KR101620953B1 - Image captcha providing method and server - Google Patents

Abstract

The present invention relates to a method and a server for providing an image CAPTCHA. The method comprises: a step of outputting an image CAPTCHA composed of a plurality of images; and a step of receiving image selection of the outputted image CAPTCHA. The images include: a first group image used for determining whether an object, which makes a connection attempt, solves an image CAPTCHA question; and a second group image not used for determining whether the object solves the image CAPTCHA question. By using the present invention, image identification exposure is prevented, and an image CAPTCHA providing method is able to distinguish whether an object, which makes a connection attempt, is a human or a computer. In addition, a new image is able to be automatically included in an image CAPTCHA.

Description

IMAGE CAPTCHA PROVIDING METHOD AND SERVER BACKGROUND OF THE INVENTION 1. Field of the Invention [0001]

The present invention relates to a method of providing an image capcha and a server for providing an image capcha. More particularly, the present invention provides images that are not used to determine whether a connected object is a person, And to provide an image-captcha providing method and an image-captcha providing server that enable a new image to be utilized in image captcha.

CAPTCHA (Automated Public Turing test to tell Computers and Humans Apart) is an automated turing test that is designed to determine whether a person is present or not from a remote location. Malicious users who distribute spam e-mail automatically try to obtain many e-mail accounts using robots that perform the specified tasks. Cap-cars can not be solved by robots, Stop work. Currently, the most widely used portal service is Google, Bing and other large portal services are used to identify people in the process of providing resources such as e-mail accounts.

There are a lot of controversies about the performance of cap cars that are widely used now. Commonly used character-based cap cars are known for their vulnerability and are also known for how to automatically interpret a capcha. This problem is due to the difference between human and robot in the ability to interpret characters, which is the core of character-based CAPTCHA, although CAP car has to submit a problem that only one can solve. Various techniques such as pattern distortion, background image addition, and color change are applied to the character-based CAPTCHA, but this causes a situation in which it is difficult for the person to solve the problem. Rather, it is pointed out that robots can solve problems more easily than people.

To improve the problem of character-based CAPTCHA, an image-based CAPTCHA emerged. The image-based cap car is mainly a problem of interpreting the meaning of the image, and the ability of interpreting the meaning of the image can make an ideal capcha because the difference between human and robot is remarkable.

There is a limit to the problem that the early image - based cap car, which presents the problem of image interpretation as a problem, can not be created quickly. This is because human interpretation of the meaning of the image is possible only by the person, and the human intervention is also necessary at the time of submitting the problem. In the early days, the image-based cap car would inevitably have a small total number of problems, and the problem-consuming attack using the computing power of the robot (the method of finding the right answer to the image by utilizing the wrong experience before) It was possible to pass an image-based capcha.

A way to automatically create problems to overcome the limit of problem submission speeds has also been devised. These latter image-based cap cars typically present the problem of locating specific patterns in distorted or cropped images. As mentioned earlier, the superiority of human beings in interpreting images is not effective in distinguishing human and robot because it is not finding the pattern of images.

Thus, there is a need for a method of providing an image capcha and a server for providing an image capcha to overcome the limitations of the initial image-based capcha, which presents the meaning interpretation of images as problems, so that the number of problems can be automatically extended without human intervention.

SUMMARY OF THE INVENTION The present invention has been made in order to solve the above-mentioned problems, and it is an object of the present invention to provide a method and apparatus for analyzing a selection pattern of an image-based capcha, which can add a new image to be used for image analysis automatically without human intervention, An image captcha providing method and an image captcha providing server.

According to another aspect of the present invention, there is provided an image captcha providing method comprising: outputting an image captcha composed of a plurality of images; and receiving an image selection of an output image captcha, The first group of images used for the passing determination and the second group of images not used for the passing determination of the image-capturing problem of the connected object.

In order to achieve the above-mentioned object, an image captcha providing server includes an interface for outputting an image captcha composed of a plurality of images and receiving an image selection of an output image captcha, Wherein the plurality of images comprise a first group of images used for passing determination of the subject of image capping problem and a second group of images not used for passing determination of the subject of image capping problem of the connected subject.

The image captcha providing method and image captcha providing server according to the present invention can add a new image to be used for image analysis automatically without human intervention by analyzing the selection pattern of the image-based capcha, .

