JP6365830B2 - Antibacterial effect determination system, antibacterial effect determination method, and antibacterial effect determination program - Google Patents

Description

  The present invention relates to an antibacterial effect determination system, an antibacterial effect determination method, and an antibacterial effect determination program for determining an antibacterial effect of an antibacterial agent having a function of suppressing or killing the activity of bacteria related to structures and the like.

  Conventionally, the Japan Industrial Standards Committee (JIS) has specified antibacterial test methods and antibacterial effects against bacteria on the surface of antibacterial processed products such as plastic products, metal products, and ceramic products (for example, non-patent literature) 1). Hereinafter, this method is referred to as a first conventional example.

  Furthermore, JIS has prescribed | regulated about the test method, antibacterial effect, and display which evaluate the antibacterial property with respect to bacteria of the textiles which gave the antibacterial process (for example, refer nonpatent literature 2). Hereinafter, this method is referred to as a second conventional example. In the halo test of the second conventional example, the bacterial solution of the test bacteria is uniformly smeared on the surface of the agar plate medium formed in the petri dish, and then a disc-shaped test piece is left on the surface of the agar plate medium. . The test piece consists of an antibacterial and deodorized processed sample or an antibacterial processed sample. Next, the petri dish in which the test piece is allowed to stand on the surface of the agar plate medium is held at a predetermined temperature and for a predetermined time to culture the test bacteria. With respect to the petri dish in which the test bacteria are cultured, the observer visually determines the size of the area (growth inhibition zone) where the growth of the test bacteria is blocked by the test piece.

  In addition, a fully automatic microbiological test apparatus that can perform drug sensitivity in a minimum of 3 hours by measuring absorbance is sold (for example, see Non-Patent Document 3). Hereinafter, this technique is referred to as a third conventional example.

JISZ2801: 2012 "Antimicrobial processed products-Antibacterial test method / Antimicrobial effect" JISL1902: 2008 “Testing method and antibacterial effect of textile products” “Lisa Suisse“ Nissui ”(registered trademark)”, [online], [searched February 18, 2014], Internet <URL: http://www.nissui-pharm.co.jp/diagnostics/raisus.html>

  Since the first and second conventional examples are visual determinations by an observer, the determination result is poor in objectivity, greatly depends on individual differences and skill levels of observers, and causes human error. There is. In addition, in order to accurately determine the antibacterial effect of the antibacterial agent, usually a large number of samples are required. However, in the first and second conventional examples, since an observer visually determines a large number of samples, enormous labor and time are required. Furthermore, in the halo test of the second conventional example, the observer visually determines the size of the petri dish growth inhibition zone, but it is qualitative and is the inhibition of the test bacteria by the test piece itself, In order to accurately determine whether or not the test bacteria are blocked by a substance released from the test piece (for example, phosphorus), the observer needs to be skilled.

  In the third conventional example, although it is automated, an expensive redox reagent or device is required.

  The present invention has been made in view of the above-described circumstances, and an example of an object is to solve the above-described problems. An antibacterial effect determination system and an antibacterial that can solve these problems An object is to provide an effect determination method and an antibacterial effect determination program.

  In order to solve the above-mentioned problem, the invention according to claim 1 is characterized in that the antibacterial agent is allowed to stand on the surface of a medium smeared with bacteria and the image of the surface of the medium after cultivating the bacteria is used. According to an antibacterial effect determination system for determining an antibacterial effect of an antibacterial agent, a local average image that is a feature amount of brightness of a local region of the image and a local dispersion image that is a feature amount of local flatness of the image Effective for determining the antibacterial effect of the antibacterial agent for each pixel of the image based on the feature extraction unit extracted from the image, the first prior probability, the local average image, and the local dispersion image An area determination unit including a first Bayes determination unit that determines a valid image area and an invalid image area by Bayesian estimation and outputs a binary image composed of the effective area and the invalid area; and the binary image. Blob analysis Forming a blob having a maximum area connecting regions and obtaining a occupancy ratio of the blob in the image and an eccentricity ratio of the blob; a second prior probability; the occupancy ratio and the eccentricity ratio And an effect determination unit having a second Bayes determination unit that determines the shape of the effective area of the image by Bayes estimation.

  The invention according to claim 2 relates to the antibacterial effect determination system according to claim 1, wherein the feature extraction unit includes a discrete Haar wavelet transform unit that performs discrete Haar wavelet transform on the image, and the discrete Haar. A discrete Haar wavelet inverse transform unit that outputs the local average image by performing discrete Haar wavelet inverse transform on the output low-frequency image of the wavelet transform unit, and subtracting the output image of the discrete Haar wavelet inverse transform unit from the image A subtracting unit that squares the output image of the subtracting unit, and an average filter that calculates a local average value of the output image of the squared unit and outputs the local dispersion image It is characterized by.

  The invention according to claim 3 relates to the antibacterial effect determination system according to claim 1 or 2, wherein the first prior probability is a two-dimensional Gaussian function having a peak value at a central coordinate point in the image. It is characterized by being modeled by.

  Further, the invention according to claim 4 is characterized in that the antibacterial effect of the antibacterial agent is obtained based on an image obtained by photographing the surface of the medium after culturing the bacteria by allowing the antibacterial agent to stand on the surface of the medium smeared with bacteria. According to a method for determining an antibacterial effect, a local average image that is a feature value of brightness of a local region of the image and a local dispersion image that is a feature value of local flatness of the image are extracted from the image. Effective area and invalidity for determining the antibacterial effect of the antibacterial agent for each pixel of the image based on the step 1, the first prior probability, and the local average image and the local variance image A second step of determining an invalid area by Bayesian estimation and outputting a binary image composed of the effective area and the invalid area, and a maximum area obtained by connecting the effective areas by blob analysis of the binary image Forming a blob with , Based on the third step of determining the occupancy ratio of the blob in the image and the eccentricity ratio of the blob, the second prior probability, and the occupancy ratio and the eccentricity ratio of the effective area of the image. And a fourth step of determining the shape by Bayesian estimation.

  The invention according to claim 5 relates to the antibacterial effect determination method according to claim 4, wherein the first step includes a first sub-step of performing discrete Haar wavelet transform on the image, and the discrete Haar. A second sub-step of performing discrete Haar wavelet inverse transformation on the wavelet transformed low-frequency image and outputting the local average image; and a third subtracting the discrete Haar wavelet inverse transformed image from the image. A sub-step, a fourth sub-step of squaring the image of the subtraction result, and a fifth sub-step of calculating a local average value of the squared image and outputting the local dispersion image It is characterized by including.

  The invention according to claim 6 relates to the antibacterial effect determination method according to claim 4 or 5, wherein the first prior probability is a two-dimensional Gaussian function having a peak value at a central coordinate point in the image. It is characterized by being modeled by.

