KR100468858B1 - Apparatus and method for estimating noise in image - Google Patents

Abstract

The present invention estimates the noise variance of the original image from the average difference image of the original image, calculates the complexity of the image, and corrects the noise variance according to the complexity to improve the accuracy of the noise variance estimation of the original image. The present invention relates to an apparatus and method for estimating the noise level of an image, which is able to make a more complete estimation and ultimately contribute to the improvement of the image quality of the original image. According to the present invention, since it is possible to separate the flat region and the feature region which are the premise of the noise estimation of the image, it is possible to separate only the average difference image without using the gradient value of the gradation value of the original image. Since the threshold value is directly determined from the distribution curve reflecting the characteristics of each frame of the image, and the actual complexity of the image is reflected when estimating the variance of the noise, the accuracy of the estimation can be further improved.

Description

Apparatus and method for estimating noise in image}

The present invention relates to an apparatus and method for estimating noise level of an image. In particular, an accuracy of noise estimation can be obtained by reflecting a characteristic of an image to noise level estimation, which is a prerequisite for adaptive image processing for image quality improvement. An apparatus and method are provided.

Estimation of the noise level of an image is an essential procedure that must be preceded in adaptive image processing for optimal reproduction of the image. The most important factor in the estimation is the accuracy of the estimation and the minimization of the calculation amount for the estimation.

There are many conventional techniques for noise estimation, and there are many domestic and international patent applications. As an example of a number of related technologies, Olsen, in his article "Estimation of noise in images: An Evaluation", CVGIP: Graphical Models and Image Processing, Vol. 55, No. 4, July pp319-323, 1993, Five representative methods are mentioned. The five methods are called Average Method, Gradient Method, Black Method, Scatter Method, and Pyramid Method. Since it is described in detail in one paper, it is weak. Olsen introduces in his paper that the mean method is the best method for estimation in many respects.

Another example of several related technologies is the article by K.Rank, M.Lendl and R.Unbehauen, "Estimation of image noise variance", IEEE Proc-Vis. Image Signal Process, Vol. 146, No. 2 pp80-84, April 1999, iterative method, which includes the directional difference, the histogram of local standard deviation, and the iterative histogram analysis. ) To estimate the noise. This method does not consider the image complexity and has a problem in that the accuracy of estimation is inferior.

Another example of the related technologies is a signal-dependent method described in US Patent 5923775, "Apparatus and Method for Signal Dependent Noise Estimation and Reduction in Digital Images," which is an image segmentation method. ), A method of estimating noise using an image gradient histogram and a histogram threshold calculation. This method has a disadvantage in that a large amount of calculation is required for estimation.

The average method mentioned above is the most widely used for the estimation of noise. The average method obtains an average difference image of the original image by obtaining an average image of the original image and then obtaining an absolute value of the difference between the original image and the average image. When the average difference image is obtained, almost all imagefeature components of the low frequency band can be removed, and only the high frequency image signal components, that is, the feature region components and the noise components, exist in the average difference image. In addition to obtaining the mean difference image, the gradient of the original image is also calculated. The gradient is an index indicating the degree of flattening of the image, and the smaller this value is, the flatter the image is. The calculation of the gradient is performed to distinguish between the feature area and the uniform area of the original image. The distinction between these two areas in the image is very important for noise estimation because the noise component in the feature region does not have a significant effect on the image reproduction, but the noise component in the flat region has a lot of negative effects on the image reproduction quality. Therefore, the distinction between these two areas is very important for noise estimation because it can pursue accuracy and calculation efficiency in noise estimation. In the division of the area, a threshold value is given to a gradient value, and a portion having a predetermined value or less is set as a flat area, and noise variance of the original image is estimated.

However, the previous average method has the following problems.

In order to obtain an average image of the original image, linear-fit is performed using the least-square method on the original image, which causes a prolonged computation time. In order to solve this problem, the gradient of the original image must be obtained separately, and the threshold of the gradient is fixed for all image frames. Therefore, the characteristic characteristics and the flat region cannot be accurately distinguished because the image characteristics do not reflect different aspects in each frame. have. In addition, the problem of excessive estimation of noise variance in a complex image having a low noise level (image having many feature areas), that is, the complexity of the image is not reflected and also includes a problem of inaccuracy in calculating the noise variance.

