KR102357212B1 - Method and Apparatus for Auto-focusing Based on Random Forest - Google Patents

Abstract

Disclosed are a random forest-based auto-focusing apparatus and method.
This embodiment automatically adjusts the focal length of the input image using an optical system capable of adjusting the focal length of the glass substrate and a previously learned random forest-based classification model. To provide an autofocusing apparatus and method for estimating and adjusting the

Description

{Method and Apparatus for Auto-focusing Based on Random Forest}

The present invention relates to a random forest-based auto-focusing apparatus and method. More particularly, it relates to a random forest-based auto-focusing apparatus and method for being mounted on inspection equipment used for defect detection of a glass substrate.

The content described below merely provides background information related to the present invention and does not constitute the prior art.

The display industry is an industry that has grown in earnest along with the spread of TVs or computers, and is being converted to next-generation displays through flat panel displays. As the size of the raw material glass plate and plastic substrate is getting bigger and thinner in accordance with the enlargement and high quality of the display, the difficulty of detecting micro-defects that may exist in the glass substrate is also increasing accordingly.

When detecting micro-defects using inspection equipment equipped with a high magnification lens, a lens depth of field problem occurs depending on the flatness of a large thin glass substrate, and the lens is not in focus. it may not be In such a situation, if a fixed focus method is used, a blurred image may be obtained due to an incorrect focus, so that it is difficult to detect a defect.

Meanwhile, in the conventional auto-focusing method, a focus value corresponding to the position of the lens is calculated, and the position of the in-focus is searched using this. Although this method is accurate, there is a problem in that the processing time for detection increases due to a decrease in processing speed due to an increase in the amount of computation.

Accordingly, there is a need for an autofocusing apparatus capable of quickly and accurately controlling a focus for detecting defects on a glass substrate.

Non-Patent Document 1: Genuer. R, J. Poggi, C. Tuleau-Malot, N. Vialaneix, "Random Forests for Big Data", Big Data Research, Vol. 9, pp. 28-46, 2017.

The present disclosure automatically calculates a focal length for an input image using an optical system capable of adjusting a focal length for a glass substrate and a pre-trained random forest-based classification model. The main object is to provide an autofocusing apparatus and method for estimating and adjusting.

According to an embodiment of the present invention, a random forest-based classification model for estimating a focal length by acquiring an image of a region of interest (ROI) for a subject from an optical microscope ); a control unit generating a control signal for adjusting a position of the optical microscope in a direction to reduce a distance difference between the focal length and an in focus; and a Z-axis driver for adjusting the position of the optical microscope by using the control signal.

According to another embodiment of the present invention, in an auto-focusing method used by an auto-focusing device, the process of acquiring an image of a region of interest (ROI) for a subject from an optical microscope; estimating a focal length by inputting the image of the region of interest into a previously learned random forest-based classification model; generating a control signal for adjusting a position of the optical microscope in a direction to reduce a distance difference between the focal length and an in focus; and adjusting the vertical position of the optical microscope by using the control signal.

According to another embodiment of the present invention, there is provided a computer program stored in a computer-readable recording medium in order to execute each step included in the auto-focusing method.

As described above, according to this embodiment, the focal length of the input image using an optical system capable of adjusting the focal length of the glass substrate and a previously learned random forest-based classification model. By providing an autofocusing apparatus and method for estimating and adjusting a focal length, it is possible to quickly focus an image acquired within a predetermined distance from an in focus.

Also, according to the present embodiment, by repeatedly applying a random forest-based classification model to an image acquired at a distance from the original focal point, it is possible to increase the recognition rate of the focal length.

1 is an exemplary diagram of an auto-focusing apparatus according to an embodiment of the present invention.
2 is a conceptual diagram of an optical microscope according to an embodiment of the present invention.
3 is a conceptual diagram of a random forest-based classification model according to an embodiment of the present invention.
4 is a flowchart of an autofocusing method according to an embodiment of the present invention.
5 is an image of an entire image and a region of interest according to an embodiment of the present invention.
6 is a graph illustrating a verification result for the verification set 2 according to an embodiment of the present invention.
7 is a graph illustrating a verification result for the verification set 3 according to an embodiment of the present invention.
8 is a conceptual diagram for global search and hill climbing search, which are types of focus detection algorithms.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to exemplary drawings. In adding reference numerals to the components of each drawing, it should be noted that the same components are given the same reference numerals as much as possible even though they are indicated on different drawings. In addition, in the description of the present embodiments, if it is determined that a detailed description of a related well-known configuration or function may obscure the gist of the present embodiments, the detailed description thereof will be omitted.