  Further, the invention according to claim 7 is the antibacterial effect of the antibacterial agent based on an image obtained by photographing the surface of the medium after culturing the bacteria, by placing the antibacterial agent on the surface of the medium smeared with bacteria. According to an antibacterial effect determination program for determining, a local average image that is a feature amount of brightness of a local region of the image and a local dispersion image that is a feature amount of local flatness of the image are extracted from the image. Effective area and invalidity for determining the antibacterial effect of the antibacterial agent for each pixel of the image based on the step 1, the first prior probability, and the local average image and the local variance image A second step of determining an invalid area by Bayesian estimation and outputting a binary image composed of the effective area and the invalid area, and a maximum area obtained by connecting the effective areas by blob analysis of the binary image Blob with Forming and determining the occupancy in the image of the blob and the eccentricity of the blob, the second prior probability, the occupancy and the eccentricity based on the occupancy and the eccentricity. A fourth step of determining the shape of the region by Bayes estimation is executed by a computer.

  The invention according to claim 8 relates to the antibacterial effect determination program according to claim 7, wherein the first step includes a first sub-step of performing discrete Haar wavelet transform on the image, and the discrete Haar. A second sub-step of performing discrete Haar wavelet inverse transformation on the wavelet transformed low-frequency image and outputting the local average image; and a third subtracting the discrete Haar wavelet inverse transformed image from the image. A sub-step, a fourth sub-step of squaring the image of the subtraction result, and a fifth sub-step of calculating a local average value of the squared image and outputting the local dispersion image It is characterized by including.

  The invention according to claim 9 relates to the antibacterial effect determination program according to claim 7 or 8, wherein the first prior probability is a two-dimensional Gaussian function having a peak value at a central coordinate point in the image. It is characterized by being modeled by.

  According to the present invention, the antibacterial effect of an antibacterial agent can be determined automatically, quantitatively, and promptly equivalent to human visual discrimination with a simple and inexpensive configuration without using an expensive reagent or device. Can do. Thereby, development of an antibacterial agent can be made efficient.

It is a block diagram which shows the structure of the antibacterial effect determination system which concerns on the 1st Embodiment of this invention. It is a block diagram which shows the structure of the area | region determination part of FIG. It is a block diagram which shows the structure of the feature extraction part of FIG. It is a block diagram which shows the structure of the effect determination part of FIG. It is a block diagram for demonstrating the learning phase in an area | region determination part. It is a figure which shows an example of teacher data. It is a figure which shows an example of an observation image. An example of a local variance image sigma _x ² scatter to the local average image mu _x, an example of the distribution of local average image mu _x and is a diagram showing an example of the distribution of local variance image sigma _x ^2. It is a figure which shows an example of local average image (mu) _x . Is a diagram illustrating an example of a local variance image sigma _x ^2. It is a diagram illustrating an example of a prior probability P _{1 (p).} It is a figure which shows an example of likelihood function Β ₀ (μ ₀ ) Γ ₀ (σ _x ² ). It is a figure which shows an example of an area | region determination result. It is a block diagram for demonstrating the learning phase in an effect determination part. An example of application of the eccentricity e _x for occupancy a _x in the plurality of observation images, an example of the distribution of occupancy (area) a _x, a diagram illustrating an example of the distribution of the eccentricity e _x. It is a diagram illustrating an example of a likelihood function _{Β a1 (a x) Β e1} (e x). It is a figure which shows an example of the observation image in the example determined with the effect determination result being favorable. It is a figure which shows an example of the area | region determination effect in the example determined with the effect determination result being favorable. It is a figure which shows an example of the observation image in the example from which the effect determination result was determined to be effective. It is a figure which shows an example of the area | region determination effect in the example determined with the effect determination result having an effect. It is a figure which shows an example of the observation image in the example determined with the effect determination result having no effect. It is a figure which shows an example of the area | region determination effect in the example determined with the effect determination result having no effect. It is a figure which shows an example which matched the effect determination result by the visual observation of the expert with respect to the same sample, and the effect determination result of the antibacterial effect determination system which concerns on this 1st Embodiment. It is a block diagram which shows the structure of the antibacterial effect determination system which concerns on the 2nd Embodiment of this invention. It is a flowchart which shows the process which the antimicrobial effect determination system of FIG. 24 performs.

First Embodiment Hereinafter, a first embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing a configuration of an antibacterial effect determination system 1 according to the first embodiment of the present invention. The antibacterial effect determination system 1 includes an area determination unit 2 and an effect determination unit 3. The region determination unit 2 determines an effective region and an invalid region for determining the antibacterial effect of the antibacterial agent from the observation image, and supplies the region determination result to the effect determination unit 3. The effect determination unit 3 determines the antibacterial effect of the antibacterial agent based on the region determination result supplied from the region determination unit 2, and outputs the effect determination result.

  Here, an example of a method for producing an observation image will be described. First, the bacterial solution of the test bacteria is evenly smeared on the surface of the agar plate medium formed in the petri dish, and then the test piece is left on the surface of the agar plate medium. Next, the petri dish in which the test piece is allowed to stand on the surface of the agar plate medium is maintained at a temperature of 37 ° C. for 40 to 48 hours, for example, and the test bacteria are cultured. Then, the petri dish on which the test bacteria are cultured is photographed with, for example, a digital camera to obtain an observation image.

  The agar plate medium is prepared, for example, by the following procedure. First, for example, 1,000 ml of purified water, 5.0 g of meat extract, 10.0 g of peptone, 5.0 g of sodium chloride and 15.0 g of agar powder are put into a flask and mixed. Next, the flask is heated in a boiling water bath to sufficiently dissolve the contents mixed with purified water and each material, and then 0.1 mol / pH so that the pH becomes 7.0 ± 0.2 (25 ° C.). Prepare with sodium hydroxide solution. Next, the flask is stoppered with high-pressure steam sterilization with a culture stopper such as cotton, and then placed in a petri dish to be completely solidified.

  In the present embodiment, Escherichia coli (NBRC 3972) is used as the test bacterium. In the present embodiment, the test piece is formed by grinding an antibacterial agent, passing it through a sieve having a mesh size of 125 μm, and then molding it into a disk shape having a diameter of 13 mm and a thickness of 2 to 3 mm. As the antibacterial agent, for example, an antibacterial agent described in Japanese Patent No. 5282279 or an antibacterial agent produced by a manufacturing method described in Japanese Patent No. 5358770 may be used.

Next, the configuration of the area determination unit 2 will be described with reference to FIG. The area determination unit 2 includes a feature extraction unit 21 and a Bayes determination unit 22. In order to extract the feature of the observation image, the feature extraction unit 21 uses the local average image μ _x as the brightness feature amount of the local region of the observation image, and the local dispersion image σ as the feature amount of the local flatness of the observation image. _x ² is extracted, respectively.

Based on the prior probability P _k (p), the local average image μ _x and the local variance image σ _x ² supplied from the feature extraction unit 21, the Bayes determination unit 22 determines the antibacterial agent for each pixel of the observation image. An effective area and an ineffective area for determining the antibacterial effect are determined by Bayesian estimation, and an area determination result is output. In the present embodiment, the region determination result is output as a binary image in which the effective region is white and the invalid region is black.

Here, the determination by Bayesian estimation will be described. When there is an exhaustive event H ₁ and an event H ₂ and A is an arbitrary event, when event A occurs, the posterior probability P (H ₁ | A) that it is event H ₁ and event A The a posteriori probability P (H ₂ | A) that is an event H ₂ is expressed by equations (1) and (2), respectively, according to Bayes' theorem.

