JPH0255953A - Automatic identifying and processing method for bacteria - Google Patents

Abstract

PURPOSE:To distinctly binarize bacteria images by determining the differential value of dispersion ratios which are evaluation functions, binarizing the bacteria images with the point of the valley of the differential value as the candidate for the threshold value and determining the value at which the change rate of an area is of the min. value as the optimum threshold value. CONSTITUTION:The image of a sample is picked up by a camera 20 of an image pickup device 6 via a fluid system 2 to obtain the static image of the bacteria. The various density image of this image is obtd. via an A/D converter 22 and is stored in an image memory system 24. A control image processing section 8 obtains a density histogram from the various density image and binarizes the same by bisecting the histogram. The threshold value for the bisection is determined by dividing the between- class dispersion and within-class dispersion of the histogram and determining the ratio thereof to obtain the evaluation function. The density value at which the differential value of the evaluation function is min. is determined as the threshold value if the distinct binarization with the above-mentioned threshold value is not possible. The density value at which the change rate of the number of picture elements is min. when the binarization is executed is selected as the optimum threshold value. The specific bacteria is identified from the shape of the image extracted by this binarization.

Description

[Detailed Description of the Invention] <Industrial Application Field> The present invention relates to a method for identifying the types of bacteria contained in a biological sample, and in particular, a method for imaging a sample and obtaining a bacterial image from the image. This invention relates to a method for identifying bacteria by extracting them and determining the feature values of their images.

<Prior Art> Conventionally, as a method for testing bacteria, there is a method of preparing a smear-stained specimen and observing it with a microscope. However, such a method has problems in that it takes time and effort to prepare the specimen, requires skill in testing, and is also time-consuming. In order to solve this problem, the Institute of Electronics, Information and Communication Engineers journal DVol.

J70-D No. 4 Using the flow one method described on pages 827 to 829, image bacteria, extract an image of the bacteria from the obtained image, and identify the type of bacteria. An inspection device that can do this was devised.

<Problems to be Solved by the Invention> The size of bacteria is approximately as small as Igm, and most of the image obtained by the above-mentioned device is the background. Moreover,
A lot of noise and density unevenness occur in this background. In order to extract an image of bacteria, it is first necessary to binarize the captured image, but there is a problem in that it is not possible to binarize accurately if it is affected by noise or density unevenness. Also, depending on the bacteria, the density of the image and the density of the background may be relatively close, and in such cases, accurate 2
The problem was that it could not be converted into a value. Also,
Depending on the bacteria, it may be impossible to find the feature values from the image of the bacteria obtained through binarization and determine what kind of bacterium it is, or to extract the feature values directly from the binarized image depending on the bacteria. There was also the problem that there were some things.

An object of the present invention is to provide an automatic bacterial identification processing method that solves the above-mentioned problems and can more accurately identify the type of bacteria.

Means for Solving the Problems> In order to achieve the above object, the present invention converts an image signal obtained by imaging a sample collected from a living body using an imaging device, and generates a plurality of signals having a plurality of concentrations. Consists of pixels! ! By obtaining a light and dark image and setting a threshold for this gray image, an image of the bacteria in the biped sample is extracted, and the bacteria is identified based on the features obtained from the image of the bacteria. In this method, the ratio of the inter-class variance of the density histogram obtained from the grayscale image divided by the edge-based variance is used as an evaluation function, and the density value at which the evaluation function is maximized is used as the threshold value. Or when the above-mentioned grayscale image is binarized using each density value at which the difference value of the evaluation function is minimum as a threshold value, and when the density value before and after these threshold values is used as a threshold value to binarize it. The density value that minimizes the rate of change in the number of pixels is defined as the threshold value.

In addition, after performing noise removal and edge enhancement processing on the above-mentioned grayscale image, the above-mentioned threshold value is set to extract the above-mentioned bacterial image, the extracted image is subjected to degeneration and expansion processing, and then hole-filling processing is performed. It can also be used to correct the image of bacteria.

