KR100499121B1 - Apparatus and method for image segmentation using automatic thresholding technique - Google Patents

Abstract

An apparatus and method for image segmentation by automatic thresholding for fitting a curve to an image histogram.

An image segmentation apparatus using an automatic thresholding technique according to the present invention comprises a memory for storing a digital image; A histogram generator for creating a histogram of a digital image stored in a memory; A predominant region determiner for dividing the image histogram into one or more predominant regions in the image histogram; A suitable part determining unit which determines a left fitting part and a right fitting part for each predominant part; A curve fitting portion for fitting a separate curve to the left fitting portion and the right fitting portion, respectively; A threshold determination unit that determines a left threshold and a right threshold of the image histogram based on a standard deviation of the fitted curve; And a pixel classification unit for dividing the digital image by classifying pixels having pixel values between a left threshold and a right threshold corresponding to each dominant portion in the digital image stored in the memory.

According to the present invention, the image histogram is divided into several regions and the curve fitting is partially performed to increase the reliability of the thresholding.

Description

Apparatus and method for image segmentation using automatic thresholding technique

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to image signal processing, and more particularly, to an image segmentation apparatus and method by automatic thresholding for fitting a curve to an image histogram.

In general, image segmentation is an important process for feature extraction, volume measurement of a region of interest, and the like in computer vision, pattern recognition, image visualization, image analysis, and the like. The image segmentation methods known to date include histogram analysis, morphological operations, thresholding, anisotropic filtering, and active surface templates. In using these methods, in the early stages, the image was segmented using an automatic thresholding technique that fits and models the image histogram as an arbitrary curve.

Among the existing methods for segmenting images using automatic thresholding, Marizen proposed [Marijn E. Brummer, IEEE Trans. Medical Imaging, vol. 12, 1993] and Stellar's proposed method [M. Stella Atkins, IEEE Trans. Medical Imaging, vol. 17, 1998] modeled the image histogram with Gaussian curves and thresholded it through curve fitting. However, when the curve is fitted to the image histogram, the model curve is often far from the image histogram. 1 shows a Gaussian curve that is a model curve, and FIGS. 2A and 2B exemplarily show an image histogram to be fitted. As shown in FIGS. 2A and 2B, an image histogram to be fitted is often not properly modeled by a Gaussian distribution. Therefore, if Figs. 2A and 2B are fitted to Fig. 1, the fitting error is severe, which causes an error in automatic thresholding, which in turn lowers the reliability of division.

The present invention was created to solve the above problems, can be used in image processing hardware and software for computer vision, pattern recognition, image visualization, image analysis, etc., automatic thresholding technology that can increase the reliability of image segmentation An object of the present invention is to provide a video splitting apparatus and a method thereof.

In order to achieve the above object, the image segmentation apparatus using the automatic thresholding technique according to the present invention comprises a memory for storing a digital image; A histogram generator for generating an image histogram by calculating a number of pixels corresponding to each pixel value of the digital image stored in the memory; The maximum value is obtained from the image histogram, and the determination of the dominant part, which is the part between the nearest minima smaller than the maximum value of the left and right minima of the maximum value, multiplied by a predetermined first ratio, is repeated as one or more dominant parts. A dominant region determiner for dividing the image histogram; Each predominant part also takes a part with a value that is greater than the maximum value multiplied by a predetermined second ratio, and the left fit portion between the left maximum value that is the maximum value closest to the left minimum value from the left minimum value of the portion and the right minimum value of the portion. A fitting site determining unit for determining a right fitting site between the right minimum value and the right maximum value which is the closest maximum value in the first position; A curve fitting portion for fitting separate curves to the left fitting portion and the right fitting portion, respectively; A threshold determination unit that determines a left threshold and a right threshold of the image histogram based on a standard deviation of the fitted curve; And a pixel classification unit for dividing the digital image by classifying pixels having pixel values between a left threshold and a right threshold corresponding to each dominant portion in the digital image stored in the memory.