In equations (1) and (2), P (H ₁ ) is the prior probability that event H ₁ will occur, P (H ₂ ) is the prior probability that event H ₂ will occur, P (A | H ₁ ) and P (A | H ₂ ) is a likelihood function.

The ratio between the posterior probability P (H ₁ | A) and the posterior probability P (H ₂ | A) is expressed by Expression (3). If equation (3) is greater than _1, it can be determined that event H1 has occurred.

The Bayesian decision unit 22, event H ₁ valid regions to determine the antibacterial effect of the antibacterial agent, event H ₂ is an invalid region in order to determine the antimicrobial efficacy of the antimicrobial agent. On the other hand, event A is a combination of the coordinates of the pixel selected from the observed image, the local average value, and the local variance value. The prior probability P (H ₁ ) is the prior probability P ₁ (p) in which the event A is an effective region, and the prior probability P (H ₂ ) is the prior probability P ₀ (p) in which the event A is an invalid region. In the prior probabilities P ₁ (p) and P ₀ (p), the variable p means a two-dimensional vector indicating coordinates in the observation image. As will be described later, the likelihood function P (A | H ₁ ) is Β ₁ (μ ₁ ) Γ ₁ (σ _x ² ), and the likelihood function P (A | H ₂ ) is Β ₀ (μ ₀ ) Γ. ₀ (σ _x ² ). If Expression (3) is larger than 1, the pixel at the coordinate p is determined as an effective area.

In the present embodiment, the probability of becoming an effective region takes a peak value at the center coordinate point in the observation image, and performs prediction based on the subjectivity that the coordinate p decreases as the distance from the center coordinate point increases. According to this prediction, the prior probability P ₁ (p) is modeled by a two-dimensional Gaussian function having a peak value at the central coordinate point in the observed image. In the present embodiment, the value of the prior probability P ₁ (p) at the center coordinate point in the observation image is set to 0.8. By adopting prediction based on subjectivity in consideration of the above position coordinates, the determination accuracy between the effective area and the ineffective area for determining the antibacterial effect of the antibacterial agent compared to the case of calculating from only the pixel value. Can be improved.

  Next, the configuration of the feature extraction unit 21 will be described with reference to FIG. The feature extraction unit 21 includes a gray scale unit 31, a histogram equalization unit 32, a discrete Haar wavelet transform unit 33, a discrete Haar wavelet inverse transform unit 34, a subtraction unit 35, a square unit 36, And an average filter 37.

  The gray scale unit 31 converts the observation image into a gray scale and supplies it to the histogram equalization unit 32. In this gray scale conversion, numerical data constituting the observation image is converted into a real number representation (in this embodiment, double precision), and the luminance value is normalized to 0 to 1.

  Here, the data format of the observation image assumed in the present embodiment will be described. In this embodiment, the gray scale conversion unit 31 converts the observation image to gray scale with standard 256 gradations (8 bits), so that the observation image format can be converted to gray scale, Any of BMP, JPEG, GIF, TIF, PNG, etc. may be used.

  Next, the size of the observation image may be any size as long as most of the petri dish on which the test bacteria are cultured and the test piece is allowed to stand fit on the screen and is photographed at a uniform magnification. Furthermore, it is preferable that the resolution of the observed image is 1000 pixels in the vertical direction and 1000 pixels or more in the horizontal direction. The number of gradations of the observed image is preferably equal to or larger than 256 gradations (8 bits) because the grayscale unit 31 grayscales the observation image with 256 gradations (8 bits). .

  The gray scale method in the gray scale unit 31 is not particularly limited, and any of a simple averaging method, an intermediate value method, a G channel method, a weighted average method using NTSC coefficients, a weighted average using HDTV coefficients, and a correction method may be used. However, it is necessary for all observation images to be unified. In the present embodiment, when the observation image format is JPEG, the weighted average method using NTSC coefficients is used.

  The histogram equalization unit 32 reduces the density value interval in the range of density values with a large number of pixels and roughens the interval in a range with a small number of pixels in the image scaled in gray scale by the gray scale unit 31. The contrast is adjusted and then supplied to the discrete Haar wavelet transform unit 33.

The histogram equalization unit 32 cultivates the test bacteria that is the subject of the digital camera, and according to the difference in the intensity of illumination applied to the petri dish on which the test piece is stationary, the feature amount, that is, the local average image This is used for suppressing variations in μ _x and local dispersion image σ _x ² . Therefore, when the observed image is taken with a constant illumination intensity, the histogram equalization may be omitted.

  The discrete Haar wavelet transform unit 33 performs low frequency discrete Haar wavelet transform on the image data supplied from the histogram equalization unit 32 and then supplies the image data to the discrete Haar wavelet inverse transform unit 34. By performing low frequency discrete Haar wavelet transform on the output image data of the histogram equalization unit 32, the image data is locally flat subband components (LL) and other subband components (LH, HL, HH). It is divided into.

Here, the discrete Haar wavelet transform will be described. First, an array of image data supplied from the histogram equalization unit 32 is X, and a two-dimensional Z transformation of the array X is X (z _x , z _y ). For the two-dimensional Z-transform X (z _x , z _y ), a linear low-pass filtering process using a two-dimensional transfer function H (z _x , z _y ) expressed by the equation (4) and a two-dimensional thinning process (down) The result of the sequential application of (sampling processing) is expressed by equation (5). In the two-dimensional thinning process, the horizontal thinning rate and the vertical thinning rate are both 2.

  Thereafter, the linear low-pass filtering process and the two-dimensional thinning process are repeated (m−1) times, and the obtained result is expressed by Expression (6). Note that any number of repetitions m can be selected. In the present embodiment, the number of repetitions m is empirically set to 3 times.

The discrete Haar wavelet inverse transform unit 34 performs the discrete Haar wavelet inverse transform on the low-frequency image data supplied from the discrete Haar wavelet transform unit 33. Since the output image data of the discrete Haar wavelet transform unit 33 is only a locally flat subband component (LL), image data obtained by performing an inverse discrete Haar wavelet transform on the subband component (LL) is This represents the local average value of the brightness of the observed image. Therefore, the output image data of the discrete Haar wavelet inverse transform unit 34 can be used as a local average image μ _x as a feature value of the brightness of the local region of each pixel constituting the observation image.

Here, the discrete Haar wavelet inverse transform will be described. A result obtained by sequentially performing two-dimensional zero-value insertion processing (upsampling processing) and linear low-pass filtering processing on Y ^{m} (z _x , z _y ) represented by Equation (6) 7). In the two-dimensional zero value insertion process, both the horizontal direction interpolation rate and the vertical direction interpolation rate are set to 2.

  Thereafter, the two-dimensional zero value insertion process and the linear filtering process are repeated (m−1) times, and the obtained result is expressed by Expression (8).

A two-dimensional inverse Z transform of the image array of M (z _x , z _y ) represented by Expression (8) is output from the discrete Haar wavelet inverse transform unit 34 as a local average image μ _x .