Furthermore, for an unidentifiable bacterial image, the skeleton is extracted by degenerating the image multiple times, and after the image is restored by expanding the image multiple times for each skeleton, the features are again calculated for each restored image. may also be used to identify bacteria.

Effect> A sample collected from a living body is imaged with an imaging device to obtain a still image of bacteria. This is A/D converted to obtain a grayscale image consisting of a plurality of pixels having a plurality of density values.

By obtaining a density histogram from this grayscale image and dividing this histogram into two, all density values are divided into two groups and binarized. A threshold value for dividing into two must be determined, but in the present invention, for this purpose, first the interclass variance and the due variance of the histogram are determined, and the ratio thereof is also determined. This is performed for all densities on the histogram to obtain an evaluation function. When this evaluation function becomes clear, the concentration value at which this evaluation function is maximized can be obtained. When this cannot be clearly binarized using this threshold,
When the difference value of the evaluation function or the minimum density 1 is determined, and when binarization is performed using these density values as thresholds, and when the density values near these thresholds are used as thresholds, 2 The density value that minimizes the rate of change in the number of pixels compared to when the value was converted is selected as the optimal threshold value. This is because a small rate of change in the number of pixels indicates that the two regions are optimally divided.

The feature amount of the image is determined from the shape of the image extracted by binarization. Using this characteristic amount, specific bacteria can be identified.

By applying noise removal processing to the grayscale image, the noise can be reduced (and by applying edge enhancement processing, the edges of the m bacteria can be made clear. Therefore, it becomes easier to perform binarization.2) By performing degeneration and expansion processing on the digitized image, irregularities on the edges of the extracted image are smoothed, and by further performing hole filling processing, extraction omissions within the image are corrected.

For images of unidentifiable bacteria, the features of the image are clarified by extracting the skeleton of the image and expanding the image for each skeleton.

<Example> This example describes the use of cocci (e.g. alpha streptococcus, beta streptococcus,
Neisseria, Staphylococcus leucosum, and quadruple coccus! This invention was carried out to identify Bacillus (e.g., Pseudomonas aeruginosa, Klebsiella pneumoniae, Escherichia coli, Salmonella typhi, Bacillus subtilis, etc.), and Figure 2 shows a block diagram of the equipment used for this purpose. It is a diagram.

This device consists of a fluid system 2, an optical system 4, a single image capturing section 6, and a control image processing section 8.

The fluid system 2 is a part through which bacteria flow, and has a sheath flow chamber IO. This chamber IO is made of quartz glass, and has a bacteria injection port 12 and a sheath liquid injection port 1 at the top.
4 and a tubule at the bottom? It has B15. When the sheath liquid is sucked from the lower part by the suction pump 16, the sheath liquid forms a sheath and flows through the thin tube portion 15, enveloping the bacteria, as shown in FIG. That is, the bacteria flow through the center of the thin tube portion 15, and since they flow in the same place as a trickle, it is easy to focus for imaging, which will be described later, and there are no simultaneous passages.

The optical system 4 has a nitrogen pulsed light laser 18 as a light source. The bacteria to be measured are extremely small, about IJLm, and require high-resolution photography. Since the resolution limit is proportional to the wavelength of the light source, short wavelengths are advantageous. There,
This nitrogen pulsed light laser has a wavelength of: l:17n+s. This nitrogen pulse light is transmitted to the thin tube section 1 of the chamber 10 through a plurality of condensing lenses (not shown).
5, and further enters the CCD television camera 20 of the image capturing section 6 via an objective lens (not shown).

The image capturing unit 6 is the CCD television camera 2 described above.
0, and the image signal is converted into a digital image signal by the A/D converter 22 and stored in the image memory system 24. Additionally, the image memory system 24 may be used as needed.
The digital image signal stored in the digital image signal is converted into an analog image signal and is displayed on a monitor television receiver 26 and recorded on a video tape recorder 28.