Therefore, the present invention was devised to solve the above problems of the prior arts, and an object of the present invention is to reduce the amount of computation required for noise variance estimation and to improve the accuracy of estimation. The present invention provides an apparatus and method for estimating noise level of an image.

1A is a block diagram of the present invention.

1b is a detailed configuration diagram of an image characteristic analyzer;

1C is an example of an original image.

1D is an example of an average image.

1E is an example of an average difference image.

1F is an example of a distribution curve for determining a distribution of average difference image gray values and a threshold value.

1G is a detailed configuration diagram of a correction variance calculation unit.

1H is an example of the graph for calculation of a correction constant.

2A is a flow chart of the present invention.

Figure 2b is a detailed flow diagram of the original image analysis step.

Figure 2c is a detailed flowchart of the correction variance calculation step.

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

10: image characteristic analysis unit 11: correction variance calculation unit

101: LPF 102: ABS 103: threshold value calculation unit

104: average / complexity calculation unit 105: estimated variance calculation unit

111: correction constant calculation unit 112: multiplier

In order to achieve the above object, the noise level estimation apparatus of an image provided by the present invention estimates the complexity of the original image and the estimated variance of the noise of the original image from the distribution of the gray level values of the average difference image obtained from the gray level values of the original image. And a correction variance calculator for calculating a correction variance corrected according to a function of complexity.

In addition, in order to achieve the above object, the noise level estimation method of an image provided by the present invention includes a method for estimating the original image from a gray scale distribution of an average difference image obtained from gray scale values of an estimated target image including noise having a known variance. And an image characteristic analysis step of calculating the estimated variance of the noise and the complexity of the original image, and a correction variance calculating step of calculating the correction variance by reflecting the complexity in the estimated variance.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT Hereinafter, the configuration, operation, and optimal embodiments of the present invention will be described in detail with reference to the accompanying drawings, in which like reference numerals are used to designate like elements. It is noted that in the description of the drawings, components of other drawings may be cited as necessary.

Figure 1a is a block diagram of the present invention and Figure 1b is a detailed block diagram of the original image analysis unit.

The image characteristic analyzer 10 may use pixel values of the original image frame ( After obtaining the pixel value of the average difference image of the original image from the estimated distribution of the original image noise from the distribution of this pixel value ) And complexity ( ) Is calculated. Here, the subscript k is the sequence number of the original image frame. In other words, Is the estimated variance of the kth frame Denotes the complexity of the k-th frame. The acquisition of the average difference image pixel value is preferably performed for each pixel value of the original image ( ) And low pass filtering (LPF, 101) ). There are various methods of low pass filtering. For example, a low pass filter matrix (for example, a 3x3 low pass filter matrix) is scanned into a frame of an original image. It may be a way to. After filtering, the pixel value of the original image ( ) And pixel values of the filtered image ( ), And the absolute value (ABS) 102 is taken as the difference to obtain the pixel value (| - Is obtained. In this case, the pixel value generally represents a gray level of the pixel. In the noise estimation of the image, the gray value is mainly used for the estimation process, and therefore, the gray value is used instead of the term 'pixel value'. Exemplary views of the original image, the filtered image, and the average difference image are shown in FIGS. 1C to 1E, respectively.

The threshold calculating unit 103 calculates a threshold value for distinguishing the flat region and the feature region of the original image. The threshold value calculating unit 103 first obtains the gray level value of the average difference image (| - With respect to |), a gray scale value distribution curve of the average difference image is extracted, in which the horizontal axis is the gray scale value magnitude (x) of the average difference image and the vertical axis is the frequency value y of the gray scale value. At this time, the distribution curve may be embodied as a histogram curve. An example of this histogram curve is shown in FIG. 1F. The threshold for distinguishing the flat and feature areas of the original image from this curve ). Although the present invention is based on the previous average method, the noise is estimated by histogramizing the gray level value of the average image as it is, without obtaining a gradient of the gray level value of the original image. Because it is close to the layout. Therefore, the flat area and the feature area of the original image can be distinguished only by the gray level value of the average image.