The subtracting unit 35 subtracts the output image data of the discrete Haar wavelet inverse transform unit 34 from the output image data of the histogram equalizing unit 32. The output image data of the discrete Haar wavelet inverse transform unit 34 is a local average image μ _x composed of only locally flat subband components (LL). Therefore, the output image data of the subtraction unit 35 can be said to be equivalent to subband components (LH, HL, HH) other than the subband component (LL) of the output image data of the histogram equalization unit 32. The output image data of the subtraction unit 35 represents a local deviation (contrast) of the observation image.

The squaring unit 36 squares the output image data of the subtracting unit 35. The average filter 37 calculates a local average value of the output image data of the square unit 36. In the present embodiment, a moving average filter is used as the average filter 37. Since the output image data of the subtracting unit 35 represents a deviation from the local average value of each pixel constituting the observation image, the squared unit 36 squares this, and the average filter 37 then compares the values with the surrounding values. By calculating the average value, the local dispersion image σ _x ² can be obtained. The local dispersion image σ _x ² can be used as a feature amount of local flatness of each pixel constituting the observation image.

Here, processing of the subtracting unit 35, the square unit 36, and the average filter 37 will be described. Similarly to the above, when the arrangement of the image data supplied from the histogram equalization unit 32 is X, the output image data of the subtraction unit 35 is an arrangement (X−μ _x ). Next, the squaring unit 36 takes the square of each element of the array (X−μ _x ). Therefore, the output image data of the squaring unit 36 is expressed by Expression (9) as an array Q obtained by taking the square of each element of the array (X−μ _x ).

The average filter 37 outputs a local dispersion image σ _x ² as a result of calculating the local average value of the array Q. That is, the two-dimensional Z-transform S (z _x , z _y ) of the array σ _x ² of the local dispersion image is expressed by Expression (10).

In Expression (10), Q (z _x , z _y ) is a two-dimensional Z transformation of the array Q, and A (z _x , z _y ) is a two-dimensional transfer function of the average filter 37. The size of the moving average filter constituting the average filter 37 may be arbitrary. In this embodiment, the size of the moving average filter is empirically set to 5 × 5. That is, the two-dimensional transfer function A (z _x , z _y ) is expressed by Expression (11).

  Next, the configuration of the effect determination unit 3 will be described with reference to FIG. The effect determination unit 3 includes a morphological closing unit 41, a blob analysis unit 42, and a Bayes determination unit 43. The morphological closing unit 41 is supplied from the region determination unit 2, and in the region determination result that determines the effective region and the invalid region for determining the antibacterial effect of the antibacterial agent for each pixel of the observation image, the effective region The shape of the effective area is adjusted by filling a small hole that does not fit in the structural element.

  The blob analysis unit 42 first divides each pixel constituting the output image data of the morphological closing unit 41 into an object pixel and a background pixel.

  Next, the blob analysis unit 42 performs connected blob analysis to form object pixels into a group (blob) of connected object pixels. In the present embodiment, it is assumed that the effective region for determining the antibacterial effect of the antibacterial agent for each pixel of the observation image is one connected region, and only the blob having the maximum area in the observation image is analyzed. ing.

The blob analysis unit 42, connected blob analysis result, and outputs the occupancy (area) _{a x} of the formula (12), and eccentricity _{e x} of the formula (13).

In the formula (12), n _b is the number of pixels blob (effective area) having the largest area in the observation image, N is the number of all the pixels constituting the observed image.

In the equation (13), the eccentricity e _x is the eccentricity of an ellipse having the same second moment as the blob (effective region), a is 1/2 of the major axis of the ellipse, and b is 1/2 of the minor axis of the ellipse. It is.

Based on the prior probability P _k , the occupation ratio (area) a _x and the eccentricity e _x supplied from the blob analysis unit 42, the Bayes determination unit 43 determines the shape of the effective region of the observation image by Bayes estimation. Outputs the result of judging the antibacterial effect of the antibacterial agent. In the present embodiment, the effect determination of the antibacterial agent is performed based on the shape data of the effective region in the region determination result, and the Bayes determination unit 43 is determined as one of good (◯), effective (Δ), and no effect (×). Is output from.

The Bayesian decision unit 43, in the above formula (1) to (3), that event H ₁ have a antibacterial effect of an antibacterial agent, event H ₂ is the lack of antimicrobial effect of an antimicrobial agent. On the other hand, the event A is a combination of the occupation ratio (area) a _x and the eccentricity e _x obtained from the observation image. Further, the prior probability P (H ₁ ) is the prior probability P ₁ where the antibacterial effect of the antibacterial agent is the event A, and the prior probability P (H ₂ ) is the prior probability P ₀ where the antibacterial effect of the antibacterial agent is the event A. As described later, the likelihood function P (A | _{H 1)} is _{Β a1 (a x) Β e1} (e x), the likelihood function P (A | _{H 2)} is Β _{a0 (a x)} Β _e0 a _{(e x).} If the expression (3) is larger than 1, it is determined that the antibacterial effect of the antibacterial agent is present in the observed image.

In the present embodiment, both the prior probability P ₁ and the prior probability P ₀ are set to 0.5. This is to make sure that there is no bias in either good (◯) or no effect (×) when determining the antibacterial effect of the antibacterial agent. In effect, this means that the antibacterial effect of the antibacterial agent is determined using only the feature amount, and the subjectivity is removed. However, it is possible to improve the determination performance of the antibacterial effect by adjusting the prior probability P ₁ and the prior probability P ₀ as parameters. For example, when the determination with the effect is excessively performed, the value of the prior probability P ₁ is decreased, and conversely, when the determination with no effect is excessively performed, the value of the prior probability P ₀ is decreased, thereby reducing the antibacterial effect. The determination accuracy can be adjusted.

Next, the operation of the antibacterial effect determination system having the above configuration will be described with reference to the drawings. First, the learning phase in the region determination unit 2 will be described with reference to FIGS. FIG. 5 is a block diagram for explaining a learning phase in the region determination unit 2. In this learning phase, estimation parameters {a _μ0 , b _μ0 }, {a _μ1 , b _μ1 }, {aσ ² ₀ , bσ ² ₀ } and {aσ ² ₁ , bσ ² are based on the teacher data and the observed image. ₁ } is obtained.

  The teacher data is an 8-bit grayscale image. The teacher data is composed of a value 0 indicating an invalid area, a value 255 indicating an effective area, and a value 128 indicating Don't care. The size of the teacher data is the same as the size of the observation image. The teacher data format may be any format such as BMP, JPEG, GIF, TIF, and PNG as long as it is grayscaled. In this embodiment, TIF is adopted as the format of the teacher data. FIG. 6 shows an example of the teacher data, and FIG. 7 shows an example of the observation image.

The estimation parameters {a _μ0 , b _μ0 } and {a _μ1 , b _μ1 } are the beta functions _{ｋ k} (μ that constitute the likelihood function _{ｋ k} (μ _k ) Γ _k (σ _x ² ) used in the determination phase described later. _k ). On the other hand, the estimated parameters {aσ ² ₀ , bσ ² ₀ } and {aσ ² ₁ , bσ ² ₁ } are represented by the gamma function Γ _k (σ, which constitutes the likelihood function _{ｋ k} (μ _k ) Γ _k (σ _x ² ). _x ² ) estimation parameter. Therefore, Bayesian decision unit 22, in the learning phase, the beta distribution maximum likelihood estimation unit 22 _1, gamma distribution is constituted by the maximum likelihood estimation unit 22 _2.