The control/image processing unit 8 has a microcomputer system 30, which controls the nitrogen pulsed light laser 18 and the image memory system 24, and also performs image processing as described later. This control/image processing section 8 is
It also has a monitor television receiver 32 for displaying the progress of image processing and the results of the image processing, and a flow disc 34 for recording the processing results.

FIG. 1 is a flowchart showing image processing performed by the microcomputer system 30. First, a digital imaging signal is input from the image memory system 24 (step St). This input digital imaging signal has deterioration at the edges of the bacteria being imaged and noise in the background. If these are left as they are, the exact 2
Since it is difficult to convert into values, noise removal and edge emphasis are performed (step S2.3).

Noise removal in step S2 is performed using an SNN-MEAN filter (Symm
eLric Nearest Neighbor-Mean
Filter). That is, as shown in FIG. 4, the density value g (X, Y) of the pixel (X, Y)
is (2n"l)
) is expressed as an average value of the density values of pixels selected by the following formula.

Ig(X, Y) g(X+i, Y+j) I<Ig(X
, Y)-g(X-i, Y-j) When l, g(X+
i, Y+j) Select.

I g(X, Y) −g(X+i, Y+j) I >
I g(X, Y) -g(X-i, Y-j) I
f) Go and select g (X-i, Y-j).

I g(X, Y) −g(X+i, Y+j) l =
l g(X, Y) -g(X-i, Y-j) l
17) When (g (X+i, Y+j) + g (X
−i, Y−j))/2. However, −n≦(i, j
)≦+n. For example, Fig. 4 is a diagram when n is 1, and if the density value g (X, Y) is one mouth, then 1s-2,
When j=1(7), I g(X, Y)-g(X-2
,Y+1) l becomes 110-81=2, and 1
g (X, Y) - g (X+2.Y-1) + 4;
i I 10-5=5, and l g (X, Y)-g
(X-2, Y+l) + (7) is smaller, so
Select g (X-2, Y, l) = 8. i-2,
When jm2, l g(X, Y)-g(X-2, Y
+2) l becomes l 10-81=2, and I g(X
, Y)-g(X+2.Y-2) l is 110-61=
4, l g (X, Y)-g (X-2, Y◆2
) + 77) is smaller, so g (X-2, Y+
2) Select =8.

When 1=-1,j-1, I g(X, Y)-g(X-
1, Y+1) l is 110-51=5, l g
(X, Y)-g (X+l, Y-.

is 110-61=4'), l g (X, Y
)-g (X+I, Y-1)+77) or small innote,
Select g (X+l, Y-1) = 6, i-1
,j-2(7) I g (X, Y)-g (X
-1, Y-2) 1 becomes t I to-71=3,
l g (X, Y) - g (X+1.Y2) I is 1
10-81=2, l g (X, Y)-g CX
÷1. Y-2) + 773 is smaller innote, g (
X+1. Select Y-2) = 8. i=0. j=l
(7) When l g (X, Y) - g (X, Y, 1
) I becomes t l 10-71 = 3, and l g
(X, Y)-g(X, Y-1) l is 110-81
= 2, I g (X, Y)-g (X, Y-1)
l is smaller, g (X, Y-1) = 8
Select. i=o, j=2 (7) When I g (
X, Y)-g (X, Y+2) I is 110-61=
4, I g (X, Y) - g (X, Y - 2)
l becomes 110-91=1, l g (X, Y) - g
(X, Y-2) + (7) is smaller, g
Select (X, Y-2) = 9. Also, i=1.
When j=0 (1), l g(X, Y)-g (X+1
．． Y) l becomes 110-71=3, and Ig(X.

Y) - g (X - 1, Y) I becomes 110 - 61 = 4, 1g (X, Y) - g (X + l, Y) + (7
) side or small innote, g(X+l.