Also, in describing the components of the present embodiments, terms such as first, second, A, B, (a), (b), etc. may be used. These terms are only for distinguishing the elements from other elements, and the essence, order, or order of the elements are not limited by the terms. Throughout the specification, when a part 'includes' or 'includes' a certain element, this means that other elements may be further included, rather than excluding other elements, unless otherwise stated. . In addition, the '... Terms such as 'unit' and 'module' mean a unit that processes at least one function or operation, which may be implemented as hardware or software or a combination of hardware and software.

DETAILED DESCRIPTION The detailed description set forth below in conjunction with the appended drawings is intended to describe exemplary embodiments of the present invention and is not intended to represent the only embodiments in which the present invention may be practiced.

This embodiment discloses a random forest-based auto-focusing apparatus and method. In more detail, the focal length of the input image is estimated using an optical system capable of adjusting the focal length of the glass substrate and a pre-trained random forest-based classification model ( An autofocusing apparatus and method for estimating and controlling are provided.

A machine learning model using a random forest may be used as a classification model for classifying categories or a regression model for inferring values for a specific dependent variable. Assuming the problem of estimating the focal length for autofocusing as a process of finding an image similar to the input image among previously stored images for each focal length, this embodiment uses a random forest-based classification model.

In the present invention, it is assumed that the subject is a patternless glass plate.

Autofocusing has an active method and a passive method. In the active method, the camera emits infrared (or ultrasonic) to measure the focal length, measures the time of the signal reflected from the subject, and calculates the distance using this. The active method has the advantage that it can be used in places where there is insufficient light regardless of the amount of light. On the other hand, if there is a transparent glass between the camera and the subject, or if there is a subject at a far distance beyond the range limit that infrared (or ultrasonic) can reach, it may be difficult to adjust the focus.

The manual method is a method of adjusting focus using light reflected from a subject, and there are a phase difference detection method and a contrast detection method.

The phase difference detection method, which is used in most single lens reflex (SLR) cameras, divides the light that passes through the lens into two and compares the waveforms between the two images. The phase difference detection method uses a lens in an autofocusing detection sensor to make two images from image information of a subject that has passed through the lens, then measures the distance between the two phases, that is, the phase Calculate the extent The phase difference detection method allows for faster focusing than the contrast detection method, but has disadvantages in that it is difficult to adjust the focus when the contrast or the amount of light is small, and a separate device for measuring the phase difference is required.

In the contrast detection method, the contrast is repeatedly calculated using the focus value of a part of the subject while the lens is moved, and the point at which the contrast is maximized is determined as the point of focus. Since the contrast is calculated using the image acquired by the camera, it is also called an image processing type autofocusing method. The contrast detection method is mainly used in compact digital cameras and mirrorless cameras, etc. However, when the amount of light is insufficient, it is difficult to adjust the focus, and since the contrast is repeatedly calculated while moving the lens, the speed is slow compared to the phase difference detection method. However, the contrast detection method has advantages in that the autofocusing performance is excellent in a situation where the amount of light is secured, and a separate device for measuring the phase difference is not required, so the configuration is simple.

Meanwhile, as a focus search algorithm, there are a global search and a hill climbing search. Global search, which is the most common focus search algorithm used in the contrast detection method, calculates a focus value for all lens position areas within an inspection section, and then determines a point at which the focus value is maximum as the in-focus point. As shown in (a) of FIG. 8 , the global search method calculates the focus values for all lens position regions within the inspection section, so it is possible to check an error with respect to the focus value having a local maximum, but the search time required There is a downside to this length.

On the other hand, as shown in (b) of Figure 8, the hill climbing search uses the fact that the graph of the focus value is distributed in the form of a hill. This method can reduce the time required compared to the global search, but has a disadvantage in that, if the starting point is set incorrectly, the maximum value may be misjudged as the original focus.