In the learning phase, a random extraction unit 23 is inserted between the feature extraction unit 21 and the Bayes determination unit 22. In the learning phase, the Bayesian determination unit 22 determines, based on the teacher data, a valid region and an invalid region for determining the antibacterial effect of the antibacterial agent for each pixel of the observation image by Bayesian estimation, and feature extraction Statistical properties of the local average image μ _x and the local variance image σ _x ² supplied from the unit 21 are analyzed. In the learning phase, processing all the pixels of the observation image requires a huge amount of memory and a long time. Therefore, in order to easily calculate the statistics of each of the effective region and the invalid region, a random extraction unit 23 is provided, and necessary data is randomly extracted from the local average image μ _x and the local variance image σ _x ^2. Yes.

Beta distribution maximum likelihood estimation unit 22 _1, and the teacher data, based on the local average image mu _x supplied from the random extracting unit 23 performs a beta distribution maximum likelihood estimation, beta function beta _k of _{(mu k)} Estimate parameters {a _μ0 , b _μ0 } and {a _μ1 , b _μ1 } are obtained. Since the local average image μ _x is obtained from an image in which the luminance value is normalized between 0 and 1 in the gray scale unit 31, the luminance value is within the range of 0 to 1 as well. On the other hand, the beta distribution is a probability model typically employed as a probability density function of a random variable whose value is restricted between 0 and 1. Therefore, in this embodiment, the beta distribution is adopted, but a gamma distribution may be adopted.

The gamma distribution maximum likelihood estimation unit 22 ₂ performs gamma distribution maximum likelihood estimation based on the teacher data and the local variance image σ _x ² supplied from the random extraction unit 23, and generates a gamma function Γ _k (σ ² _x ) Estimation parameters {aσ ² ₀ , bσ ² ₀ } and {aσ ² ₁ , bσ ² ₁ }. The local dispersion image σ _x ² is guaranteed to have a positive luminance value. On the other hand, the gamma distribution is a probability model typically employed as a probability density function of a random variable having only positive values. Therefore, in this embodiment, a gamma distribution is adopted. Since the local dispersion image σ _x ² has a luminance value in the range of 0 to 1 for the same reason as the local average image μ _x , a beta distribution may be adopted.

Here, FIG. 8 shows an example of the distribution of the local dispersion image σ _x ² with respect to the local average image μ _x , an example of the distribution (histogram) of the local average image μ _x , and an example of the distribution (histogram) of the local dispersion image σ _x ^2. Indicates. A curve drawn with a solid line corresponds to the invalid area, and a curve drawn with a dotted line corresponds to the valid area.

  Next, the determination phase in the region determination unit 2 will be described with reference to FIGS. 2, 3, 7 and 9 to 13. In this determination phase, for example, the observed image shown in FIG. 7 is grayscaled by the grayscaler 31 shown in FIG.

Next, the output image data of the histogram equalization unit 32 is subjected to discrete Haar wavelet transform by the discrete Haar wavelet transform unit 33 and then subjected to discrete Haar wavelet inverse transform by the discrete Haar wavelet inverse transform unit 34. Thereby, the output image data of the discrete Haar wavelet inverse transform unit 34 is supplied to the Bayes determination unit 22 shown in FIG. 2 as a local average image μ _x . FIG. 9 is an example of the local average image μ _x .

On the other hand, the output image data of the histogram equalization unit 32 is subtracted by the subtraction unit 35 from the output image data of the discrete Haar wavelet inverse transformation unit 34, that is, the local average image μ _x . Next, the output image data of the subtraction unit 35 is squared by the squaring unit 36 and then passes through the average filter 37. Thereby, the average filter 37 supplies the output image data to the Bayes determination unit 22 shown in FIG. 2 as the local dispersion image σ _x ² . FIG. 10 is an example of the local dispersion image σ _x ² .

Based on the prior probability P _k (p), the local average image μ _x and the local variance image σ _x ² supplied from the feature extraction unit 21, the Bayes determination unit 22 determines the antibacterial agent for each pixel of the observation image. An effective area and an ineffective area for determining the antibacterial effect are determined by Bayesian estimation, and an area determination result is output. FIG. 11 is an example of the prior probability P ₁ (p).

The estimation parameters {a _μ0 , b _μ0 } and {a _μ1 , b _μ1 } obtained in the learning phase are assigned to the beta functions Β ₀ (μ ₀ ) and Β ₁ (μ ₁ ), respectively. On the other hand, the estimated parameters {aσ ² ₀ , bσ ² ₀ } and {aσ ² ₁ , bσ ² ₁ } obtained in the learning phase are converted into gamma functions Γ ₀ (σ _x ² ) and Γ ₁ (σ _x ² ). Assigned respectively.

In this determination phase, in equation (3), prior probability P (H ₁ ) is prior probability P ₁ (p), prior probability P (H ₂ ) is prior probability P ₀ (p), and likelihood function P (A | H ₁ ) is a likelihood function Β ₁ (μ ₁ ) Γ ₁ (σ _x ² ), and the likelihood function P (A | H ₂ ) is a likelihood function Β ₀ (μ ₀ ) Γ ₀ (σ _x ² ). . FIG. 12 is an example of the likelihood function Β ₀ (μ ₀ ) Γ ₀ (σ _x ² ).

The prior probability P ₁ (p), prior probability P ₀ (p), likelihood function Β ₁ (μ ₁ ) Γ ₁ (σ _x ² ) and likelihood function Β ₀ (μ ₀ ) Γ ₀ ( Substituting (σ _x ² ) for transformation yields equation (14).

  The left and right sides of Expression (14) are compared, and if the left side of Expression (14) is large, the pixel of the coordinate p is determined as an effective area, while if the right side of Expression (14) is large, the pixel of the coordinate p is determined. Are determined to be invalid areas. FIG. 13 shows an example of the region determination result.

Next, the learning phase in the effect determination unit 3 will be described with reference to FIGS. 14 and 15. FIG. 14 is a block diagram for explaining a learning phase in the effect determination unit 3. In this learning phase, estimation parameters {a _a0 , b _a0 }, {a _a1 , b _a1 }, {a _e0 , b _e0 } and {a _e1 , b _e1 } are set based on the teacher data and the region determination result. Ask.

The teacher data was obtained by performing blob analysis of the effective area in the blob analysis unit 42 on the image for which the antibacterial effect of the antibacterial agent is already judged as good (◯) or not effective (×). This is numerical data in which the occupation ratio (area) a _x and the eccentricity e _x are stored as a two-dimensional vector.

The estimated parameters {a _a0 , b _a0 } and {a _a1 , b _a1 } are the beta functions Β _ak (a _x ) that constitute the likelihood function Β _ak (a _x ) Β _ek (e _x ) used in the determination phase described later. ) Estimation parameter. On the other hand, the estimation parameters {a _e0 , b _e0 } and {a _e1 , b _e1 } are the estimation parameters of the beta function Β _ek (e _x ) that constitutes the likelihood function （ _ak (a _x ) Β _ek (e _x ). It is. Therefore, Bayesian decision unit 43, in the learning phase, the beta distribution maximum likelihood estimation unit 43 ₁ is constituted by a Beta distribution maximum likelihood estimation unit 43 _2.