Select Y)=7. When i-1, jgl, I g
(X, Y)-g (x++, y+t) 1 is +1O-
61=4, 1g(X,Y)-g(X-1,Y-
1) l is 110-71=3, and g (X, Y)
-g (X-1, Y-1) l is smaller, g
Select (X-1, Y-1) =7. i=1. j-
At 2, g (X, Y) - g (X+l, Y+2)
I becomes 110-71=3, l g(X, Y)-g
(X-1, Y-2) I is +1O-81=2 tons,
I g (X, Y)-g (X-1, Y-2) lc7)
Since the one is smaller, g (X-1, Y-2) = 8 is selected. i”21j-0ノ, l g(X, Y
)-g(X+2.Y) I is +1O-61=4, and l g(,X,Y)-g(X-2,Y) l is +1
0-91 = 1, and I g (X, Y)-g (
X-2, Y) + 77) or small, g (X-2,
Select Y) = 9. i=2. When j=1, 1
g(X,Y)-g(X+2.Y+1) I is +1O
−71=3, and I g(X, Y)−g(X−2
, Y-1) l is l 10-71 = 3, equal innote,
(g (X+2.Y+1) + g (X-2,Y-1
))/2=7. i=2. When j=2, 1g
(X, Y)-g (X+2.Y+2) l is 110-
81=2, g (X, Y)-g (X-2, Y-
2) l becomes 110-61=4, and I g (X,
Y)-g (X+2.Y+2)+77) is smaller.

gCX◆2. Select Y, 2) = 8. Then, these selected density values 8.8, 6.8, 8.9, 7.7
．． Let g(X, Y) be the average value of 7.75 of 8,9.7.8. In this way, the density value of each pixel is determined and noise is removed. In this case, only the background noise can be removed while preserving the edges of the object (bacteria image).

The edge enhancement in step S3 is performed by performing differential processing on the original image. That is, the density value of the image consisting of r
(i, j), the difference value g(i, j) is g(i, j) = [(Δ i)2 + (Δj)2
]1/2. However, Δi and Δj are as follows: Δi = [r(i+1.jd)+2f(i+I,j)
+f(i+I,j-1) ]-[f(i-1,j+1)
+2f(i-1,j)+f(i-1,j-1) ]Δ
j = [f(i-1,j+1)+2f(i,j+
1)+f(i+l,j+1)]-[f(i-1,j
-1)+2f(i,j-1) tenf(i+1.j-1)
]. This is known as the Sobel process.

By calculating the density sum of the image from which noise has been removed in step S2 and the image from which edges have been emphasized in step S3, an image in which noise has been removed from the original image and edges have been emphasized is obtained.

In order to binarize the image corrected in this way, a threshold value is set (step S4). This setting is based on an automatic threshold selection method using the variance ratio as an evaluation function. That is, the ratio F between the first class dispersion σa' and the intraclass dispersion σw'. (=σ, +1/σ
, 2) is set as the threshold value. However, the interclass variance σ8 and the due variance σw2 are as follows: σ112= Σ J (xm-x)2/Σ nkkE
NrlkεNn Nn= (1, 2...N) (rw2= Σ Σ(Xkt-Xk)"/Σ n
kkENn uεHz kEN.

It is. Here, N is the total number of classes, nk is the number of pixels in the second class, × is the average density value of pixels in the class, and Xbt
is the value of the intersecting pixel in the class or the average value of the density values of all pixels. For example, there is 1 pixel with density 1, 3 pixels with density 2, 2 pixels with density 3, 1 pixel with density 4, 4 pixels with density 5, 6 pixels with density 6, etc. When considering an image (all pixels a25) consisting of 5 pixels with density 7 and 3 pixels with density 8, the average value of the density values of all pixels or 5
．． It is 28. Here, N is 2 because it is binarized.

If the density l is a threshold value (divided into a class with a density of 1 or less and a class with a density greater than 1), the average value of the density values of pixels in the first class is 1, and the density value of pixels in the second class is The average value O is 5.46. Therefore, the interclass variance is (l(1-5,
28) "÷2.i(5,46-5,28)")/25 snow 0.76.

