JPH07249123A - Item identification system - Google Patents

Abstract

(57) [Summary] (Correction) [Purpose] To be able to automatically judge whether the selected tablets match the designated brand. Furthermore, it enables accurate identification regardless of the position and orientation when the tablet is supplied. [Configuration] The camera 1 captures a color image of the tablet, converts the image into a gray image, and further the tablet area in the image Is corrected to a fixed position and direction, and a feature vector is extracted from the image. The gray value vector of the image is used as this feature vector. This feature vector and dictionary file 110
The distance from the reference vector of the designated brand stored in is calculated, and it is determined that the tablet belongs to the designated brand based on the magnitude of this distance. In the position and direction correction, a dynamic binarization method is used to determine a threshold value for the contrast evaluation value for each small area obtained by dividing the gray image of the tablet, and this small threshold value is used to binarize each small area. As a result, the area of the identification code written on the tablet surface is extracted and the direction of the tablet is corrected.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an article identification system for identifying an article using image processing / identification technology, and particularly in an automatic packaging machine for pharmaceutical tablets and capsules (hereinafter simply referred to as tablets). The present invention relates to an article identification system suitable for tablet identification for inspecting whether a selected tablet is correct for an input name pattern.

[0002]

2. Description of the Related Art At present, in many general hospitals, an automatic tablet packing machine for automatically packing a plurality of tablets of a designated brand in a packaging paper is used. The automatic tablet packing machine receives a plurality of tablet brand information inputted by the operator from the terminal, carries the corresponding tablets from the doublet case in the apparatus, and performs a process of packing the tablets in the same packaging paper. For general automatic tablet packaging machines, see, for example, “Toshikazu Seino, Takao Orii, et al .: Use of a computer for dispensing operations-Trial of an audit system for automatic dispensing packaging machines, Proc. Of the 8th Medical Information Union Conference. , 565-568, 1988 ".

By introducing a drug system using such an automatic tablet packing machine into a hospital, the tablet packing which has been performed by a pharmacist until now is automated, and the waiting time is greatly shortened. It becomes possible and the quality of medical treatment can be improved.

[0004]

In the above automatic tablet packing machine, a mechanical operation is included in the process of taking out a specified tablet from the tablet case and the process of carrying the taken-out tablet to a position for packaging. Therefore, there is a possibility that the tablets may be taken out incorrectly or missing. In addition, although the human beings supply the tablets to the tablet case, there is a possibility that the brand to be supplied at that time may be wrong. However, even if there is such a mechanical or human error, it is necessary for the package to be finally provided to the patient to definitely contain the tablet of the designated name. Therefore, conventionally, when operating an automatic tablet packaging machine, a pharmacist has visually performed an identification test to determine whether the packaged tablet is a designated brand.

Therefore, the labor of the identification test is troublesome,
There was a problem that the inspection took time. Further, since the inspection is performed after the packaging, if the wrong packaging of tablets is found, the packaging has to be broken and the packaging of correct tablets has to be made again.

Therefore, it is an object of the present invention to automatically determine whether or not a selected tablet matches a designated brand in the operation of the automatic tablet packing machine, and thereby, the automatic tablet packing machine always specifies the selected tablet. The goal is to get the correct packaging out of the street tablets.

The more general purpose of the present invention is to
In applications such as selecting items in a specified category from a multi-category item group, an identification test is performed to confirm whether the selected item belongs to the specified category. It is to be able to do it automatically and accurately using technology.

Further, a secondary object of the present invention is to enable an accurate identification result to be obtained regardless of the posture and position of the sorted article when it is supplied to the image processing apparatus. Especially.

[0009]

SUMMARY OF THE INVENTION An article identification system according to the present invention comprises an image input means for inputting an image of an article, a category input means for inputting a category of an article, and a constant article area in the input image. Position / orientation correction means for correcting the position and direction of the image, feature extraction means for extracting the gray value vector of the image for which the position and direction has been corrected as a feature vector, and the extracted feature vector and a reference vector of an input category prepared in advance. And an identification unit that determines whether or not the input article is related to the input category by determining the distance between the input item and the distance.