In order to achieve the above object, the image segmentation method using the automatic thresholding technique according to the present invention comprises the steps of (a) generating an image histogram by calculating the number of pixels corresponding to each pixel value of the digital image; (b) determining the dominant region, which is the area between the closest local minimums, which is obtained from the image histogram and is smaller than the maximum of left and right minima of the maximum value multiplied by a first ratio. Dividing the image histogram into a dominant region; (c) each predominantly separate site shall have a site having a value greater than the maximum value multiplied by a second predetermined ratio, and the left fit site and its site between the left maximum of the site and the left maximum which is the closest maximum to the left minimum; Determining a right fit region between the right minimum value and the right maximum value that is the closest maximum value in the right minimum value of the right value; (d) fitting separate curves to the left fitting portion and the right fitting portion, respectively; (e) determining a left threshold and a right threshold of the image histogram based on the standard deviation of the fitted curve; And dividing the digital image by classifying pixels having pixel values between a left threshold and a right threshold corresponding to each dominant portion in the digital image.

Hereinafter, with reference to the accompanying drawings will be described in detail the present invention.

According to FIG. 3A, an embodiment of an image segmentation apparatus using an automatic thresholding technique according to the present invention includes a memory 10, a histogram generator 20, a dominant position determiner 30, and a suitable position determiner 40. , The curve fitting unit 50, the threshold determining unit 60, and the pixel classification unit 70. The histogram generator 20, the dominant site determiner 30, the suitable site determiner 40, the curve fit unit 50, the threshold determiner 60, and the pixel classifier 70 are each an information processing system. It can be implemented as a combination of software running on the information processing system.

The memory 10 stores an input digital image.

The histogram generator 20 generates an image histogram by calculating the number of pixels corresponding to each pixel value of the digital image stored in the memory 10.

The preponderance determiner 30 obtains the maximum value from the image histogram and repeats the determination of the predominant region, which is the area between the closest local minimum values smaller than half of the maximum value on the left and right of the maximum value, and the one or more predominant regions. The image histogram is divided into.

The suitable site determining unit 40 takes a site having a value greater than half of the maximum value of each of the predominant sites, and the left fit site between the left maximum value which is the maximum value closest to the left minimum value and the left maximum value of the site Determine the right fit between the right minimum and the right maximum, which is the closest maximum.

The curve fitting portion 50 fits separate curves to the left fitting portion and the right fitting portion, respectively.

The threshold determination unit 60 determines the left threshold and the right threshold of the image histogram based on the standard deviation of the fitted curve.

The pixel classifier 70 divides the digital image by classifying pixels having pixel values between the left and right thresholds in the digital image stored in the memory 10.

Hereinafter, with reference to Figure 3b will be described in detail the operation of the present invention.

First, a single or multiple set of digital images to be divided are loaded into a memory (step 100).

Next, the number h (f) of pixels for all pixel values f is calculated for the image loaded in the memory 10 (step 200). The calculated image histogram is stored in the memory for processing of the next step.

Then, the image histogram is divided. The segmentation of an image histogram is a process of dividing a histogram portion necessary for determining a threshold value in a created image histogram. The segmentation of the image histogram is performed by finding a dominant portion of the image histogram (step 300), and then extracting a suitable portion from the dominant portion (step 400).

As shown in FIG. 4, the maximum value H _max = h (f _max ) of the histogram h (f) is found, and then, if defined as one chunk by taking up to half of the maximum value, the whole is set as the dominant portion, and If it is a chunk, the pixel values f are divided in order from the smallest and set to the dominant regions. The boundary f _min between one dominant region and a neighboring dominant region is determined as follows. Here, when the left position and the right position corresponding to the value corresponding to half of the maximum value H _max are f _minL and f _minR

for (i = f _minR ; i <N; i ++) {

   flag = 1;

for (j = i; j <i + T; j ++) {

if (h (i)> h (j)) {flag = 0; break; }

   }

if (flag) {f _min = i; break; }

}

for (i = f _minL; i>0; i--) {

   flag = 1;

for (j = i; j> i-T; j--) {

if (h (i)> h (j)) {flag = 0; break; }

   }

if (flag) {f _min = i; break; }

}

Herein, N denotes a range of pixel values f, T denotes a constant, and a constant T varies depending on an application image, but in an embodiment of the present invention, T = 20. As f _min is illustrated in the image histogram illustrated in FIG. 4, f _min means a position of a minimum value located between two dominant regions, and determines a range of dominant regions. Thereafter, the predominant region found in the image histogram h (f) is removed, and the above-described predominant region determination process is repeated.