The due variance is 1(1-1)”/I+(:1(2-5,4
6) 2+2 (3-5,46) 241 (4-5,46)
2◆4(S-5,46)2◆6(6-5,46)”+5
(7-5,46)2, :I(8-5,46)2)/24
-3.50, and the value of the evaluation function is 0.76/3.5
It becomes 0-0.22. If density 2 is the threshold value, then the first
The average density value of the pixels in the second class is l, 75, and the average density value of the pixels in the second class is 5.95. Therefore, the interclass variance is (4(1,75-5,28)2+21
(5,95-5,28)”)/25-2.:17, the due variance is (1(1-1,75)2.:1(2-1,75
) 2)/4+ (2(:l-5,95)2+1 (4
-5,95) 2+4 (5-5,95)246(6-
5,95)"+5(7-5,95)"+3(8-5°,
95)2)/21=2.2:l, and the value of the evaluation function is 2.3'l/2.23=1.06.

The values of the interclass variance, marginal variance, and evaluation function for each threshold value obtained in the same manner are shown below.

As is clear from this table, the value of the evaluation function is maximum when the threshold value is 3, so this density 3 is used as the threshold value.

When binarization is performed using the threshold value set in this manner, the binarization is normally performed clearly, but in some cases, it may not be possible to binarize clearly. Figure 5(a),
(b) shows the density histogram and evaluation function values when clear binarization is possible, and Figures 6 (a) and (b) show clearly binarized values.
These are the density histogram and the evaluation function value when it could not be converted into a value. When it can be clearly binarized, the evaluation function is
forms a beautiful mountain, and the threshold value takes the concentration value of the maximum value of the variance ratio. However, if the bacterial concentration value is close to the background concentration value, the evaluation function forms a hill in the middle of the mountain. has appeared.

For this reason, it is thought that clear binarization is not possible.

Therefore, after determining the threshold value in step S4, it is determined whether the threshold value is appropriate (step S5), and if it is not appropriate, the threshold value is optimized (step S6). Therefore, after smoothing the values of each evaluation function, the difference value is determined. Note that although the smoothing was done in Figures 5(b) and 6(b), the evaluation function is actually drawn in a zigzag manner, which has a negative effect when calculating the difference value. This is to remove it. The difference values obtained in this way are shown in Figure 7. In the same figure, the valley a is based on the threshold determined by the automatic threshold method, and at a higher concentration, b, c
, d valleys are formed. So, these, c, d
are threshold candidates, binarization is performed using these candidate thresholds (step S7), the rate of change in area in that case is investigated, and the threshold when the rate of change is the smallest is determined as the optimal threshold. The value was determined. That is, first, the threshold value is moved back and forth slightly to perform binarization, and the number of pixels (area) that was set as l at that time was counted, and the number of pixels (area) that was set as 1 when the threshold value was set was counted. The number of pixels (area) is also counted and the rate of change in the area is determined. The rate of change in area is determined in the same manner for other threshold values. And among the thresholds S, c, and d,
The smallest rate of change in area is set as the optimal threshold.

If the threshold value is optimal, for example, Fig. 6(a)
As shown in the figure, the threshold value is located in the valley of the histogram, so even if you move the threshold value back and forth and perform binarization, the number of pixels that become l does not change much. This is because if it is not the threshold value, the rate of change per area will increase when the threshold value is moved back and forth as shown in FIG.

The edges of the bacteria extracted by binarization in this way are extremely uneven as shown in Figures 8(a) and (b), so the edges are corrected by shrinking and expanding. Step S8). In FIG. 8(a), the solid line indicates a binary image of bacteria, which is degenerated based on 8-connection. That is, for each pixel, pixels in which all eight adjacent pixels at the top, bottom, left, right, top right, bottom right, top left, and bottom left are part of the bacterial image are left, and the other pixels are deleted. The deleted state is shown by hatching. From this state, each pixel is expanded in the eight directions mentioned above. This state is shown by a chain line. As a result, an image with unevenness removed can be obtained. Figure 8(b) shows the upper and lower
A case is shown in which degeneration and expansion are performed based on four left and right connections. The hatching indicates a degenerate state, and the one-dot chain line indicates an expanded state of the degenerate state. In step S8, if there is a hole in the center of the binarized image of bacteria, the hole is also filled.

