="\{\{state\}\}">G</figure-callout>}7+G8)-3\&times;(\mathrm{<figure-callout\; id="4"\; label="G"\; filenames="S200680049994XE00111.png"\; state="\{\{state\}\}">G</figure-callout>}4+G5+\mathrm{<figure-callout\; id="9"\; label="G"\; filenames="S200680049994XE00051.png"\; state="\{\{state\}\}">G</figure-callout>}9+G10)}{12}$

[formula 11]

$\frac{\mathrm{<figure-callout\; id="6"\; label="G"\; filenames="S200680049994XE00051.png"\; state="\{\{state\}\}">G</figure-callout>}6+G7+G8+G11+G12+G13}{6}-\frac{G9+G10}{2}$

$=\frac{2\&times;(\mathrm{<figure-callout\; id="6"\; label="G"\; filenames="S200680049994XE00051.png"\; state="\{\{state\}\}">G</figure-callout>}6+\mathrm{<figure-callout\; id="7"\; label="G"\; filenames="S200680049994XE00101.png"\; state="\{\{state\}\}">G</figure-callout>}7+\mathrm{<figure-callout\; id="8"\; label="G"\; filenames="S200680049994XE00051.png"\; state="\{\{state\}\}">G</figure-callout>}8+G11+G12+G13)-6\&times;(\mathrm{<figure-callout\; id="9"\; label="G"\; filenames="S200680049994XE00051.png"\; state="\{\{state\}\}">G</figure-callout>}9+G10)}{12}$

Herein, the dual numbers row has been used the calculating of (estimated value)-(mean value), and the calculating that odd-numbered line has been used (mean value)-(estimated value), because as shown in Figure 1, even number line generally has the mean value less than estimated value, and odd-numbered line generally has the mean value greater than estimated value.Therefore, use in the back during the correction coefficient, odd-numbered line is carried out subtraction, and the dual numbers row is carried out addition.

Correction coefficient generation unit 440 among Fig. 4 utilizes the difference that obtains by formula 9-11 to generate the correction coefficient alpha that will be used for the Bayer format-pattern.As shown in Equation 12, correction coefficient alpha is half of the mean value that obtains by formula 9-11.

[formula 12]

$\mathrm{\&alpha;}=\frac{2G1-2\mathrm{<figure-callout\; id="2"\; label="G"\; filenames="S200680049994XE00051.png"\; state="\{\{state\}\}">G</figure-callout>}2+2G3-9G4-9G5+8\mathrm{<figure-callout\; id="6"\; label="G"\; filenames="S200680049994XE00051.png"\; state="\{\{state\}\}">G</figure-callout>}6+8G7-8G8-9G9-9G10+2G11+3G12+2G13}{12\&times;3\&times;2}$

Fig. 5 has carried out diagram to this formula, wherein illustration this correction coefficient generation unit be how to generate correction coefficient.As shown in the figure, can multiply by a constant by 5 * 5 masks and generate correction coefficient as shown in Figure 3 Bayer form.In the drawings, " .X " expression is positioned at the multiple of the component of same position.

The Bayer format-pattern that the correction coefficient alpha that correcting unit 450 obtains aforementioned calculation is applied to import is to remove grid noise.In other words, correcting unit 450 correction coefficient alpha that will obtain at last is added to the green component of Bayer format-pattern as shown in Figure 3 or deducts this last correction coefficient alpha that obtains from the green component of this Bayer format-pattern.Below with reference to Fig. 6 this is further described.

Fig. 6 illustration the correcting unit 450 among Fig. 4 be how to carry out correction.

As shown in the figure, correction coefficient alpha is added to the green component of Bayer format-pattern or deducts from the green component of Bayer format-pattern.As previously mentioned, when using correction coefficient, odd-numbered line is carried out subtraction, and the dual numbers row is carried out addition.

Fig. 7 is the curve chart of expression according to the mean value of the green pixel on horizontal direction behind the removal grid noise of one embodiment of the present invention, and Fig. 8 has represented the zoomed-in view of the part of the with dashed lines mark among Fig. 7.

