JPH0159616B2 - - Google Patents

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention is applicable to the analysis of the background of an image in which the difference from the background is not clear among the objects to be analyzed such as blood cells and urine sediment obtained during an optical field of view. This relates to a method and device that accurately determines the difference between the object and the object, and that enables much faster processing than conventional methods.
In particular, it is an object of the present invention to provide an image analysis method and apparatus for performing region segmentation of an object to be measured in a microscopic image at high speed.

[Prior art] Conventionally, as in white blood cell classification, objects such as white blood cells in a smear sample are classified and counted based on morphological differences or differences in size, density, etc.
When providing clinical data, microscopic images are analyzed using various algorithms and classified and counted.

Furthermore, Japanese Patent Application Laid-Open No. 49-93095 describes an analysis method and apparatus using color algebra and image processing technology, which algebraically combines signals in a first wavelength band and a second wavelength band to classify areas. has been done. However, this publication does not describe anything about obtaining a threshold value as in the present invention.

[Problem to be solved by the invention]

In cases such as white blood cells that are stained at relatively high concentrations, there is a clear optical difference between the cells and the background, whereas when classifying and counting substances that contain nearly transparent substances such as urine sediment, Since there is almost no difference between the object to be measured and the background, it is difficult to analyze all images using a fixed threshold set in advance. Particularly when performing classification and counting, when dividing regions into areas such as the background and the object, parts of the object with relatively high concentration such as the nucleus, and parts with low concentration such as protoplasm. For example, the image 1 obtained depending on the degree of staining, the quality of the smear, the drift of the optical system, and the position on the smear.
Since the density of the area is detected differently for each image, it is necessary to find and set an optimal threshold value for each image. Therefore, the density information of each pixel is captured into the memory address corresponding to each pixel of the image, and then a density histogram is created using a computer or the like, or similar pixels are The optimal threshold value for performing region segmentation focusing on the continuity of features was determined by calculation, that is, by computer software. In any case, in order to obtain the optimum value, arithmetic processing is performed by changing the conditions little by little for the same image, which has the disadvantage that the calculation takes an extremely long time. In other words, the method of inputting information into a computer's memory and calculating the threshold value for analysis allows for much more accurate analysis than the method of using a pre-fixed threshold value. Although it is possible to add precise correction calculations, etc., it takes a considerable amount of time to process the calculations, and the analysis speed per sample becomes extremely slow. It was not practical for use in hospital laboratories that require processing.

The present invention has been developed in view of the above points, and it is possible to determine the optimal threshold value and improve the image quality even when the difference between the object in the image and the background is not clear, such as a glass cylinder in urine sediment. Processing can be done almost simultaneously when it is imported into memory, significantly reducing the time required from several minutes to several tens of minutes to achieve the same level of accuracy to about 1/60 to 1/30 seconds. The object of the present invention is to provide an image analysis method and an apparatus for the same.

[Means and actions for solving the problem]

In order to achieve the above object, the image analysis method of the present invention calculates a density difference value between each pixel of an image obtained by an imaging device and a previously measured pixel, and has a density difference value that is less than or equal to a predetermined value. The method is characterized in that a density histogram of pixels is obtained, the position of a valley in this density histogram is detected, and the image is analyzed using the density value at the position of this valley as a new threshold value to divide the image into regions.

Further, the image analysis device of the present invention will be described with reference to FIG. 7, and will be described with reference to FIG. 7. an imaging device 1 that generates an image; a shading correction circuit 2 that is connected to the imaging device and sends a correction signal to the imaging device to correct the output signal according to the position of the image, and corrects it so that it outputs the original signal; A sampling pulse generator 3 receives a horizontal synchronizing pulse signal from an imaging device and generates sampling pulses corresponding to each pixel, and AD converts the density of each pixel using this sampling pulse to generate a logical value representing the density. The output AD converter 4 and this
A read access memory 6 connected to an AD converter via a first memory address register 5, and an AD
An image memory 8 connected to the converter 4 via the first memory data register 7, two registers 10 and 11 connected to the AD converter 4, and data in the two registers connected to these registers. a subtracter 12 for calculating a density difference value by comparing and subtracting the density difference values; Digital comparator 1 that removes things
3, and is connected to this digital comparator and read access memory 6 via a second memory data register 14, and is preset to a first memory address register 5.
The number already written in the read access memory 6 is read out using the logical value specified by the address, 1 is added to that value, and the number is read out again in the read access memory 6.
a digital adder 15 for writing to the second
a computer 16 connected to the memory data register 14 and the image memory 8 via the first memory data register 7; an X counter 17 provided between the image pickup device 1 and the image memory 8;
It includes a Y counter 18 and a second memory address register 20.