For the bacterial image processed in this way, the surface MA,
The 0th plane MA, which has a circumference and a circularity R1 and a distance variance V, is the sum of the number of pixels of bacteria. Therefore, for example, in the case of FIG. 9, the area is 64. The perimeter is the sum of the distances around the bacterial image, where the distance between pixels adjacent to the bacterial image on the top and bottom or left and right is 1, and the distance between diagonally adjacent pixels is 5. In the case of FIG. 9, the zero circularity R that is zo+sf is 4πX A / L 2. The distance dispersion V is
=Σ(D(i,j)-D)2/A. D(i,j)
is the Euclidean distance from pixel (i, j) of the bacterial image to the center of gravity, and D is the average distance.

Identification is performed using these feature amounts (step S t
O). A detailed flowchart of this identification is shown in FIG. 1O. First, distinguish between a single coccus and a single Bacillus longus (step Sll, 12). - Two-dimensional distribution of circularity 1 Perimeter - Two-dimensional distribution of circularity, distance variance - 0 based on two-dimensional distribution of perimeter (The number of pixels) 100 or the circumference of 45 is set as a set plane, and anything smaller than this can be identified as a single sphere, and an area of 500 is set as a set plane, and anything larger than this can be identified as a single Bacillus longus. Furthermore, when using the two-dimensional area-distance distribution distribution shown in FIG. 12, the set plane is 1 plane gxoof and the distance variance is 6.0, and anything smaller than this can be identified as a single coccus. In addition, in the area-circularity two-dimensional distribution in Figure 13, circularity of 88% or less can be identified as a single sphere.
The above can be identified as Bacillus longus. Furthermore, when the two-dimensional distribution of circumference-circularity shown in FIG. 14 is used, one having a circumference of 40 or less can be identified as a single coccus. Furthermore, when using the two-dimensional distribution of distance variance - perimeter length in Figure 15, the distance variance is 4.5.
The following can be identified as a single coccus, and the distance variance of 35.0 or more can be identified as a single bacillus.

Things other than single cocci and single long bacilli (aggregates of cocci and short bacilli) cannot be identified as they are, so distance conversion is performed (step S1: l) and skeleton extraction is performed (
Step 514).

When the distance conversion in step 13 is performed by density-weighted distance conversion using parallel processing, a grayscale image of size MXN f(
i, j) (see Figure 16(a), however, the 0 part is omitted), provided that f(i, j)≧0, i
E N l= (1, 2,...M), j
εN,=(l, 2,...N). Further, the number of connected pixels is assumed to be four. The initial image g'(i, j) is defined as follows for all iEN(,jEN3).

However, C is a sufficiently large positive integer (Figure 16(b) is 3
0). Sequence of images: g'(i, j) (to=
0゜l, 2...).

In Fig. 16(c), gl*I(jj), g2"'(i, j
), g”l(i, j). All s E N +
, JεN, g−1(i,j)=gk(i,j
), the process ends. g'(i, j) at this time
The density weighting of f(i, j) is a distance image, and skeleton extraction in step S14 is performed by extracting the maximum density of f(i,j). The skeletons thus extracted are labeled (step 515), and each skeleton is extended multiple times to restore the object (step S!6). 1st
Figure 7 (a) shows the state after distance conversion, Figure 7 (b) shows the state in which the skeleton has been extracted, and Figure 7 (C) shows the state in which the extracted skeleton is expanded. The first expansion part, the hatching on the right is the second expansion part,
The dotted area indicates the third expansion, and this expansion is 4J! This is done based on one knot.