For each predominant portion of the split image histogram to determine the fit region of the image histogram, more than half of its maximum is taken, as illustrated in FIG. 5. When the minimum value La of the left and the right of the minimum value from the region H _minL, H _minR, is obtained as follows: the maximum value closest to the closest peak value _maxR H H _maxL, right and left minimum minimum, respectively. Range of suitable areas of the histogram h _L (f) and h _R (f) is Fig. 6a, a, f from the right and from the left to _{maxL minR} f f f _minL to _maxR as Figure 6b illustratively shown in.

for (i = f _minL ; i <N; i ++) {

   flag = 1;

for (j = i; j <i + T; j ++) {

if (h (i) <h (j)) {flag = 0; break; }

   }

if (flag) {f _maxL = i; break; }

}

for (i = f _minR ; i>0; i--) {

   flag = 1;

for (j = i; j> i-T; j--) {

if (h (i) <h (j)) {flag = 0; break; }

   }

if (flag) {f _maxR = i; break; }

}

Thereafter, curve fitting is performed on the histogram of the separated fit site (step 500). Curve fit can be done in many ways, but the Levenberg-Marquardt method [W. H. Press, Numerical Recipes in C, 1988]. Various curves may be applied to the curve to be fitted, but it is preferable to use a Gaussian curve, and FIGS. 7A and 7B show that the Gaussian curve is fitted to the histogram of the separated fitting site, respectively.

The threshold is determined from the left and right Gaussian curves g _L (f) and g _R (f) fitted to the fit site (step 600). When the range of f = -2σ to 2σ, which occupies 95.9% of the Gaussian curve area, is used as the reference, the left and right thresholds are determined as T _L = -2σ _L and T _R = 2σ _R. Where T _L and T _R are the thresholds of the left and right h _L (f) and h _R (f) of the histogram, respectively, and σ _L and σ _R are the fitted Gaussian curves g _L (f) and g _R (f respectively). Standard deviation.

As a result, the pixels of the image are classified into corresponding classes based on a threshold selected at each dominant region (step 700).

Experimental results of the conventional method for the T1 and PD MR images of FIGS. 8A and 8B and the method of the present invention are shown in FIGS. 9A, 9B, 10A, and 10B. The results obtained by the Gaussian curve using the conventional method are the same as those shown in FIGS. 2A and 2B, and the thresholds divided by ± 2 σ are divided into FIGS. 9A and 9B. As shown in FIG. 2A, the left threshold T _L = 90 fits relatively well but the right threshold T _R = 390 does not fit well with respect to the T1 image, and the upper threshold is incorrect in FIG. 9A. As shown in FIG. 2B, the left threshold T _L = 100 and the right threshold T _R = 850 do not fit well with respect to the PD image, and the upper and lower thresholds are also incorrect in FIG. 9B.

The results using the method of the present invention are shown in Figs. 7a, 7b, 10a and 10b. As shown in FIG. 7A, the left threshold T _L = 90 and the right threshold T _R = 320 were obtained by separately applying Gaussian to the left and right sides of the histogram for the T1 image. Accordingly, in FIG. 10A, an improved result can be seen from the result of the conventional method. Also for PD images, left and right threshold values T _L = 340 and T _R = 630 were obtained by Gaussian fitting the left and right sides of the histogram separately as shown in FIG. 7B. Accordingly, in FIG. 10B, the improved results can be seen from the results by the conventional method.

On the other hand, the embodiments of the present invention described above can be written as a program that can be executed on a computer. And, it can be implemented in a general-purpose digital computer for operating the program using a medium used in the computer. The media may be stored such as magnetic storage media (e.g., ROM, floppy disk, hard disk, etc.), optical reading media (e.g., CD-ROM, DVD, etc.) and carrier waves (e.g., transmission over the Internet). Media.

So far I looked at the center of the preferred embodiment for the present invention. Those skilled in the art will appreciate that the present invention can be implemented in a modified form without departing from the essential features of the present invention. Therefore, the disclosed embodiments should be considered in descriptive sense only and not for purposes of limitation. The scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the scope will be construed as being included in the present invention.

According to the present invention, the image histogram is divided into several regions and the curve fitting is partially performed to increase the reliability of the thresholding.

First, the reliability of image segmentation is improved by reducing the error of curve fitting in image segmentation using threshold value.

Second, it was easy to fit a Gaussian curve because a class could not be modeled as a Gaussian curve in an image histogram.

Third, the image histogram was systematically divided to improve the difficulty of curve fitting.

1 is a graph illustrating a general Gaussian distribution curve.

2A and 2B are graphs showing a Gaussian curve fitted to a histogram of an image.

3A is a block diagram of an image splitting apparatus according to the present invention.