The first decision is to find the value on the horizontal axis (Xpeak) that corresponds to the maximum frequency value (y) on the histogram distribution curve (shown in Figure 1f) and then (Xpeak + a = X1, Y1), Xpeak spaced a distance from Xpeak. Find (Xpeak + b = X2, Y2) separated by b from. The decision criteria for a and b are values determined experimentally, not based on any mathematical principle. However, the derivative of the mean value of the frequency value (y) and the gradation value (x) of the image ( In other words, a and b are determined within a slope of a tangent line with respect to any point of the curve is less than (or more than) a predetermined value. Otherwise, an error may occur in which the feature region may actually be determined to be a flat region. The predetermined value is determined according to the shape of the distribution curve, and the point at which the slope changes rapidly will be the standard for determining the predetermined value. Next, find the equation of the line connecting these two points, and the coordinates of the point where the line meets the horizontal axis of the Will be the value corresponding to.

The average calculator 104 calculates a threshold value from the threshold calculator 103. ) Is the average of the average difference image gray value ( The complexity calculator 105 calculates a threshold value from the threshold calculator 103. ), The complexity of the kth image frame (complexity, ) Are preferably calculated by Equation 1 below.

here Is an index indicating the smoothing degree of the k-th frame of the original image. According to Equation 1 Has a range from 0 to 1 Is closer to 1 ( The closer to is 0, the more the original image has a flat area, which means that the complexity is low. Here, Xmax means the maximum value of the average difference image gray value.

The estimated variance calculator 106 estimates the variance from the mean (Xcent, k) of the mean difference image gray values. This can be obtained by Equation 2 below.

Where const is a constant determined by experiment.

The image characteristic analyzer 10 shown in FIG. 1B may function as an image characteristic analyzer by itself. That is, the complexity and estimation variance obtained using the image characteristic analyzer may be combined with other image processing apparatuses to be associated with performance of functions of other processing apparatuses if necessary. For example, if there is a chromaticity adjusting device for an image, and when complexity and estimation dispersion are required to adjust chromaticity, the chromaticity adjusting device receives these two values from the image characteristic analyzer and performs a predetermined chromaticity adjusting function. do.

1G is a detailed configuration diagram of the correction dispersion calculator.

The necessity of calculating the variance variance is to ensure accuracy in estimating the noise variance of an image having a large number of feature areas (low noise level). As mentioned above, the conventional average difference method has a problem that the noise variance is excessively estimated for the image having many feature regions. The reason for this is that noise is estimated in a uniform manner without considering the complexity of the image. Therefore, it is not possible to distinguish between the original image and the feature region. Therefore estimated variance For areas with many feature areas For small areas), the correction needs to be reduced, exactly.

The correction variance calculator 11 estimates the variance calculated by the estimated variance calculator 105 ( ) On complexity ( This part calculates the correction variance reflecting). The correction constant calculator 111 calculates from the estimated variance calculator 105. Is inputted using a function of complexity set in a predetermined manner. Correction constant reflecting complexity from The function of complexity in calculating a may be set by a predetermined modeling process and provides an example of the modeling method.

The modeling is an estimated variance calculated by the estimated variance calculator 106. And complexity It is the task of establishing the relationship. First, baseline variance ( ) And estimated variance Of ( ) At this time Is the original image gradation value This is a randomly set value of the actual noise variance of. The reason for setting the reference variance is that the degree of correction of the estimated variance can be predicted based on the reference variance. this Although it is possible to predict the degree of correction of the estimated variance alone, the complexity of the image There is a problem to take this as a correction constant since it is not reflected. therefore on Should be used to calculate the correction constant.

1H Wow This is an example of a graph for setting the relationship of.

Wow The relationship setting of each Corresponding to The points corresponding to (x points on FIG. 1H) are shown on the coordinate plane, and then the points are connected to extract a curve having a predetermined shape. This curve is approximated by a curve having a certain regularity (bold solid line in Fig. 1H). From approximated curve of It can be approximated as a function of. This function is usually represented by an nth order constant and is also given as a second or lower order constant for most images. An example of a quadratic constant function is shown in equation (3).

1H shows the i-th frame , , Value , , Is presented.

This function is set as default in the correction constant calculating section 111 and is a value obtained by this function. Will be taken. At this time , , Can be obtained by using various statistical analysis techniques such as regression technique, and the detailed analysis is weak because statistical analysis techniques are described in detail in the literature on statistics.

The multiplier 112 from the correction constant calculation unit 111 Input and estimate it Correction variance by multiplying by Will yield.

2A is a flowchart of the present invention, and FIG. 2B is a detailed flowchart of an image characteristic analysis step.