Here, when correcting the position and direction of the article area, a predetermined mark area existing in the article area is used in the correction of at least one of the position and the direction. For example, in the preferred embodiment, the area of the identification code present on the tablet surface is utilized to correct the orientation of the tablet area within the image taken of the tablet. That is, the direction of the tablet area is corrected so that the identification code area faces a certain direction.

When the predetermined stamp area is used for correcting the position or direction, it is necessary to extract the predetermined stamp area from the article area. In the extraction processing, the dynamic binarization method is used. it can.

The dynamic binarization processing method is a processing method in which a target image is divided into a plurality of small areas and a binarization threshold value is set for each of these small areas to perform binarization.

The contrast evaluation value can be used when the predetermined mark area is extracted from the article area by using the dynamic binarization method. The contrast evaluation value is a numerical value representing the degree of contrast at each place in the image.

Further, when the threshold value for extracting the predetermined mark area is set using the contrast evaluation value, the zero point method based on the appearance frequency of the contrast evaluation value can be used. The zero point method is a term uniquely used in this specification, but for example, in the case of a contrast evaluation value, the frequency of appearance of the contrast evaluation value is scanned from the smallest and the appearance frequency is initially zero. This is a method of setting the contrast evaluation value at the point of failure as a threshold.

[0015]

In the present invention, an article (for example,
An image of the tablet) and the category of the article (eg, tablet brand) are input to the system. Then,
First, the article region in the image is corrected to a fixed position and direction, and then the feature vector of the article is extracted from the corrected image. A gray value vector having gray values in the image as elements is used as the feature vector. The distance between the feature vector and the reference vector prepared in advance is obtained, and it is determined whether or not the input category corresponds to the article depending on whether or not the distance is within a predetermined allowable value.

By carrying out the above-mentioned processing, it becomes possible to automatically carry out the identification inspection of the article without the visual inspection of a person. Moreover, since the position / orientation of the article in the image is constantly corrected, it is not necessary to strictly adjust the position / orientation of the tablet with respect to the camera when the image is captured.

Further, since a predetermined mark in the article is used in the position correction or the direction correction of the article, even if there are various types of the outer shape of the article, different correction processing is prepared for each outer shape. Even if it does not, one kind of common processing algorithm can cope with a certain kind of article.

For example, in the preferred embodiment, an identification code on the tablet surface is utilized to effect tablet orientation correction. Although there are various outer shapes of tablets, generally, each tablet has an identification code character string written in the center thereof. Therefore, if the direction is determined by relying on the identification code, the direction of almost any tablet can be corrected. In this way, even for articles other than tablets, the position and direction of the article can be corrected by a common processing algorithm by using some mark having an image feature.

If a dynamic binarization method is used when extracting the predetermined mark from the article area, it becomes easy to remove unnecessary areas other than the mark and extract only the mark skillfully. In particular, if the surface of the article is not a flat surface but a curved surface, or if the surface of the article is colored with different colors or tones, the shade of the surface itself will differ depending on the location. In order to extract, it is effective to set a unique binarization threshold for each small area at a different location by the dynamic binarization processing method.

Further, if the contrast evaluation value is used when extracting the predetermined mark, it becomes easy to extract a mark of a type that can be easily identified by the naked eye. That is, when a person looks at an object and grasps its shape or pattern, one of the image-like elements that most contributes to the grasping is the contrast. Therefore, if the contrast evaluation value is used for the image processing, it becomes easy to skillfully extract the constituent elements of the article that a person will see and grasp the article. As a result, especially when a person intentionally tries to extract the mark from an article on which the mark is intentionally written (for example, a tablet with an identification code), the mark should be placed on the surface of the article so that it is easy to grasp with the naked eye. On the other hand, since it should have a large contrast, such marks can be extracted well.