The present invention is carried out in the following manner. First, a smear such as urine created on a slide is
Photoelectric conversion is performed using a microscope camera or the like, and the density of the light is extracted as an electrical signal. These electrical signals are pulse signals corresponding to each pixel, that is, each dot on the photoelectric conversion surface, and the height of this pulse signal is proportional to the intensity of light. Assuming that the density of each point, that is, the pixel, of the image shown in Figure 1 is fi, j, the scanning of this plane advances in the i direction from 1 to m, advances by 1 in the j direction, and then again in the i direction from 1 to m. Proceed, total m×n
signals for each pixel can be obtained. When using a general television camera, m×
The time to obtain n pieces of information, that is, the time to scan a surface once, is about 1/6 second.
The electrical signal thus obtained is converted into a digital logical value by performing AD conversion on the pulse signal of each pixel. Conventionally, this logical value was stored directly in memory, and the computer read each address and performed all arithmetic processing, which took a very long time, but the present invention speeds up the process as described below. I tried to increase my popularity. In other words, whether each pixel in an image belongs to the background, the protoplasm, or the nucleus can be determined by setting a predetermined threshold, but as mentioned above, this threshold Values had to be set for each image, and in the past, all of this had to be computed on a computer, and then the information in memory had to be divided, which required twice the effort.

In the present invention, at the same time as the AD conversion value of the image signal is taken into the memory, the difference in density between each pixel is stored in another memory as a histogram. In other words, if the density of a certain point in the image is fi,j, then the difference Δfi, determined by j=fi, j−fi, ₁ and j, is
j is determined, and the fixed number stored at the intersection of the memory address corresponding to Δfi,j and the address corresponding to the concentration fi,j is read out, 1 is added thereto, and the fixed number is stored at the same address again. Therefore, the first memory contains mere density information corresponding to one side of the image.
The second memory stores histogram information of the density and difference value of each point of the image. Therefore, the address of the first memory corresponds to the two-dimensional plane of the image, but the address of the second memory does not correspond to the two-dimensional plane of the image at all, but is related to the density distribution information once converted. It is a street address. The contents stored in this second memory are as shown in FIG. 2, and indicate how many pixels there are corresponding to a predetermined density value and density difference value. Note that the numbers in FIG. 2 indicate degrees. A difference value of zero indicates that there is no density difference between adjacent pixels, which occurs, for example, on a uniform background or inside a cell. In Figure 2, the part near the center that is shifted to the right corresponds to pixels on the boundary between the cell and the background, and it is possible to obtain a remarkable histogram that could not be obtained with a simple histogram of concentration distribution. can. Furthermore, in FIG. 2, by setting a threshold value for the density difference value in advance,
By removing pixels having a difference value greater than a predetermined value, that is, a density gradient, a histogram for pixels less than the threshold value dh can be obtained. This allows the background and the object to be clearly distinguished.