Beta distribution maximum likelihood estimation unit 43 _1, and the teacher data, based on the occupancy supplied (area) a _x from blob analysis unit 42 performs a beta distribution maximum likelihood estimation, beta function beta _{ak (a x} ) Estimation parameters {a _a0 , b _a0 } and {a _a1 , b _a1 }. Since the occupancy (area) a _x is obtained from the image in which the luminance value is normalized between 0 and 1 in the blob analysis unit 42, the luminance value is also in the range of 0 to 1. . On the other hand, the beta distribution is a probability model typically employed as a probability density function of a random variable whose value is restricted between 0 and 1. Therefore, in this embodiment, the beta distribution is adopted, but a gamma distribution may be adopted.

Beta distribution maximum likelihood estimation unit 43 _2, and the teacher data, based on the supplied eccentricity e _x from blob analysis unit 42 performs a beta distribution maximum likelihood estimation, estimation of beta function beta _{ek (e x)} Parameters {a _e0 , b _e0 } and {a _e1 , b _e1 } are obtained. Since the eccentricity e _x is obtained from an image in which the luminance value is normalized between 0 and 1 in the blob analysis unit 42, the luminance value is also in the range of 0 to 1. On the other hand, the beta distribution is a probability model typically employed as a probability density function of a random variable whose value is restricted between 0 and 1. Therefore, in this embodiment, the beta distribution is adopted, but a gamma distribution may be adopted.

Here, FIG. 15 shows an example of the distribution of the eccentricity e _x with respect to the occupation ratio (area) a _x in a plurality of observation images, an example of the distribution (histogram) of the occupation ratio (area) a _x , and the distribution of the eccentricity e _x . An example of (histogram) is shown. The curve drawn with a solid line corresponds to an observation image where the antibacterial effect of the antibacterial agent is good (◯), and the curve drawn with a dotted line corresponds to an observation image where the antibacterial effect of the antibacterial agent is not present (×).

Next, the determination phase in the effect determination unit 3 will be described with reference to FIGS. 4, 13, and 16. In this determination phase, for example, the region determination result shown in FIG. 13 is obtained after the morphological closing unit 41 shown in FIG. The blob analysis unit 42 analyzes the connected blob. Accordingly, the blob analysis unit 42 outputs the occupation ratio (area) a _x and the eccentricity e _x .

Next, the Bayesian determination unit 43 performs Bayesian estimation on the shape of the effective region of the observation image based on the prior probability P _k , the occupation ratio (area) a _x and the eccentricity e _x supplied from the blob analysis unit 42. The effect determination result which determined the antibacterial effect of an antibacterial agent is output.

The estimation parameters {a _a0 , b _a0 } and {a _a1 , b _a1 } obtained in the learning phase are assigned to the beta functions ａ _a0 (a _x ) and Β _a1 (a _x ), respectively. On the other hand, the estimated parameters {a _e0 , b _e0 } and {a _e1 , b _e1 } obtained in the learning phase are assigned to the beta functions ｅ _e0 (e _x ) and Β _e1 (e _x ), respectively.

In this determination phase, in Equation (3), prior probability P (H ₁ ) is prior probability P ₁ , prior probability P (H ₂ ) is prior probability P ₀ , and likelihood function P (A | H ₁ ) is likelihood. function _{Β a1 (a x) Β e1} (e x), the likelihood function P (a | _{H 2)} is the likelihood function _{Β a0 (a x) Β e0} (e x). Figure 16 is an example of a likelihood function _{Β a1 (a x) Β e1} (e x).

Priori probability _{P 1} in formula (3), the prior probability _{P 0,} by substituting the likelihood function _{Β a1 (a x) Β e1} (e x) and the likelihood function _{Β a0 (a x) Β e0} (e x) When deformed, equation (15) is obtained.

  The left and right sides of Expression (15) are compared, and if the left side of Expression (15) is large to some extent, it is determined that the antibacterial effect of the antibacterial agent is good (O) in the observed image, while the right side of Expression (15) is If it is large to some extent, it is determined that there is no antibacterial effect of the antibacterial agent in the observed image (×).

  Note that if no significant difference is found between the left side and the right side of Equation (15), the determination is rejected. In the present embodiment, when a difference of 0.2 or more is not seen, it is rejected. When the determination is rejected, it is determined that there is an effect (Δ). The value of whether or not to reject can be adjusted.

  Here, an example of the effect determination result of the antibacterial effect determination system 1 according to the present embodiment will be described with reference to FIGS. FIG. 17 is a diagram illustrating an example of an observed image in an example in which the effect determination result is determined to be good (◯), and FIG. 18 is a diagram illustrating an example of an area determination effect in an example in which the effect determination result is determined to be good (◯). It is. FIG. 19 is a diagram illustrating an example of an observation image in an example in which the effect determination result is determined to be effective (Δ), and FIG. 20 illustrates an area determination effect in an example in which the effect determination result is determined to be effective (Δ). It is a figure which shows an example. On the other hand, FIG. 21 is a diagram illustrating an example of an observation image in an example in which the effect determination result is determined to have no effect (×), and FIG. 22 illustrates an area determination effect in an example in which the effect determination result is determined to have no effect (×). It is a figure which shows an example. From FIG. 17 to FIG. 22, it can be confirmed that the antibacterial effect determination system 1 according to the present embodiment is operating properly.

  FIG. 23 shows an example in which an expert's visual effect determination result for the same sample is associated with the effect determination result of the antibacterial effect determination system 1 according to the present embodiment. From FIG. 23, the antibacterial effect determination system 1 according to the present embodiment can confirm that an effect determination result almost equivalent to that of a human being can be obtained.

As described above, according to the first embodiment of the present invention, it is possible to automatically and quantitatively and quickly make a determination equivalent to the determination by human visual observation. Thereby, development of an antibacterial agent can be made efficient. Further, according to the first embodiment of the present invention, since the automatic determination using image processing is performed, the determination is quickly performed by a simpler method than the visual determination by the observer as in the first and second conventional examples. It can be carried out. Further, since the determination is made using the grayscale image, it is not necessary to stain the inspection object as in the third conventional example.
According to Embodiment 1 of the present invention, the antibacterial effect of an antibacterial agent can be determined with a simple and inexpensive configuration without using an expensive redox coloring reagent or apparatus as in the third conventional example. it can.

  Furthermore, according to Embodiment 1 of the present invention, it can be mechanically determined that the test bacteria directly under the test piece are killed and the growth inhibitory zone is not diffused around the concentric circles of the test piece. Therefore, it is accurately determined whether the test is qualitative as in the second conventional example of hello test, or the test bacteria are blocked by the test piece itself, or the test bacteria are blocked by a substance released from the test piece. Therefore, the problem that the observer requires skill is unlikely to occur.