In the image characteristic analysis step (S20), first, a gray value of the original image ( ) Is input and low-pass filtered to adjust the grayscale value of the filtered image ( ). As mentioned in the device invention, there are various methods of low pass filtering. For example, a low pass filter matrix is scanned into a frame of an original image. It may be a way to. After filtering Wow Finds the difference between the two differences and takes the absolute value of the difference. - Obtain (S201). Next, the distribution curve for the gray level distribution of the average difference image is extracted to determine the threshold for separation of the flat region and the feature region (S202). can do. Once the threshold is determined, the mean (Xcent, k) and the complexity of the original image ) Is calculated as described above (S203) and the estimated variance ( ), The image characteristic analysis step S20 is terminated. Where Xcent, k And Can be obtained in the same manner as above.

The image characteristic analysis step S20 shown in FIG. 2B may function as an image characteristic analysis method by itself. That is, the complexity and the estimated variance obtained using the image characteristic analysis method may be combined with other image processing methods if necessary to be related to the performance of other processing methods. For example, if there is a method of adjusting the chromaticity of the image, when the complexity and the estimated dispersion are required to adjust the chromaticity, the chromaticity adjustment method receives the two values through the image characteristic analysis method and performs a predetermined chromaticity adjustment function. You can do it.

2C is a detailed flowchart of the correction variance calculation step.

This step (S21) is the complexity of the original image ( ) Is a step of estimating the noise level for the original image by correcting the estimated variance to reflect the estimated variance. Correction constant from Calculating (S211) by multiplying the correction constant and the estimated variance Comprising the step (S212). Can be obtained as in the device invention.

The present invention can also be embodied as computer readable code on a computer readable recording medium. The computer-readable recording medium includes all kinds of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disks, optical data storage devices, and the like, which may also be implemented in the form of carrier waves (for example, transmission over the Internet). do. The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

Although the preferred mode of operation of the present invention has been described in detail by way of example, those skilled in the art will appreciate that the present invention may be made without departing from the spirit and scope of the invention as defined in the appended claims. It will be appreciated that the various modifications or changes may be made. Accordingly, modifications to the foregoing embodiments of the present invention will not depart from the techniques of the present invention.

The use of the present invention provides the following advantages.

Unlike in the past, it is possible to separate the flat area and the feature area, which are the premise of the noise of the image, without using the gradient of the gradation value of the original image. In addition, since the threshold is directly determined from the distribution curve reflecting the characteristics of each frame of the image and the actual complexity of the image is reflected when the variance of the noise is estimated, the accuracy of the estimation can be further improved.

Claims (19)