Figure 3 shows the distribution curve for density values for pixels below the threshold dh.
The value fv (=30) is a stable boundary level between the background and the object. If this determined boundary level, that is, density fv, is used as a new threshold and the density of the image already captured in the first memory is binarized (to 0 or 1), the background and object can be separated. Ru. FIG. 4 shows an example of a case where an image is divided into three types of density (for example, background, protoplasm, and nucleus), and FIG. 5 shows a density histogram obtained from this. Note that the sample is a binuclear transitional epithelial cell. In Figure 4, each number and A, B, C, and D indicate how many pixels corresponding to each density and difference value are included in the image; 1 to 9 and A are 10, and B is 11, C is 12, D is
Showing 13 pieces. Note that the threshold value dh is set in advance so that the stable region of the background and the object is included. In other words, if the threshold value dh is set too large, noise components (information due to dust, etc.) will be picked up, and if it is too small, even the main background and object information will be picked up. Setting is difficult. Furthermore, averaging is an important element in processing for obtaining a density histogram. That is, when the numbers below the threshold value dh in Figure 4 are accumulated for each concentration, the unevenness is severe as shown in the top number (sum of threshold values) written at the bottom of Figure 4; By repeating the addition of adjacent cumulative values twice (averaging), a relatively smooth density histogram as shown in FIG. 5 can be obtained. In FIG. 5, the left side is a group of background densities, and the right side is a group of object densities. The valley of the right curve is the boundary value of the concentration difference.

The median value of the valley obtained in the above manner is used as the threshold for the next image analysis, and an example of the result obtained by processing one screen already stored in the first memory is shown in the sixth example. As shown in the figure. In FIG. 6, the background is shown as blank, the cytoplasm is shown as +, and the nucleus is shown as *. Once the regions are divided in this way, the processing for classification is easy. For example, in classification based on area, even if the image quality varies somewhat depending on the staining situation, the above processing can always obtain a constant threshold value, so accurate area can be obtained. Furthermore, in feature extraction based on the presence or absence of a nucleus, information from high-concentration dust and scratches other than the nucleus is excluded when determining the above threshold (because the difference value exceeds the threshold dh). ) Since only continuous information of two or more consecutive pixels is used as information for threshold determination, it is possible to accurately determine the nucleus and divide the area. can be obtained accurately. The above method can also be applied to objects that are semitransparent or nearly transparent, such as glass cylinders in urine sediment, where there is no noticeable difference in staining, and where it is difficult to set a threshold value depending on the background density. According to the method, the threshold value is determined only based on valid information from which unstable portions of boundaries and noise components are removed, so that it is possible to reliably divide regions. In the subsequent processing, determination is performed using conventional pattern recognition or other feature extraction techniques, and fractional counting is performed.

〔Example〕

Next, FIG. 7 shows an example in which the above method is applied to an actual circuit. The image analysis device shown in FIG. 7 includes an imaging device 1 such as a television camera, a shuddering correction circuit 2 connected to the imaging device 1, and a pixel-by-pixel unit that receives a horizontal synchronization pulse signal from the imaging device 1. A sampling pulse generator 3 generates a sampling pulse corresponding to
An AD converter 4 that performs AD conversion and outputs a logical value representing concentration; a read access memory 6 connected to this AD converter 4 via a first memory address register 5; an image memory 8 connected via a register 7;
two registers 10 connected to the AD converter 4,
11, a subtractor 12 connected to these registers 10 and 11 for comparing and subtracting the data in the two registers to obtain a density difference value, and a subtractor 12 that is connected to these registers 10 and 11 to obtain a density difference value by comparing the data of the two registers and performing subtraction, and A digital comparator 13 is connected to the digital comparator 13 and the read access memory 6 via a second memory data register 14, and is connected to the digital comparator 13 and the read access memory 6 through a second memory data register 14. a digital adder 15 connected to the second memory data register 14 and a computer 16 connected to the image memory 8 via the first memory data register 7; X counter 17 and Y counter 18 provided
and a second memory address register 20.