Second Embodiment Next, a second embodiment of the present invention will be described with reference to the drawings. In addition, about the structure same or equivalent to 1st Embodiment, while using an equivalent code | symbol, the description is abbreviate | omitted or simplified. The antibacterial effect determination system 51 according to the present embodiment is different from the antibacterial effect determination system 1 according to the first embodiment in that the antibacterial effect determination system 51 according to the present embodiment is realized by the same configuration as a general computer or a device such as a microcomputer. ing.

  FIG. 24 is a block diagram showing a configuration of the antibacterial effect determination system 51. As shown in FIG. 24, the antibacterial effect determination system 51 includes a CPU (Central Processing Unit) 61, a main storage unit 62, an auxiliary storage unit 63, a display unit 64, an input unit 65, and an interface unit 66. It is roughly structured.

  The CPU 61 executes processing to be described later on the input observation image according to the program stored in the auxiliary storage unit 63. The main storage unit 62 includes a RAM (Random Access Memory) and the like, and is used as a work area for the CPU 61.

The auxiliary storage unit 63 is a semiconductor memory such as ROM or RAM, an HD drive to which an HD (hard disk) is mounted, a CD-ROM (Compact Disk Read-Only Memory), a CD-R (Recordable), or a CD-RW. (ReWritable), a DVD (Digital Versatile Disk) -ROM, a DVD-R, a DVD-RW, a CD / DVD drive or the like to be mounted. The auxiliary storage unit 63 stores programs executed by the CPU 61 and various parameters. In addition, information regarding observation images input from the outside and information including processing results by the CPU 61 are sequentially stored. Examples of data stored in the auxiliary storage unit 63 include teacher data used in steps S1 and S2, which will be described later, prior probabilities P _k (p) and P _k , a local average image μ _x , a local variance image σ _x ^{2, and the} like. There is.

  The display unit 64 includes a liquid crystal display, an organic EL display, a plasma display, and the like, and displays a processing result of the CPU 61 and the like. The input unit 65 includes a pointing device such as a keyboard and a mouse. An operator's instruction is input via the input unit 65 and notified to the CPU 61. The interface unit 66 includes a serial interface or a LAN (Local Area Network) interface.

  The flowchart shown in FIG. 25 corresponds to a series of processing algorithms of a program executed by the CPU 61. Hereinafter, processing executed by the antibacterial effect determination system 51 having the above configuration will be described with reference to the flowchart shown in FIG.

The CPU 61 preliminarily estimates parameters {a _μ0 , b _μ0 }, {a _μ1 , b _μ1 }, {aσ ² ₀ , bσ ² ₀ } and {aσ ² ₁ , bσ ² based on the teacher data and the observed image. ₁ } region determination learning processing is performed in advance. That is, the CPU 61 executes a learning phase process performed by the feature unit 21, the random extraction unit 23, and the Bayes determination unit 22 described with reference to FIGS. The estimated parameters {a _μ0 , b _μ0 } and {a _μ1 , b _μ1 } are assigned to the beta functions Β ₀ (μ ₀ ) and Β ₁ (μ ₁ ), respectively, and the estimated parameters {aσ ² ₀ , bσ ² ₀ } and {Aσ ² ₁ , bσ ² ₁ } is assigned to the gamma functions Γ ₀ (σ _x ² ) and Γ ₁ (σ _x ² ), respectively.

The CPU 61 preliminarily determines the estimation parameters {a _a0 , b _a0 }, {a _a1 , b _a1 }, {a _e0 , b _e0 } and {a _e1 , b _e1 based on the teacher data and the region determination result. } Is executed in advance to determine the area. That is, the CPU 61 executes a learning phase process performed by the morphological closing unit 41, the blob analysis unit 42, and the Bayes determination unit 43 described with reference to FIGS. The estimated parameters {a _a0 , b _a0 } and {a _a1 , b _a1 } are assigned to the beta functions Β _a0 (a _x ) and Β _a1 (a _x ), respectively, and the estimated parameters {a _e0 , b _e0 } and {a _{e1, b e1}} is substituted into each of the beta function Β _{e0 (e x)} and Β _{e1 (e x).}

  In step S1, the CPU 61 executes gray scale / contrast adjustment processing on the input observation image. That is, the CPU 61 executes the processing performed by the gray scale conversion unit 31 and the histogram equalization unit 32 described with reference to FIGS.

In step S2, the CPU 61 performs a feature amount extraction process on the image obtained in step S1, and extracts a local average image μ _x and a local variance image σ _x ² that are feature amounts. That is, the CPU 61 is performed by the discrete Haar wavelet transform unit 33, the discrete Haar wavelet inverse transform unit 34, the subtraction unit 35, the square unit 36, and the average filter 37 described with reference to FIGS. Execute the process.

In step S3, the CPU 61 executes region determination processing based on the prior probability P _k (p), the local average image μ _x, and the local variance image σ _x ² to obtain an image as a region determination result. That is, the CPU 61 executes processing performed by the Bayesian determination unit 22 described with reference to FIGS. 11 to 13 and the expressions (3) and (14).

  In step S4, the CPU 61 executes an effective area shaping process on the image obtained in step S3. That is, the CPU 61 executes processing performed by the morphological closing unit 41 described with reference to FIGS. 4 and 13.

In step S5, the CPU 61 performs a feature amount extraction process on the image obtained in step S4, and extracts an occupancy (area) a _x and an eccentricity e _x that are feature amounts. That is, the CPU 61 executes the process performed by the blob analysis unit 42 already described.

In step S6, CPU 61 has a prior probability _{P k,} on the basis of the occupation ratio (area) _{a x} and eccentricity _{e x,} executes the effect determination process, and outputs the effect determination result. That is, the CPU 61 executes processing performed by the Bayesian determination unit 43 described with reference to FIG. 16 and Equations (3) and (15).

  As described above, according to the second embodiment, the effects obtained in the first embodiment can be obtained, and an expensive redox reagent or apparatus such as the third conventional example can be provided. Without using it, the antibacterial effect of the antibacterial agent can be determined with a simple and inexpensive configuration using a general computer or a microcomputer.

  As described above, the embodiments of the present invention have been described in detail with reference to the drawings. However, the specific configuration is not limited to these embodiments, and the design can be changed without departing from the scope of the present invention. Is included in the present invention.

  For example, in the second embodiment, the program stored in the auxiliary storage unit 63 of the antibacterial effect determination system 51 is stored and distributed in a computer-readable recording medium such as a CD-ROM or DVD-ROM, An apparatus that executes the above-described processing may be configured by installing the program in a computer.

  Further, the program may be stored in a disk device or the like included in a predetermined server device on a communication network such as the Internet, and may be downloaded onto a computer by being superimposed on a carrier wave, for example. The program may be activated and executed while being transferred via a communication network. Furthermore, the program may be executed entirely or partially on the server device, and the above-described image processing may be executed while transmitting / receiving information related to the processing via a communication network.

  Moreover, in each above-mentioned embodiment, although the example which determines the antibacterial effect of an antibacterial agent based on only the shape of an effective area | region was shown, it is not limited to this. For example, the transparency of the effective area and the blurring degree of the outline of the effective area may be digitized, and the antibacterial effect of the antibacterial agent may be determined based on these values and the shape of the effective area. If comprised in this way, it can determine more correctly.