Further, when the zero point method based on the appearance frequency of the contrast evaluation is used in the threshold setting for extracting the predetermined mark, the contrast of the surface area other than the mark such as a tablet is For articles that are extremely small compared to the area of, the extraction of the mark becomes easier. Even for articles other than tablets, the zero point method is applied unless there are circumstances where the surface irregularities are extremely severe or there are multiple colors that are greatly different in tone (this is true for many articles). Better results can be expected.

[0022]

Embodiments of the present invention will now be described in detail with reference to the drawings.

FIG. 1 shows a functional configuration of a tablet identification system according to an embodiment of the present invention.

This tablet identification system operates on a computer 100 equipped with the NTSC TV camera 1 as an external device. The computer 100 includes an image input unit 10
1, name pattern input unit 102, inverse γ correction unit 103, RGB gray conversion unit 104, position / orientation correction unit 105, feature extraction unit 106, learning unit 107, dispersion degree calculation unit 108, identification unit 109, and dictionary file 110. Prepare

The operation of this system can be roughly divided into a learning phase and an identification phase. Hereinafter, the functions and operations of each unit will be described separately for (A) learning phase and (B) identification phase.

(A) Learning Phase The system must first go through a learning phase as initialization or preparation before starting its operation. In the learning phase, for each brand of tablets to be identified, multiple samples and accurate brand names are entered,
Based on this, the reference vector and the degree of dispersion of each brand are calculated and stored as the reference vector file 10.

First, the image input section 101 inputs a color analog image obtained by photographing a tablet placed on a plain black background by the NTSC television camera 1, and converts it into an RGB digital image. Further, the brand input unit 102 inputs the brand name of the tablet into the system.

Next, the inverse γ correction unit 103 removes the influence of the γ correction applied to each of the RGB planes, and performs the inverse γ correction so that the input / output characteristic of the camera 1 becomes substantially linear. At this time, for example, 2.2 is used as the value of γ. This is because the NTSC television camera 1 is usually configured to perform γ correction of γ = 2.2 on a captured image. By performing this inverse γ correction, the contrast between the tablet region and the background region (black background) in the image is increased, and the subsequent processing can be effectively performed.

Next, the RGB gray conversion unit 104 causes the RG
Using the grayscale value of the G component of the B component as it is, RGB
Converts the image to a gray image with only grayscale components. Where G
The reason for using a component is that G
This is because the components have a wider dynamic range than the other two components, and thus it is considered that a subtle difference in white can be reflected in the light and shade in a tablet having an overwhelmingly large amount of white. By converting an RGB image into a gray image, the following processing amount can be reduced to 1/3.

Next, the position / orientation correction unit 105 performs processing so that the position and orientation of the tablet in the image will always be the same position and orientation if they have the same name. FIG. 2 shows a detailed processing flow of this position / orientation correction.

As shown in FIG. 2, first, the gray image is binarized by using a predetermined threshold value,
Only the tablet area is extracted (step 201). At this time, for example, Otsu's method based on the discrimination criterion is used as a method of selecting the binarization threshold value. This method is a very effective method for an image that can be clearly divided into two areas, a background area and a tablet area. For details, see “Nobuyuki Otsu: Automatic Threshold Selection Method Based on Discrimination and Least Squares Criterion, IEICE Transactions Vol. J63-D”.
No. 4, pp. 349-356, 1980 ".

Next, the center of gravity of the tablet area in the binarized image is obtained, and the center of gravity is moved to the center of the image (step 202). Then, the gray image is binarized in the tablet area to extract the area of the identification number code printed or stamped on the tablet surface (step 203). A dynamic binarization processing method is used as the binarization method at this time. Dynamic 2
The binarization processing method is a method in which a grayscale image to be binarized is divided into a plurality of rectangular regions (not necessarily rectangular) and a binarization threshold value is determined for each rectangular region. For details, see “Akio Shio: Dynamic Binarization Method for Detecting Characters in Scenes, IEICE Transactions Vol. J71-DN.
o. 5, pp. 863-873, 1988 ".