Hereinafter, this embodiment uses a contrast detection method based on global search in the learning stage of the random forest-based classification model.

1 is an exemplary diagram of an auto-focusing apparatus according to an embodiment of the present invention. In an embodiment of the present invention, the auto-focusing apparatus 100 uses a pre-trained random forest-based classification model to determine a focus position for an input image. Estimate and adjust

The auto-focusing apparatus 100 includes all or a part of an optical unit 103 , a classification model 104 , a control unit 105 , and a Z-axis driving unit 106 . Here, the components included in the auto-focusing apparatus 100 according to the present embodiment are not necessarily limited thereto. For example, the auto-focusing apparatus 100 may further include an objective lens 101 and a camera 102 for acquiring an input image for the classification model 104 from the subject. That is, the image for the region of interest (ROI) of the subject is taken by the image sensor of the camera 102 after passing through the object lens 101, and the classification model 104. transmitted to the side. In addition, a training unit (not shown) for learning the classification model 104 may be additionally provided on the auto-focusing apparatus 100 , or may be implemented in the form of interworking with an external training unit.

In the case of patterned glass, it is possible to calculate the focus value by using the contrast difference of the field of view at the boundary of the pattern according to the focus state. In the case of a glass substrate without a pattern, it is difficult to calculate a focus value using the contrast difference because there is no contrast difference between the focused state and the non-focused state. In order to perform autofocusing on a glass substrate without a pattern, the image of the region of interest should be able to include the contrast of the boundary according to the state of focus.

An optical microscope is composed of an objective lens that is an image formation lens for magnification observation, a light source irradiating light to an observation object, and a lens for illumination. used in light microscopy The illumination method is Kohler illumination. Kohler illumination is a method for realizing bright and uniform illumination in the entire area of the subject, and the illumination light source also forms an image inside the optical system. An image of illumination is induced to form an image on the rear focal plane of the objective lens, and it becomes possible to irradiate the entire subject with uniform illumination, so that most optical microscopes employ Kohler illumination.

The optical unit 103 according to the present embodiment provides Kohler illumination to the subject.

2 is a conceptual diagram of an optical microscope according to an embodiment of the present invention.

As shown in FIG. 2 , the subject is enlarged or reduced in the image pickup device through the lens, which is referred to as imaging. Since the position of the field stop as well as the position of the subject and the image pickup device is the image forming position, the field stopper (field diaphragm) determines the range of the subject to be imaged. That is, if the object is positioned on the field stop, it is possible to identify the object in the image pickup device. 2 , an aperture diaphragm determines the brightness and depth of an image by controlling the amount of light incident on the lens.

In the case of a microscope employing Kohler illumination, after reducing the field stop, the user can manually adjust the focus by moving the condenser up and down until the blade of the field stop is clearly visible. By applying this method, the autofocusing apparatus 100 according to the present embodiment uses the region including the blade of the field stop as the region of interest in learning and verifying the random forest-based classification model 104 . That is, learning and verification of the classification model 104 are executed based on the contrast provided by the blade of the field stop.

The random forest-based classification model 104 according to the present embodiment estimates a focal length using the obtained ROI image. The autofocusing apparatus 100 estimates the focal length of the subject by inputting the image of the blade region of the field stop to M (M is a natural number) decision tree included in the pre-trained classification model 104 . can do.

The random forest method can be widely used for classification problems and regression problems, and shows excellent performance with only a few parameter adjustments (see Non-Patent Document 1). Since the random forest method is based on several independent trees, it is possible to increase the processing speed by using a parallel structure. However, if the size of the bootstrap sample for each tree is large, it may be difficult to directly execute parallel learning on the tree. In addition, since the random forest includes repeated sampling, a bootstrap method adapted for a large-scale online situation in addition to parallel processing may be considered.

M decision trees included in the random forest-based classification model 104 according to the present embodiment may be trained in advance to estimate the focal length using the training image. Details of the learning process will be described later.

The controller 105 according to the present embodiment generates a control signal for adjusting the position of the optical microscope in a direction to reduce the difference between the estimated focal length and the in-focus distance, and provides it to the Z-axis driving unit 106 . In addition, the control unit 105 generates an illumination control signal for controlling the intensity of the Kohler illumination provided by the optical unit 103 and transmits it to the optical microscope using the Z-axis driving unit 106 . Here, it is assumed that the optical microscope including the objective lens 101 , the camera 102 , and the optical unit 103 is attached to the Z-axis driving unit 106 .