3B is a flowchart of an image segmentation method according to the present invention.

4 is a graph illustrating a method of selecting a dominant region in a histogram of an image.

5 is a graph for explaining a method of selecting a suitable site in a dominant histogram.

6A and 6B are graphs showing selected fit sites.

7A and 7B are graphs illustrating fitting a Gaussian curve to a suitable site.

8A and 8B illustrate MR T1 and PD images.

9A and 9B illustrate the result of image segmentation by applying a conventional method to the MR T1 and PD images illustrated in FIGS. 8A and 8B.

10A and 10B illustrate the results of image segmentation by applying the method of the present invention to the MR T1 and PD images illustrated in FIGS. 8A and 8B.

Claims (7)

A memory for storing digital images; A histogram generator for generating an image histogram by calculating a number of pixels corresponding to each pixel value of the digital image stored in the memory; The maximum value is obtained from the image histogram, and the determination of the dominant part, which is the part between the nearest minima smaller than the maximum value of the left and right minima of the maximum value, multiplied by a predetermined first ratio, is repeated as one or more dominant parts. A dominant region determiner for dividing the image histogram; Each predominant part also takes a part with a value that is greater than the maximum value multiplied by a predetermined second ratio, and the left fit portion between the left maximum value that is the maximum value closest to the left minimum value from the left minimum value of the portion and the right minimum value of the portion. A fitting site determining unit for determining a right fitting site between the right minimum value and the right maximum value which is the closest maximum value in the first position; A curve fitting portion for fitting separate curves to the left fitting portion and the right fitting portion, respectively; A threshold determination unit that determines a left threshold and a right threshold of the image histogram based on a standard deviation of the fitted curve; And And a pixel classifier for dividing the digital image by classifying pixels having a pixel value between a left threshold and a right threshold corresponding to each dominant portion in the digital image stored in the memory. Image splitting device. The method of claim 1, wherein the first ratio and the second ratio are respectively Image segmentation apparatus using an automatic thresholding technology, characterized in that 0.5. The method of claim 1, wherein the curve fitting portion An image segmentation apparatus using an automatic thresholding technique, wherein the curve is fitted using a nonlinear least-squares method, and the curve is a Gaussian curve. (a) generating an image histogram by calculating the number of pixels corresponding to each pixel value of the digital image; (b) determining the dominant region, which is the area between the closest local minimums, which is obtained from the image histogram and is smaller than the maximum of left and right minima of the maximum value multiplied by a first ratio. Dividing the image histogram into a dominant region; (c) each predominantly separate site shall have a site having a value greater than the maximum value multiplied by a second predetermined ratio, and the left fit site and its site between the left maximum of the site and the left maximum which is the closest maximum to the left minimum; Determining a right fit region between the right minimum value and the right maximum value that is the closest maximum value in the right minimum value of the right value; (d) fitting separate curves to the left fitting portion and the right fitting portion, respectively; (e) determining a left threshold and a right threshold of the image histogram based on the standard deviation of the fitted curve; And (f) dividing the digital image by classifying pixels having a pixel value between a left threshold and a right threshold corresponding to each dominant portion in the digital image. Split method. The method of claim 4, wherein the first ratio and the second ratio are respectively Image segmentation using automatic thresholding technology characterized in that 0.5. The method of claim 4, wherein the curve fitting portion A method of segmenting an image using an automatic thresholding technique, characterized by fitting a curve using a nonlinear least-squares method, wherein the curve is a Gaussian curve. (a) a module for generating an image histogram by calculating the number of pixels corresponding to each pixel value of the digital image; (b) determining the dominant region, which is the area between the closest local minimums, which is obtained from the image histogram and is smaller than the maximum of left and right minima of the maximum value multiplied by a first ratio. A module for dividing the image histogram into a dominant region; (c) each predominantly separate site shall have a site having a value greater than the maximum value multiplied by a second predetermined ratio, and the left fit site and its site between the left maximum of the site and the left maximum which is the closest maximum to the left minimum; A module for determining a right fit region between the right minimum value and the right maximum value that is the closest maximum value in the right minimum value of; (d) a module for fitting a separate curve to the left fit portion and the right fit portion, respectively; (e) a module for determining a left threshold and a right threshold of the image histogram based on the standard deviation of the fitted curve; And and (f) a module for dividing the digital image by classifying pixels having pixel values between a left threshold and a right threshold corresponding to each dominant portion in the digital image. A computer-readable recording medium that records a divided program.