This dynamic binarization method is very suitable for extraction of the identification code area of the tablet for the following reasons. That is, the surface of the tablet is generally a curved surface rather than a flat surface, and the identification code is printed or stamped on the curved surface. When the surface of such a tablet is photographed, a high place such as the central portion appears relatively bright (a large gradation value), and a place such as a peripheral portion appears relatively dark (a small gradation value). As described above, the surface of the tablet is displayed in different shades depending on the place, and the identification code is printed over the different shade areas or in the bright central portion. When a static binarization method that binarizes the entire image with a single threshold value is applied to such an image of the tablet surface, the dark peripheral part is extracted together with the identification code, or the dark part is extracted. There arises a problem that a certain identification code cannot be extracted. On the other hand, according to the dynamic binarization processing method, an appropriate threshold value can be set for each rectangular area, and therefore only the identification code area can be satisfactorily extracted.

In the dynamic binarization processing method, a contrast evaluation value (index indicating the magnitude of contrast) for each pixel is obtained in each rectangular area, and a threshold value for the contrast evaluation value is set as the binarization threshold value. . In order to automatically set this threshold, a method called "zero point method" is used in this specification.

The zero point method will be described. In the zero-point method, first, the contrast evaluation value of each pixel is calculated in each square area in the tablet area, and the appearance frequency of these contrast evaluation values is obtained. Then, the appearance frequency distribution of the contrast evaluation values as shown in the histogram of FIG. 3 is obtained for each rectangular area.

Based on such appearance frequency distribution, next, a threshold value for the contrast evaluation value for separating the identification code and the tablet surface is determined as follows. That is, the appearance frequency value is scanned from the one having the smaller contrast evaluation value (the contrast is smaller), and the evaluation value of the contrast at the point where the appearance frequency becomes zero first is determined as the threshold value. In the example of FIG. 3, threshold =
It becomes 0.175.

According to this zero point method, the identification code area can be satisfactorily extracted from the tablet area. The reason is as follows. Generally, in the tablet area, the area of the mere surface of the tablet other than the identification code area (hereinafter referred to as the surface area) has a very low contrast as compared with the identification code area (that is, the contrast evaluation value is very high). small). Therefore, when the appearance frequency distribution of the contrast evaluation value in the tablet area is examined, as shown in FIG.
One large distribution peak formed by the surface area appears. Then, unevenness having a relatively sparse distribution due to the identification code region is formed in a range where the contrast evaluation value is farther away from the mountain. Therefore, if the zero point method is used, the edge of a large mountain due to the surface area is determined as the threshold value, and therefore only the identification code area in the range where the contrast evaluation value is larger than this threshold value can be satisfactorily extracted.

After the identification code area has been extracted as described above, the second moment of the extracted identification code area is then obtained, and the direction in which the second moment becomes maximum coincides with the X axis of the image. , The entire image is rotated by affine transformation (step 204).

By correcting the position and direction of the tablet area in this manner, the identification accuracy described later is improved. This is because the appearance characteristics of tablets depend on the position and the direction. In addition, since the same tablet always has the same position and direction, the exact position of the tablet with respect to the camera 1 at the time of shooting
No need to adjust orientation.

Referring again to FIG. 1, next, the feature extraction unit 1
06 performs processing for extracting the feature vector of the tablet from the position / orientation-corrected gray image. FIG. 4 shows a detailed processing flow of this feature extraction. In addition, FIGS. 5 to 7 show image examples for explaining the process of feature extraction step by step.

First, as shown in FIG. 6, a rectangular area 500 of a predetermined size is cut out from the gray image 400 whose position and orientation are corrected as shown in FIG. 5 (FIG. 4, FIG. 4). Step 301). Here, the rectangular area 5
The size of 00 is determined by a statistical method as a size that enables the most accurate feature extraction based on the result of performing feature extraction on tablets of various brands in advance.

Next, as shown in FIG. 7, the rectangular area 5
00 is divided into small sections 600 of arbitrary size N × M (step 302). Here, the size N of the small section 600
× M is determined in proportion to the processing capacity of the computer.