The Z-axis driving unit 106 according to the present embodiment executes a final autofocusing function on the subject by adjusting the vertical position of the optical microscope. To this end, the Z-axis driving unit 106 may include a motor and a motor driving unit for vertical position adjustment.

1 is an exemplary configuration according to the present embodiment, and other components or differences between the components according to the fixing method of the subject, the objective lens, the camera, the optical unit, the classification model, the control unit and/or the Z-axis driving unit are different. Implementations involving connections are possible.

It is assumed that the auto-focusing apparatus 100 according to the present embodiment is mounted on a programmable system. It is assumed that the programmable system acquires data for an input image from an optical microscope including an objective lens, a camera, and an optical unit using a wired or wireless transmission method.

Hereinafter, a learning process for the random forest-based classification model 104 will be described.

3 is a conceptual diagram of a random forest-based classification model according to an embodiment of the present invention.

As shown in FIG. 3 , the training unit constructs a bootstrap group by randomly and redundantly sampling n (n is a natural number) data from the entire training image dataset, but generates M groups. After learning a random binary decision tree for each bootstrap group, the training unit proceeds to combine all M trees in the final step.

The training image dataset can be created by global searching the search range near the focal point. In addition, each training image includes the blade of the field stop so that the contrast provided by the blade of the field stop can be used for learning the decision tree.

로 나타낸다. 여기서

은 M 개의 트리 각각에 대한 독립 랜덤 변수이다. 각 트리를 생성하기 전, 중복 샘플링을 실행하여 학습용 데이터세트로부터 j 번째 부트스트랩 그룹을 생성하기 위해 변수

가 이용될 수 있다. In the random forest-based classification model 104 composed of M trees, the predicted value of the j-th tree for the input image x is

is indicated by here

is an independent random variable for each of the M trees. Before creating each tree, we run redundant sampling to create the j-th bootstrap group from the training dataset.

can be used

은 학습용 데이터로서,

는 독립 랜덤 변수에 기반하여 생성될 수 있는 i 번째 영상의 특징(feature)으로써 특징의 개수는 영상의 해상도(resolution)와 동일하다.

는 i 번째 영상의 초점거리이다.

is data for learning,

is a feature of the i-th image that can be generated based on an independent random variable, and the number of features is equal to the resolution of the image.

is the focal length of the i-th image.

The training unit may train the j-th tree according to Equation 1 to estimate the estimated value of the input image x.

는 트리 형성 전에 선택된 데이터세트를 의미하며,

은 x를 포함하는 전체 집합이고,

은

에 해당하는 픽셀의 수를 의미한다. 수학식 1에 따르면, 트레이닝부는 입력 이미지의 특징과 j 번째 트리의 학습에 이용된 부트스트랩 그룹의 각 구성요소 간의 유사 확률을 산정한 후, 각 구성요소 별로 산정된 확률과 각 구성요소에 대응되는 초점거리의 곱을 결합하여, j 번째 트리의 예측값을 추정한다. here

means the dataset selected before tree formation,

is the whole set containing x,

silver

is the number of pixels corresponding to . According to Equation 1, after calculating the similarity probability between the features of the input image and each component of the bootstrap group used for learning the j-th tree, the training unit calculates the probability calculated for each component and the corresponding probability for each component. By combining the products of focal lengths, the predicted value of the j-th tree is estimated.

For all M trees, by combining the estimated values of each tree according to Equation 1 as in Equation 2, the classification model 104 can estimate the focal length.

4 is a flowchart of an autofocusing method according to an embodiment of the present invention.

The autofocusing apparatus 100 according to an embodiment of the present invention acquires an image of a region of interest (ROI) for a subject from an optical microscope (S401). The autofocusing apparatus 100 uses a blade region of a field stopper or field diaphragm as a region of interest. An optical microscope may generate an image of a region of interest using Kohler illumination.