The image obtained by the imaging device 1 usually has a shading characteristic, and since the brightness of the image is not uniform, a phenomenon occurs in which the obtained image is dark at the periphery and bright at the center. Therefore, the shading correction circuit 2 is a circuit that sends a correction signal for correcting the output signal according to the position of the image to the imaging device 1, and performs correction so that the original signal is output. Generally, a method is used in which the output signal for one screen without an object is stored, and the image signal obtained later is corrected by a computer and output, but this method has problems in terms of speed. Therefore, if the correction is performed in synchronization with the scanning time of the imaging device 1, the speed will be increased. The imaging device 1 outputs an analog signal corresponding to the density of a pixel of an image, and a horizontal synchronization pulse signal and a vertical synchronization pulse signal necessary for indicating the position of the pixel. In response to the horizontal synchronizing pulse signal, the sampling pulse generator 3 generates sampling pulses corresponding to each pixel as shown in FIG. 8. The density of each pixel is AD converted by this sampling pulse, and the logical value f representing the density is
It is output from the AD converter 4. The output f of the AD converter 4 is sent to the image memory 8, and also to the read access memory 6 and the register 10 in order to perform the above-mentioned area division threshold setting.
In the above-mentioned method, in order to make the explanation easier to understand, a histogram including the difference values as shown in FIGS. 2 and 4 is shown, and based on this, the number of particles is calculated for the concentration as shown in FIGS. 3 and 5. We have explained how to obtain a density histogram that displays the difference value, but as a result, it is sufficient to cut out the part where the difference value is too large and create a density histogram of only the main part, so that part can be omitted in the actual circuit. Note that the first memory address register 5 and the second memory address register 20 are temporary storage elements for specifying an address in the memory, and the former specifies the address corresponding to the density, and the latter specifies the memory address corresponding to the pixel position. Specify the address within.
On the other hand, the second memory data register 14 and the first memory data register 7 are temporary storage elements for information to be written into the memory, and writing is performed when signals to the address register and data register are aligned. Registers 10 and 11 are so-called registers to which data is transferred, and when current data is input, previous data is transferred from register 10 to register 11, and the stored contents change one after another. This timing is provided by the pulses of the sampling pulse generator. The previous data is stored in the register 11, so the data in the registers 10 and 11 are compared, and the subtracter 12
By performing subtraction, the difference value Δf can be found. This difference value Δf is compared with a preset threshold value dh by a digital comparator 13, and only those below the threshold value are passed, and those above the threshold value dh are removed. In other words, below the threshold dh is 1, and the threshold dh is
The above shall be set as zero. Next digital adder 15
has already been accessed in the read access memory 6 by the logical value of the address specification in the memory address register 5.
This is to read out the number written in , add 1 to that value, and write it into the read access memory 6 again. Therefore, when the difference value Δf is larger than the threshold value dh, it is zero, so
The contents of read access memory 6 remain unchanged. In other words, the original memory contents remain the same. On the other hand, the specified address counted by the X counter 17 and Y counter 18, that is, the v, y coordinate data corresponding to the pixel position of the image, is sent to the image memory 8 via the second memory address register 20. , while A.D.
The output f of the converter 4 is sent to the first memory data register 7
and stored as density information in the image memory.

As described above, image information is stored in the image memory 8, while density histogram information of pixels having a difference value below the threshold value dh is stored in the read access memory 6. Each image is processed at a high speed of 1 second. When memory writing for one image is completed, the density histogram is averaged as described above, the valley position is detected according to the main program built into the computer, and then the valley position is updated. The density of the pixel stored in the image memory 8 is processed using a new threshold value, and is converted into a binary value of 0 or 1 or 0, +,
The region is determined by performing multivalue conversion such as ternary conversion such as *. An example of the output obtained by this process is as shown in FIG. 6, and subsequent processes such as classification are also performed according to the program stored in the computer 16. In conventional classification and counting devices of this type, areas were determined by computer software programs without using the methods and devices described above, or all processing was done using fixed thresholds. The latter took too much time, while the latter did not provide accuracy.
According to the method and device of the present invention, accuracy is improved and time is significantly shortened, and multi-screen processing of multiple samples can be performed in a short time. The processing capacity of devices, etc. can be significantly improved.

[Effects of the Invention] In the method and apparatus of the present invention, remarkable histograms that cannot be obtained with conventional density histograms, such as a uniform background, the inside of a cell, or the boundary between a cell and a background, can be obtained. be able to. Further, since it is possible to obtain a histogram in which pixels having a density difference value exceeding a predetermined value are removed, the influence of noise such as high density dust can be removed.

Therefore, when the difference between the object and the background in the image is not clear (for example, when the image contains near-transparent substances such as urine sediment), or when there is noise, the optimal threshold value cannot be set. It has the effect of being able to make decisions.