Next, for each small section 600, a representative value is calculated from the gray values of all the pixels (step 303). The median is used as the representative value, for example. Even if the distribution of gray values is obviously biased like the tablet area and the background, the median value is
This is because the entire distribution can be well reflected.
Then, a grayscale value vector having the representative value of each small section 600 as an element is extracted as a feature vector (step 30).
4).

In the learning phase, a large number of learning samples of various brands of tablets to be identified are supplied to this system, and the above-described processing from photographing to feature extraction is performed on each learning sample.

Referring again to FIG. 1, in the learning section 107,
For each issue, principal component analysis is performed on the total gray value vector extracted from many learning samples of that issue to obtain the eigenvalue and the eigenvector. Then, the contribution ratio is calculated based on the eigenvalue, several principal components are selected from the one having the highest contribution ratio, and the above-mentioned eigenvector is used to convert the above all gray value vector into the principal component vector. As described above, by selecting a significant feature based on the result of the principal component analysis, the dimension of the feature vector can be reduced, and the amount of processing thereafter can be reduced. Next, the obtained main component vector is averaged for each issue to be a reference vector, and this reference vector is stored in the dictionary file 110 as a reference vector of the issue input from the issue input unit 102. Further, the eigenvector of each issue is also stored in the dictionary file 110.

From the distribution of the distance between the reference vector and the principal component vectors of a large number of learning samples, the dispersion degree calculation unit 108 obtains the standard deviation σ when the average is 0, for each issue. Here, the distribution of the distance of the learning sample with respect to the reference vector is as shown in FIG. 8, for example. The standard deviation of such distance distribution is calculated for each stock,
This is stored in the dictionary file 110 as the dispersion degree of the brand input from the brand input unit 102.

Hereinafter, the reference vectors and the degree of dispersion of all brands stored in the dictionary file 110 are collectively called a dictionary.

(B) Identification Phase In this face, the correctness of the correspondence is inspected for the tablet and the name pattern which are not known whether they correspond. That is, using the dictionary created in the learning phase, an identification test is performed to determine whether or not the tablets supplied to the system correspond to their brands. Here, the tablet to be identified and the brand name are input to the system.

Referring to FIG. 1, first, with respect to the input tablets, starting from the camera photographing and the input of the brand name as in the learning phase, the image input section 101 to the feature extraction section 10 are started.
A series of image processing up to 6 is performed and the gray value vector is extracted.

Next, the identification unit 109 uses the eigenvector obtained in the learning phase to convert the grayscale value vector into a principal component vector. Next, the distance between this principal component vector and the reference vector corresponding to the input brand is calculated. At this time, Euclidean distance, similarity, etc. are used for the distance function. A general distance function is described, for example, in “Noboru Funakubo: Visual Pattern Processing and Recognition, published by Keigaku Shuppan, 1990”.

By the way, in the present embodiment, since the gray value vector is used as the basic feature vector, the distance calculation process is performed by superimposing the reference image and the identification target image to calculate the degree of coincidence. It is none other than. Such a feature vector has hardly been used in the conventional pattern identification system, but it is effective when a large category is to be processed because it can maximize the use of image information.

Further, the identification unit 109 compares the distance thus obtained with an allowable value r.sigma. Determined from the dispersion degree (standard deviation) .sigma.
If it is within σ, it is determined that the tablet is a tablet that corresponds to the input brand, and if not, it is determined that it is a tablet of another brand that does not correspond, and the determination result is output.
It is desirable to utilize this result in the automatic tablet packing machine for selecting whether to pack the identified tablet, discharge it without packing, or return it to the original tablet. Alternatively, all tablets are packaged regardless of the identification result, but the identification result may be printed on the surface of the packaging paper and checked by the pharmacist later.