The auto-focusing apparatus 100 estimates a focal length by inputting the image of the ROI into a random forest-based classification model (S402). The autofocusing apparatus 100 inputs the image of the blade region of the field stop to M (M is a natural number) decision tree including the pre-trained classification model 104, and the focal length of the subject length) can be estimated.

The autofocusing apparatus 100 generates a control signal for adjusting the Z-axis position of the optical microscope using the estimated focal length (S403). The autofocusing apparatus 100 may generate a control signal for adjusting the Z-axis position of the optical microscope in a direction to reduce the difference between the estimated focal length and the in-focus distance. In addition, the autofocusing device 100 provides an illumination control signal for controlling the intensity of Kohler illumination to the optical microscope.

The autofocusing apparatus 100 adjusts the vertical position of the optical microscope based on the control signal for adjusting the Z-axis position (S404).

The auto-focusing apparatus 100 checks whether the calculated focal length is estimated within the allowable depth range ( S405 ). When it is estimated that the depth of field of the optical microscope is within the allowable range, the autofocusing apparatus 100 ends the autofocusing process.

On the other hand, when the depth is not estimated within the allowable range, the autofocusing apparatus 100 repeats steps S401 to S404. By iteratively executing the autofocusing process, it is possible to reduce the difference between the estimated focal length and the in-focus range to within the allowable depth of field.

Hereinafter, an experimental example for showing the performance of the auto-focusing apparatus 100 according to the present embodiment will be described.

In the experimental example, a Sentech STC-CMB2MPOE was used as a camera for image acquisition. This camera is an Ethernet interface-based camera, and it is possible for the autofocusing apparatus 100 to acquire an image by directly connecting the camera to the computer side. As a lens, an M Plan APO 5X lens, an objective lens of Mitutoyo, was used. As a light source, white LED (Light Emitting Diode) coaxial box spot lighting was used, and an LED lighting controller controlled in 255 steps for lighting brightness control was used. For Z-axis motion control, Fastech's Ezi-Servo Plus-R motor driver and EzM42 series motor were used.

㎛ 내에 위치하는 경우 오토포커싱이 되었다고 가정한다. The cell size of the used Ethernet camera is 5.5 μm, and when the aperture stop value of the 5 magnification objective is set to 17.86 for image magnification, the depth of the acquired image is 47.15 μm. In this experimental example, since images were acquired while moving in units of 10 μm, a depth range of -20 μm to +20 μm was set as the autofocusing target. There was no significant difference in sharpness between images acquired at the in-focus (0 μm) and images taken at -20 μm, -10 μm, +10 μm, and +20 μm within the depth range. So, at the starting point

It is assumed that auto-focusing is performed when it is located within μm.

㎛,

㎛,

㎛ 범위 인식률은 목표값과 추정값 간의 차이가 각각 허용 범위 내로 인식된 영상 개수의 비율로 정의한다. 실험예에서는 각 초점거리별로 구성된 데이터에 대해서 정 인식률,

㎛ 범위 인식률,

㎛ 범위 인식률,

㎛ 범위 인식률로 구분하여 본 실시예의 성능이 평가되었다.The static recognition rate is defined as the ratio of the number of images in which the target value and the estimated value coincide with respect to the data configured for each focal length. Also

μm,

μm,

The μm range recognition rate is defined as the ratio of the number of images in which the difference between the target value and the estimated value is within an acceptable range, respectively. In the experimental example, the positive recognition rate,

μm range recognition rate,

μm range recognition rate,

The performance of this example was evaluated by dividing by the recognition rate in the μm range.

In the experimental example, 21 images were acquired per set while moving in increments of 10 µm over the search range from -100 µm to +100 µm from the in-focus (0 µm) on the glass substrate. In the entire image as shown in Fig. 5 (a), the blade region of the field stop as shown in Fig. 5 (b) was selected as the region of interest and used for learning and verification. Even at the same location, fine changes may occur in the image due to changes in lighting and mechanical changes, so in this experimental example, images of the same location were acquired several times and used for learning the random forest-based classification model 104 .

Table 1 shows the composition of the dataset used for training and validation.

As shown in Table 1, the training dataset was repeatedly acquired 500 times while moving by focal length in the search range to generate a total of 500 sets. In addition, for the performance analysis of the classification model 104, an image not used for learning and an image from another nearby location were obtained at the location where the training data was obtained, and a verification dataset was generated. The verification set 1 is image data obtained at the same location as the training data, and the verification sets 2 and 3 include images obtained at a different location from the training data.