As shown in the figure, as can be seen, compared to Figure 1, the difference of the mean value of the pixel on the horizontal direction has diminished.

Fig. 9 is that expression is according to the example that has been removed the image of grid noise of the present invention.As shown in the figure, as can be seen, compare with the image processing system of routine, grid noise has been removed basically.

Figure 10 is the optimized example of illustration to the correction coefficient of acquisition according to the present invention.In this example, at system the correction coefficient alpha that obtains by above-mentioned steps is carried out optimization.

Although 5 * 5 masks are described as example, the present invention is not limited to 5 * 5 masks.

Figure 11 represents the example according to the correction coefficient that obtains from 6 * 6 masks of one embodiment of the present invention.

As shown in the figure, according to the device that is used to remove grid noise of the present invention, the green component that can also this correction coefficient be added to input picture by the correction coefficient of obtaining N * N mask removes grid noise.

As mentioned above, the present invention has improved traditional filtered method of removing grid noise, and this traditional filtered method can damage image detail (for example, having the edge or the border of a large amount of high fdrequency components), can cause deterioration of image quality.The present invention is by proofreading and correct and adjusting the attribute that keeps the pixel of even number line and odd-numbered line with even number line in the delegation and the difference between the odd-numbered line simultaneously, can remove grid noise and can not damage image detail.

Drawings and detailed description only are examples of the present invention, only are used to describe the present invention and by no means to the spirit and scope of the present invention restrictions or restriction.Thus, any one technical staff of this area should understand can have a large amount of replacements and other to be equal to execution mode.True scope of the present invention should be only limited by the main idea of appended claim.

Claims (5)

1. device that is used to remove grid noise, this device comprises:

Average calculation unit, it is used to calculate mean value of each row of the Bayer format-pattern of input;

Estimation unit, it is used to calculate estimated value, and this estimated value is estimated the mean value that is arranged on the even number line between the odd-numbered line of Bayer format-pattern and the mean value that is arranged on the odd-numbered line between the even number line of Bayer format-pattern;

Difference computational unit, it is used to calculate by the estimated value of this estimation unit calculating with by the difference between the mean value of this average calculation unit calculating;

Correction coefficient calculation, it is used for by utilizing the difference that is generated by this difference computational unit to come the calculation correction coefficient, wherein, the correction coefficient that is obtained by this correction coefficient calculation is the value that the mean value of the difference that will be calculated by this difference computational unit obtains divided by 2; And

Correcting unit, this correcting unit is used for this correction coefficient is applied to the Bayer format-pattern that this Bayer format-pattern and output calibration are crossed, wherein, this correcting unit deducts this correction coefficient from each green component of the odd-numbered line of Bayer format-pattern, and this correcting unit adds this correction coefficient to each green component of the even number line of Bayer format-pattern.

2. device according to claim 1, wherein, this difference computational unit deducts the mean value that is calculated by this average calculation unit by the estimated value from even number line and comes calculated difference.

3. device according to claim 1, wherein, this difference computational unit deducts estimated value by the mean value by this average calculation unit calculating from odd-numbered line and comes calculated difference.

4. device according to claim 1, wherein, for each green component of this Bayer format-pattern is determined this correction coefficient.

5. method of removing grid noise, this method may further comprise the steps:

Calculate the mean value of every row of the Bayer format-pattern of importing;

Calculate estimated value, this estimated value is estimated the mean value that is arranged on the even number line between the odd-numbered line of Bayer format-pattern and the mean value that is arranged on the odd-numbered line between the even number line of Bayer format-pattern;

Difference between the mean value that calculates this estimated value and calculate by the mean value calculation step;

Mean value by the difference that will be calculated comes the calculation correction coefficient divided by 2; And

Deduct this correction coefficient from each green component of the odd-numbered line of Bayer format-pattern, and add the Bayer format-pattern that this correction coefficient and output calibration are crossed to each green component of the even number line of Bayer format-pattern.

Images