(I) Transparency When the petri dish on which the test bacteria are cultured is photographed with a digital camera, it is assumed that the stage on which the petri dish is placed is plain black. After the effective area is determined, the average luminance value μ _e of the effective area is calculated. Assuming that the luminance value of the effective area is normalized to a value of 0 to 1, the transparency T expressed by the equation (16) is calculated. In the ineffective region, the white turbidity progresses due to the propagation of the test bacteria, so the value of the transparency T approaches 0.
T = 1-μ _e (16)

(Ii) the contour blurring of using the area A and the eccentricity e _x of the effective area obtained by connecting blob analysis of degree effective area defines a degree B blurring of outline of the formula (17).
B = constant × e _x / A (17)
In the present embodiment, since the shape of the effective area is basically a circle, the shape is distorted from the circle when the blur occurs, so the blur condition B has a large value. Further, even if the area A of the effective area is reduced due to the blur, the blur condition B is a large value.

DESCRIPTION OF SYMBOLS 1 ... Antibacterial effect determination system, 2 ... Area | region determination part, 3 ... Effect determination part, 21 ... Feature extraction part, 22 ... Bayes determination part, 22 ₁ ... Beta distribution maximum likelihood estimation part, 22 ₂ ... Gamma distribution maximum likelihood estimation part , 23 ... Random extraction unit, 31 ... Gray scale conversion unit, 32 ... Histogram equalization unit, 33 ... Discrete Haar wavelet transform unit, 34 ... Discrete Haar wavelet inverse transform unit, 35 ... Subtraction unit, 36 ... Square unit , 37 ... mean filter, 41 ... morphological closing portion, 42 ... blob analysis unit, 43 ... Bayesian decision unit, ₄₃ 1, 43 2 _... beta distribution maximum likelihood estimation unit, 51 ... antimicrobial effect determination system, 61 ... CPU, 62 ... main storage unit, 63 ... auxiliary storage unit, 64 ... display unit, 65 ... input unit, 66 ... interface unit

Claims (9)

An antibacterial effect determination system for determining an antibacterial effect of the antibacterial agent based on an image obtained by photographing the surface of the medium after cultivating the bacteria, leaving the antibacterial agent on the surface of the medium smeared with bacteria,
A feature extraction unit that extracts from the image a local average image that is a feature value of brightness of a local region of the image and a local dispersion image that is a feature value of local flatness of the image; and a first prior probability And, based on the local average image and the local variance image, for each pixel of the image, a valid area and an invalid invalid area for determining the antibacterial effect of the antibacterial agent are determined by Bayesian estimation. An area determination unit having a first Bayes determination unit that outputs a binary image composed of the effective area and the invalid area;
Blob analysis of the binary image to form a blob having a maximum area connecting the effective areas, and a blob analysis unit for obtaining an occupancy rate of the blob in the image and an eccentricity rate of the blob; An effect determination unit including a second Bayes determination unit that determines the shape of the effective area of the image by Bayes estimation based on the prior probability, the occupation rate, and the eccentricity. Antibacterial effect judgment system. The feature extraction unit includes:
A discrete Haar wavelet transform unit for transforming the image into a discrete Haar wavelet transform;
A discrete Haar wavelet inverse transform unit that performs discrete Haar wavelet inverse transform on the output low-frequency image of the discrete Haar wavelet transform unit and outputs the local average image;
A subtractor that subtracts the output image of the discrete Haar wavelet inverse transform unit from the image;
A squaring unit that squares the output image of the subtraction unit;
The antibacterial effect determination system according to claim 1, further comprising: an average filter that calculates a local average value of the output image of the square part and outputs the local dispersion image.   The antibacterial effect determination system according to claim 1 or 2, wherein the first prior probability is modeled by a two-dimensional Gaussian function having a peak value at a central coordinate point in the image. An antibacterial effect determination method for determining an antibacterial effect of the antibacterial agent based on an image obtained by photographing the surface of the medium after culturing the bacteria, leaving the antibacterial agent on the surface of the medium smeared with bacteria,
A first step of extracting, from the image, a local average image that is a feature amount of brightness of a local region of the image and a local dispersion image that is a feature amount of local flatness of the image;
Based on the first prior probability and the local average image and the local dispersion image, an effective area and an invalid invalid area that are effective for determining the antibacterial effect of the antibacterial agent for each pixel of the image. A second step of determining by Bayesian estimation and outputting a binary image composed of the effective area and the invalid area;
Blob analysis of the binary image to form a blob having a maximum area concatenating the effective areas, and determining a occupancy ratio of the blob in the image and an eccentricity ratio of the blob;
An antibacterial effect determination method comprising: a fourth step of determining a shape of an effective area of the image by Bayes estimation based on a second prior probability and the occupation ratio and the eccentricity. The first step includes
A first sub-step of discrete Haar wavelet transform of the image;
A second sub-step of performing discrete Haar wavelet inverse transform on the discrete Haar wavelet transformed low frequency image and outputting the local average image;
A third sub-step of subtracting the discrete Haar wavelet inverse transformed image from the image;
A fourth sub-step of squaring the image of the subtraction result;
The method according to claim 4, further comprising: a fifth substep of calculating a local average value of the squared image and outputting the local dispersion image.   The antibacterial effect determination method according to claim 4 or 5, wherein the first prior probability is modeled by a two-dimensional Gaussian function having a peak value at a central coordinate point in the image. An antibacterial effect determination program for determining an antibacterial effect of the antibacterial agent based on an image obtained by photographing the surface of the medium after cultivating the bacteria, leaving the antibacterial agent on the surface of the medium smeared with bacteria,
A first step of extracting, from the image, a local average image that is a feature amount of brightness of a local region of the image and a local dispersion image that is a feature amount of local flatness of the image;
Based on the first prior probability and the local average image and the local dispersion image, an effective area and an invalid invalid area that are effective for determining the antibacterial effect of the antibacterial agent for each pixel of the image. A second step of determining by Bayesian estimation and outputting a binary image composed of the effective area and the invalid area;
Blob analysis of the binary image to form a blob having a maximum area concatenating the effective areas, and determining a occupancy ratio of the blob in the image and an eccentricity ratio of the blob;
Antibacterial effect determination characterized by causing a computer to execute a fourth step of determining, by Bayesian estimation, the shape of the effective area of the image based on the second prior probability and the occupancy rate and the eccentricity rate program. The first step includes
A first sub-step of discrete Haar wavelet transform of the image;
A second sub-step of performing discrete Haar wavelet inverse transform on the discrete Haar wavelet transformed low frequency image and outputting the local average image;
A third sub-step of subtracting the discrete Haar wavelet inverse transformed image from the image;
A fourth sub-step of squaring the image of the subtraction result;
The antibacterial effect determination program according to claim 7, further comprising: a fifth sub-step of calculating a local average value of the squared image and outputting the local dispersion image.   The antibacterial effect determination program according to claim 7 or 8, wherein the first prior probability is modeled by a two-dimensional Gaussian function having a peak value at a central coordinate point in the image.

Images