Furthermore, in the device of the present invention, the image signal
At the same time as the AD conversion value is taken into the image memory, the difference in density of each pixel is stored as a histogram in the read access memory. Further, a digital comparator excludes data with density difference values exceeding a predetermined value from being stored in the read access memory. These things shorten processing time.

In this way, the processing time can be significantly reduced by setting thresholds for region division, which was conventionally done by software, by using hardware.

As explained above, according to the present invention, threshold settings for dividing the object to be measured, which were conventionally done using software (computer programs), can now be done using hardware (circuits). Therefore, what used to take several minutes to several tens of minutes can now be done in about 1/60 to 1/30 seconds, significantly reducing the time. In addition, conventional methods using fixed thresholds, which are fast, do not have accuracy when processing samples that change or translucent objects, but the method and apparatus of the present invention This allows highly accurate processing to be performed in accordance with the concentration of the sample without changing the speed, and is less susceptible to the effects of noise components such as dust and scratches.

[Brief explanation of drawings]

Fig. 1 is an explanatory diagram showing pixels in an image taken by an imaging device such as a television camera, Fig. 2 is an explanatory diagram showing the number of pixels corresponding to a predetermined density value and density difference value, and Fig. 3 is an explanatory diagram showing the number of pixels corresponding to a predetermined density value and density difference value. FIG. 4 is an explanatory diagram showing an example of a case where an image is divided into three types of density (for example, background, protoplasm, and nucleus);
FIG. 5 is a density histogram obtained from FIG. 4, FIG. 6 is an explanatory diagram showing an example of the result of processing one screen, and FIG. 7 is an example of the apparatus of the present invention. A systematic explanatory diagram, FIG. 8, is a waveform diagram of sampling pulses corresponding to each pixel in the method of the present invention. 1... Imaging device, 2... Shading correction circuit, 3... Sampling pulse generator, 4...
AD converter, 5... first memory address register, 6... read access memory, 7... first memory data register, 8... image memory, 10,
11...Register, 12...Subtractor, 13...Digital comparator, 14...Second memory data register, 15...Digital adder, 16...Calculator, 17...X counter, 18...Y counter ,
20...Second memory address register.

Claims (1)

[Claims] 1. A density difference value between each pixel of an image obtained by an imaging device and a previously measured pixel is obtained to obtain a density histogram of pixels having a density difference value equal to or less than a predetermined value. Detect the position of the valley of
An image analysis method characterized by analyzing the image and dividing the region by using the density value at the position of this valley as a new threshold. 2. An imaging device 1 that generates an analog signal corresponding to the density of a pixel in an image and a horizontal synchronization pulse signal and a vertical synchronization pulse signal necessary to indicate the position of the pixel, and an imaging device connected to this imaging device that A shading correction circuit 2 that sends a correction signal for correcting the output signal to the imaging device and corrects it so that it outputs the original signal, and a sampling circuit that receives a horizontal synchronization pulse signal from the imaging device and corresponds to each pixel. A sampling pulse generator 3 that generates a pulse, an AD converter 4 that performs AD conversion on the density of each pixel using this sampling pulse and outputs a logical value representing the density, and a first memory address register 5 connected to this AD converter. an image memory 8 connected to the AD converter 4 via the first memory data register 7, and two registers 10 and 11 connected to the AD converter 4. , a subtracter 12 connected to these registers for comparing and subtracting the data in the two registers to obtain a density difference value, and a threshold value for comparing this density difference value with a preset threshold value. A digital comparator 13 that passes only those that are below a threshold value and removes those that are above a threshold value, and a first memory address register that is connected to this digital comparator and read access memory 6 via a second memory data register 14 and that is connected to the first memory address register in advance. a digital adder 15 for reading out the number already written in the read access memory 6 according to the logical value of address designation 5, adding 1 to that value and writing it to the read access memory 6 again; and a second memory data. a computer 16 connected to the register 14 and the image memory 8 via the first memory data register 7; an X counter 17 and a Y counter 1 provided between the image pickup device 1 and the image memory 8;
8 and a second memory address register 20.