As described above, a total of 10,500 ROI images for a total of 21 positions, 500 for each focal length, were used as training data. Each region of interest image has a resolution of 80x40, and 3,200 feature elements can be generated. When the tree is grown in the learning process, the stopping criterion is set so that the depth of the tree does not exceed a maximum of 5, and the minimum number of learning images reaching a separation node is limited to be more than 10. In the experimental example, as a result of executing training using 500 training datasets, a total of 63 trees were generated.

For the random forest-based classification model 104 according to the present embodiment, learning was performed by setting the focal length of the input image as a target value. When a new image is input, the pre-trained classification model 104 may calculate a focal length estimation value. Therefore, if the optical microscope is moved by the difference in distance from the estimated focal length to the in-focus, it can be positioned in the in-focus.

Validation was performed on set 1, which was acquired from the same location as the training data but was not used for training. As a result of verification using 2,100 images composed of 100 sets, the recognition rate for the entire section was 100%. Therefore, it was confirmed that the reproducibility of the random forest-based classification model 104 is high.

6 is a graph illustrating a verification result for the verification set 2 according to an embodiment of the present invention.

㎛ 범위 내로 추정된 인식률은 100 %였다. 이는 심도 허용 범위 이내이므로, 해당 초점거리 구간에서는 본 실시예에 따른 오토포커싱 장치(100)의 초점거리 추정 성능이 우수함이 확인되었다.Verification was performed using 2,100 images of set 2 for verification, which consists of 100 sets. As shown in FIG. 6, the recognition rate for the entire section was 51.05%, which was a low ratio, but considering the focal length -70 μm to +70 μm section,

The recognition rate estimated within the μm range was 100%. Since this is within the allowable depth range, it was confirmed that the focal length estimation performance of the autofocusing apparatus 100 according to the present embodiment was excellent in the corresponding focal length section.

7 is a graph illustrating a verification result for the verification set 3 according to an embodiment of the present invention.

㎛ 범위 내로 추정된 인식률은 100 %였다. 이는 심도 허용 범위 이내이므로, 해당 초점거리 구간에서는 본 실시예에 따른 오토포커싱 장치(100)의 초점거리 추정 성능이 우수함이 확인되었다. Verification was performed using 2,100 images of set 3 for verification consisting of 100 sets. As shown in FIG. 7, the recognition rate for the entire section was 33.86%, which was a low ratio, but considering the focal length -70 μm to +70 μm section,

The recognition rate estimated within the μm range was 100%. Since this is within the allowable depth range, it was confirmed that the focal length estimation performance of the autofocusing apparatus 100 according to the present embodiment was excellent in the corresponding focal length section.

As described above, according to this embodiment, the focal length of the input image using an optical system capable of adjusting the focal length of the glass substrate and a previously learned random forest-based classification model. By providing an autofocusing apparatus and method for estimating and adjusting a focal length, it is possible to quickly focus an image acquired within a predetermined distance from an in focus.

Also, according to the present embodiment, by repeatedly applying a random forest-based classification model to an image acquired at a distance from the original focal point, it is possible to increase the recognition rate of the focal length.

Although it is described that each process is sequentially executed in each flowchart according to the present embodiment, the present invention is not limited thereto. In other words, since it may be applicable to change and execute the processes described in the flowchart or to execute one or more processes in parallel, the flowchart is not limited to a time-series order.

Various implementations of the systems and techniques described herein may be implemented in digital electronic circuitry, integrated circuitry, field programmable gate array (FPGA), application specific integrated circuit (ASIC), computer hardware, firmware, software, and/or combination can be realized. These various implementations may include being implemented in one or more computer programs executable on a programmable system. The programmable system includes at least one programmable processor (which may be a special purpose processor) coupled to receive data and instructions from, and transmit data and instructions to, a storage system, at least one input device, and at least one output device. or may be a general-purpose processor). Computer programs (also known as programs, software, software applications or code) contain instructions for a programmable processor and are stored on a "computer-readable recording medium".