An image characteristic analyzer for estimating the variance of the noise of the original image and the complexity of the original image from the distribution of the gray level values of the average difference image obtained from the gray level values of the original image; And And a correction variance calculator for calculating a correction variance correcting the estimated variance as a function of the complexity. The method of claim 1, wherein the gray value of the average difference image is And obtaining a filtered image by performing low pass filtering on the original image, obtaining a difference between a gray value of the original image and a gray value of the filtered image, and then obtaining an absolute value of the difference. The method of claim 1, wherein the image characteristic analyzer The horizontal axis extracts a gray scale value distribution curve of the average difference image having the magnitude (x) of the gray scale value of the average difference image and the frequency y of the x, and extracts the gray scale distribution curve from the distribution curve. A threshold calculator for calculating a threshold for distinguishing feature regions; An average calculator configured to calculate an average of gray levels of the average difference image from the threshold value; A complexity calculator for calculating the complexity from the threshold value; And And an estimated variance calculator for calculating the estimated variance from the average. The method of claim 1, wherein the correction variance calculator A correction constant calculator which receives the complexity and calculates a correction constant using a function of the complexity; And And a multiplier for multiplying the correction constant by the estimated variance to calculate the corrected variance. 5. The method of any one of claims 1, 3 and 4, wherein the complexity is And an apparatus for estimating a noise level of an image based on a value obtained by dividing the total number of pixels of the flat region of the original image by the total number of pixels of the original image. 4. The method of claim 3, wherein the threshold is The two x values (x1, x2) (x2> x1) spaced apart by a predetermined distance on the horizontal axis of the curve from the x value corresponding to the maximum value of y and the two y values (y1, y2) of the corresponding vertical axis are described. An apparatus for estimating noise level of an image, characterized by (x1, y1) and (x2, y2) coordinates on a curve, wherein a straight line passing through these two points is determined as an x value of a point crossing the horizontal axis. 5. The method according to any one of claims 1, 3 and 4, wherein the estimated variance is The noise level estimation apparatus of the image, characterized in that calculated from the average of the average image gray value. The method of claim 1 or 4, wherein the correction constant Original image gradation value Reference variance arbitrarily set to the actual noise variance of And the estimated variance Of ( ), And then And complexity Above for setting relations of The angle corresponding to each value in Extract the curve connecting the points by drawing the points corresponding to the coordinate plane, approximate the curve to a curve having a predetermined regularity, and then express the approximated curve as a function of the complexity Noise level estimation apparatus for an image, characterized in that it is set to a value calculated by. A filtered image acquisition unit for low-pass filtering the original image to obtain a filtered image; An absolute value calculator which calculates an absolute value of the difference by obtaining a difference between the gray value of the original image and the gray value of the filtered image; The horizontal axis extracts a gray scale value distribution curve of the average difference image having the magnitude (x) of the gray scale value of the average difference image and the frequency y of the x, and extracts the gray scale distribution curve from the distribution curve. A threshold calculator for calculating a threshold for distinguishing feature regions; An average calculator configured to calculate an average of gray levels of the average difference image from the threshold value; A complexity calculator for calculating the complexity from the threshold value; And And an estimated variance calculator for calculating the estimated variance from the average. An image characteristic analysis step of calculating the estimated variance of the noise of the original image and the complexity of the original image from the distribution of the gray level values of the average difference image obtained from the gray level values of the original image; And And calculating a correction variance for correcting the estimated variance as a function of the complexity. The method of claim 10, wherein the average difference image And obtaining a filtered image by performing low pass filtering on the original image, obtaining a difference between the gray value of the original image and the gray value of the filtered image, and then obtaining the absolute value of the difference. The method of claim 10, wherein the original image analysis step The horizontal axis is the magnitude (x) of the gray scale value of the average difference image, and the vertical axis is the frequency of the magnitude (y), and the gray scale value distribution curve of the average difference image is extracted and the flat region of the original image is extracted from the distribution curve. A threshold value calculating step of calculating a threshold for classification of the feature region; An average / complexity calculation step of calculating an average and the complexity of the gray level values of the average difference image from the threshold value; And And an estimation variance calculating step of calculating the estimated variance from the average. The method of claim 10, wherein the step of calculating the correction variance Calculating a correction constant using the function of the complexity; And And calculating the correction variance by multiplying the correction coefficient by the estimated variance. 13. The method of claim 10 or 12, wherein the complexity is The noise level estimation method of the image, characterized in that determined based on a value obtained by dividing the total number of pixels of the flat region of the original image by the total number of pixels of the original image. 13. The method of claim 12 wherein the threshold is The two x values (x1, x2) (x2> x1) spaced apart by a predetermined distance on the horizontal axis of the curve from the x value corresponding to the maximum value of y and the two y values (y1, y2) of the corresponding vertical axis are described. A method of estimating noise level of an image, characterized by (x1, y1) and (x2, y2) coordinates on a curve, wherein a straight line passing through these two points is determined as an x value of a point intersecting the horizontal axis. 14. The method according to any one of claims 10, 12 and 13, wherein the estimated variance is The noise level estimation method of the image, characterized in that calculated from the average of the average image gray value. The method according to claim 10 or 13, wherein the correction constant is Original image gradation value Reference variance arbitrarily set to the actual noise variance of And the estimated variance Of ( ), And then And complexity Above for setting relations of The angle corresponding to each value in Extract the curve connecting the points by drawing the points corresponding to the coordinate plane, approximate the curve to a curve having a predetermined regularity, and then express the approximated curve as a function of the complexity The noise level estimation method of the image, characterized in that it is set to a value calculated by. Obtaining a filtered image by performing low pass filtering on the original image to obtain a filtered image; Calculating an absolute value of the difference by obtaining a difference between the gray value of the original image and the gray value of the filtered image; The horizontal axis is a magnitude (x) of the gray level value of the average difference image, and the vertical axis is a gray level value distribution curve of the average difference image having the frequency y, and the flat region of the original image is extracted from the distribution curve. A threshold value calculating step of calculating a threshold for classification of the feature region; Calculating an average of gray levels of the average difference image from the threshold value; A complexity calculation step of calculating the complexity from the threshold value; And And an estimation variance calculating step of calculating the estimated variance from the average. 19. A recording medium having a computer-readable method of recording the noise level estimating method according to any one of claims 10 to 18, and recorded with executable program code.