In the same manner as described above, characteristic values such as perimeter, circularity, and distance dispersion are determined for the expanded specimen (step S1?). Identification is based on

<Effect of Jl> As described above, according to the present invention, when the threshold for converting a bacterial image into a binary image cannot be clearly converted into a binary image even after determining the threshold value, the evaluation function Find the difference numerator 1 of the dispersion ratio, use the points that are the valleys of these difference values as threshold candidates, perform binarization using these candidate thresholds, and find the threshold with the smallest rate of change in one area. is the optimal threshold, so
Even if the density of the bacterial image and the background density are close to each other,
Bacteria images can be clearly binarized and bacteria can be identified well.

Furthermore, even if there is noise in the background, it is removed and the edges of the bacterial image are emphasized, allowing for even clearer binarization.

In addition, for bacteria images that could not be identified, the skeleton is extracted, reconstructed based on this skeleton, and feature quantities are extracted to identify the bacteria, so the running power is improved without identifying the bacteria.

[Brief explanation of the drawing]

Fig. 1 is a flowchart of an embodiment of the automatic bacterial identification processing method according to the present invention, Fig. 2 is a diagram of the equipment used in the embodiment, and Fig. 3 is a chamber diagram of the equipment used in the embodiment. FIG. 4 is an explanatory diagram of noise removal in step S2 of FIG. 1. FIG.
FIG. 6(b) is a diagram showing the relationship between the evaluation function of the histogram in FIG. 5(a) and the density, FIG. Figure (b) is a diagram showing the relationship between the evaluation function of the histogram in Figure 6 (a) and density, Figure 7 is a diagram showing the difference value of the evaluation function in Figure 6 (b), Figure 8 (a ), (b) is step S in Figure 1.
Figure 9 is an explanatory diagram of how to calculate the area and perimeter in step S9 of Figure 1. Figure 1O is an illustration of how to calculate the area and perimeter in step S9 of Figure 1. Detailed flowchart of step SIO identification, Figure 11 is a two-dimensional distribution map of bacterial area-perimeter, Figure 2 is a two-dimensional distribution diagram of bacterial area-distance distribution, and Figure 13 is a two-dimensional distribution diagram of bacterial area-distance distribution.
Figure 14 is a two-dimensional distribution diagram of circularity; Figure 14 is a two-dimensional distribution diagram of bacterial perimeter-circularity; Figure 15 is a two-dimensional distribution diagram of distance dispersion-perimeter of m bacteria; Figures 16 (a) to (e) is an explanatory diagram of the distance conversion in step S13 in Fig. 10, and Fig. 17 (a)
7(c) are explanatory diagrams of skeleton extraction and expansion of step S14.16 in FIG. 1O. 20... CCD television camera, 24...
Image memory system, 30...Microcomputer system.

Claims (3)

[Claims] (1) A/D converting the image signal obtained by imaging a sample collected from a living body with an imaging device to obtain a grayscale image consisting of multiple pixels having multiple densities, and thresholding this grayscale image. In the method of extracting the image of the bacteria in the sample by setting and identifying the bacteria based on the feature values obtained from the image of the bacteria, the interclass variance of the concentration histogram obtained from the grayscale image is The ratio divided by the intraclass variance is used as the evaluation function, and (a) the concentration value at which the above evaluation function is maximum. or (b) When the above-mentioned grayscale image is binarized using each density value at which the difference value of the above-mentioned evaluation function is minimum as a threshold value, and when the density values before and after these thresholds are used as a threshold value, it is binary-valued. An automatic bacterial identification processing method that uses, as the threshold value, a density value that minimizes the rate of change in the number of pixels compared to when the number of pixels changes. (2) After performing noise removal and edge enhancement processing on the above-mentioned grayscale image, the above-mentioned bacterial image is extracted by setting the above-mentioned threshold, and the extracted image is subjected to degeneration and expansion processing, and then hole-filling processing is performed. 2. The automatic bacterial identification processing method according to claim 1, wherein the bacterial image is corrected by applying the following steps. (3) For an unidentifiable bacterial image, the skeleton is extracted by degenerating the image multiple times, and the image is restored by expanding the image multiple times for each skeleton, and then the characteristics are restored for each restored image. Claim 1 or 2: determining the amount and identifying bacteria.
Automatic bacterial identification processing method described.