Various implementations of the systems and techniques described herein may be implemented by a programmable computer. Here, the computer includes a programmable processor, a data storage system (including volatile memory, non-volatile memory, or other types of storage systems or combinations thereof), and at least one communication interface. For example, a programmable computer may be one of a server, a network appliance, a set-top box, an embedded device, a computer expansion module, a personal computer, a laptop, a Personal Data Assistant (PDA), a cloud computing system, or a mobile device.

The above description is merely illustrative of the technical idea of this embodiment, and a person skilled in the art to which this embodiment belongs may make various modifications and variations without departing from the essential characteristics of the present embodiment. Accordingly, the present embodiments are intended to explain rather than limit the technical spirit of the present embodiment, and the scope of the technical spirit of the present embodiment is not limited by these embodiments. The protection scope of this embodiment should be interpreted by the following claims, and all technical ideas within the scope equivalent thereto should be interpreted as being included in the scope of the present embodiment.

100: autofocusing device 101: objective lens
102: camera 103: optical unit
104: classification model 105: control unit
106: Z-axis driving unit

Claims (14)

A random forest-based classification model that estimates a focal length by acquiring an image of a region of interest (ROI) for a subject from an optical microscope, where the classification model is contains a plurality of decision trees;
a control unit generating a control signal for adjusting a position of the optical microscope in a direction to reduce a distance difference between the focal length and an in focus; and
Z-axis driving unit for adjusting the position of the optical microscope by using the control signal
including,
The classification model is
Autofocusing, comprising calculating a similarity probability between the image of the region of interest and a training image, and generating an estimate value of each of the plurality of decision trees by using a product between the similarity probability and a focal length corresponding to the training image ( auto-focusing) devices. According to claim 1,
The optical microscope is
an objective lens passing the image of the subject;
a camera that generates an image of the region of interest; and
Optics providing Kohler illumination including light source, illumination lens, aperture diaphragm and field diaphragm
Auto-focusing device comprising a. According to claim 1,
The optical microscope is attached to the Z-axis driving unit, auto-focusing device, characterized in that the position is adjusted in the vertical direction 3. The method of claim 2,
The area of interest is
Auto-focusing device, characterized in that the area including the blade (blade) of the field stop. 3. The method of claim 2,
The control unit is
Auto-focusing apparatus, characterized in that further generating an illumination control signal for adjusting the intensity of the Kohler illumination, and providing the illumination control signal to the optical microscope using the Z-axis driving unit. 3. The method of claim 2,
The classification model is
Auto-focusing apparatus, characterized in that the plurality of decision trees are learned in advance based on the learning image. 7. The method of claim 6,
The classification model is
Auto-focusing apparatus, characterized in that the focal length is estimated by averaging the estimated values of each of the plurality of decision trees. delete 7. The method of claim 6,
The learning video is
The autofocusing apparatus, comprising images of the same region as the region of interest, and being collected based on a global search method in a search range near the focal point. 7. The method of claim 6,
The classification model is
Autofocusing apparatus, characterized in that it is learned in advance to estimate the focal length by using the contrast (contrast) provided by the blade of the field stop. In the auto-focusing method used by the auto-focusing device,
a process of acquiring an image of a region of interest (ROI) for a subject from an optical microscope;
A process of estimating a focal length by inputting the image of the region of interest into a pre-trained random forest-based classification model, wherein the classification model includes a plurality of decision trees ) including;
generating a control signal for adjusting a position of the optical microscope in a direction to reduce a distance difference between the focal length and an in focus; and
The process of adjusting the vertical position of the optical microscope using the control signal
including,
The process of estimating the focal length is,
Autofocusing method comprising calculating a similarity probability between the image of the region of interest and a training image, and generating an estimate value of each of the plurality of decision trees by using a product between the similarity probability and a focal length corresponding to the training image . 12. The method of claim 11,
When the focal length is an image that is not estimated within an allowable depth of field range of the optical microscope, the autofocusing method is repeatedly performed. 12. The method of claim 11,
Autofocusing method, characterized in that it further comprises the step of generating an illumination control signal for controlling the intensity of Kohler illumination used in the optical microscope and transmitting the signal to the optical microscope. A computer program stored in a computer-readable recording medium to execute each step included in the auto-focusing method according to any one of claims 11 to 13.

Images