Here, the coefficient r which determines the allowable value r · σ can be arbitrarily set by the user. For example, r
= 1.0, the identification standard becomes strict and tablets that do not correspond to the brand can be easily repelled. However, even in the case of tablets that actually correspond, scratches on the tablet surface, When the distance from the reference vector is large due to a lack or other reasons, it is determined that the tablet does not correspond and is repelled. On the contrary, when the allowable value is large such as r = 3.0, the result opposite to the above example is produced. In this way, the behavior of this system varies greatly depending on the allowable value. Therefore, it is necessary for the user to set an appropriate allowable value depending on the purpose, for example, to repel tablets that are slightly different while giving priority to safety.

As described above, in the tablet identification system of this embodiment, the brand and tablet identification inspection can be performed without relying on human visual inspection.

The present invention is not limited to the above embodiments, but can be implemented in various different modes without departing from the scope of the invention. For example, not only tablets
It can also be used for identification of selected parts in automated production lines.

[0056]

According to the present invention, it is possible to automatically identify an article. In particular, high identification accuracy can be obtained regardless of the posture or direction of the article supplied to the system.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of a tablet identification system according to an embodiment of the present invention.

FIG. 2 is a flowchart showing processing of a position / orientation correction unit in the embodiment.

FIG. 3 is a histogram for explaining a zero point method used for determining a threshold value for separating a tablet surface area and an identification code area.

FIG. 4 is a flowchart showing processing of a feature extraction unit in the same embodiment.

FIG. 5 is a diagram showing an example of a gray image after applying a position / orientation correction unit in the embodiment.

FIG. 6 is a diagram showing a rectangular region cut out from the image of FIG. 5 in the same example.

FIG. 7 is a diagram showing a state in which the image of FIG. 6 in the embodiment is divided into small areas.

FIG. 8 is a diagram showing an example of a distance distribution of learning samples with respect to a reference vector in the embodiment.

[Explanation of symbols]

 1 NTSC TV camera 100 Computer 101 Image input unit 102 Brand input unit 103 Inverse γ correction unit 104 RGB gray conversion unit 105 Position / orientation correction unit 106 Feature extraction unit 107 Learning unit 108 Scattering degree calculation unit 109 Identification unit

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification number Office reference number FI technical display location 7459-5L G06F 15/70 320

Claims (5)

[Claims] 1. An article identification system for determining whether an article input to the system is related to an input category, and an image input means for inputting an image of the article, and a category input means for inputting the category of the article. A position / orientation correction unit that corrects the article area in the input image to a fixed position and direction; a feature extraction unit that extracts a grayscale value vector of the position and direction corrected image as a feature vector; The distance between the extracted feature vector and the reference vector of the input category prepared in advance is obtained, and whether the distance is within a predetermined allowable value or not Identifying means for determining whether or not the position / orientation correcting means exists in the article area. An article identification system which extracts a region and corrects at least one of a position and a direction of the article region by using the predetermined mark region. 2. The article identification system according to claim 1, wherein the position / orientation correction means uses a dynamic binarization processing method to extract the predetermined mark area from the article area. Product identification system. 3. The article identification system according to claim 2, wherein the position / orientation correction means separates the predetermined marking area and the other area in the article area in the dynamic binarization processing method. An article identification system characterized by using a contrast evaluation value for this purpose. 4. The article identification system according to claim 3, wherein the position / orientation correction means determines a threshold value for separating the predetermined mark area from the other area, and the appearance frequency of the contrast evaluation value. An identification system characterized by using a zero point method based on. 5. An article identifying method for determining whether an article belongs to a designated category, the step of inputting an image of the article, and the step of correcting the article area in the input image to a certain position and direction. And a step of extracting a grayscale value vector of the position and direction-corrected image as a feature vector, and obtaining a distance between the extracted feature vector and a reference vector of the designated category prepared in advance. A step of determining whether or not the input article is related to the input category depending on whether or not the article is within a predetermined tolerance, and the step of correcting the position and the direction is the article. A step of extracting an area of a predetermined mark existing in the area, and a step of correcting at least one of a position and a direction of the article area using the predetermined mark area Article identification method characterized.