1 is a diagram showing an example of a system for providing an image capcha.
2 is an exemplary block diagram of an image-captcha-providing server.
3 is a diagram illustrating an exemplary control flow for providing an image capcha.
Figure 4 is an illustration of an exemplary flow for determining the identity of a collected image and for redetermining a behavior pattern.
5 is a diagram showing an example of an image captcha that is output to the connection target device and displayed on the connection target device.
6 is a diagram showing an example of a problem passing method of an image capcha of the conventional system.
7 is a diagram showing an example of a problem passing method of an image capcha according to the present invention.
8 is a diagram showing an example of classifying behavior patterns through a clustering technique.
Figure 9 shows a typical error pattern that can be automatically determined in the reconstruction of the tree.

The above and other objects, features and advantages of the present invention will become more apparent from the following detailed description of the present invention when taken in conjunction with the accompanying drawings, in which: It can be easily carried out. In the following description, well-known functions or constructions are not described in detail since they would obscure the invention in unnecessary detail. Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a diagram showing an example of a system for providing an image capcha.

Referring to FIG. 1, an image-captcha providing system includes at least one connection target apparatus 200 and an image-captcha providing server 100. The connection target device 200 and the image capcha providing server 100 are preferably configured to be able to transmit and receive various data through the Internet.

A brief description of each system block through an example of an image captcha providing system is as follows. The connection target apparatus 200 is a device capable of connecting to a server or another apparatus through an Internet network and transmitting / receiving various data through the Internet network. The connection target device 200 may be, for example, a personal computer, a mobile phone, a smart phone, a tablet PC, a server for a specific use, and the like. The connection target device 200 is a device that can connect to at least the image captcha providing server 100 and can use a specific function provided by the image captcha providing server 100. [

For example, the connection target apparatus 200 may be a device used by a specific person (person) to newly join the image captcha providing server 100 or may be a device that is installed to maliciously utilize the function of the image capcha providing server 100 Called " robot " Malicious intentional robots are programmed to analyze images of image capcha through problem consuming attacks and to acquire multiple accounts through analysis.

The image captcha providing server 100 provides the subscription or registered user with functions according to the use of the server. The image captcha providing server 100 may be, for example, an e-mail server, a portal server, a cloud server, or the like. The image captcha providing server 100 according to the present invention is configured to output an image captcha and to determine whether the connection target apparatus 200 is a person or not by selecting images of the outputted image captcha. The image captcha providing server 100 is configured to provide an account to a user and to provide a specific function or service when an image captcha problem is passed. In addition, the image-captcha-providing server 100 is configured to determine whether a person is a behavior pattern according to image selection.

The connection target device 200 and the image capcha providing server 100 can transmit various request and response data according to a communication protocol on the Internet protocol through the Internet network. The connection target apparatus 200 transmits a request for account acquisition to the image captcha providing server 100 and the image captcha providing server 100 provides an image captcha composed of a plurality of images in response to the request. The image captcha providing server 100 determines whether or not the requested connection subject has passed the trouble and the person by receiving data or response indicating selection and selection in the connection target apparatus 200 with respect to the provided image captcha, You can provide an account.

The image captcha providing server 100 may have various interfaces with the connection target apparatus 200. [ For example, the image captcha providing server 100 may operate as a web server that provides a web page. The image-captcha providing server 100 is configured to transmit a web page to the connection target apparatus 200 and receive data according to a selection among the images displayed on the web page.

The image captcha generated according to the present invention and the specific control and data flow using the image captcha will be described in detail with reference to the following drawings.

2 is an exemplary block diagram of an image capcha providing server 100. In Fig. 2, the image captcha providing server 100 includes at least an interface unit 101, a storage unit 103, a control unit 105, and a connection unit 107, and may further include other blocks. The block diagram of Figure 2 preferably shows a functional block diagram.

2, the interface unit 101 is connected to the Internet network and transmits / receives data to / from each of the one or more connection target apparatuses 200. [ The interface unit 101 may transmit data indicating image captcha to the connection target apparatus 200 and receive data indicating selection of images included in the image captcha from the connection target apparatus 200, for example. The interface unit 101 for data transmission / reception includes a chipset capable of wired local communication (for example, Ethernet communication) or a chipset capable of wireless local area communication (e.g., Wi-Fi communication).

The storage unit 103 stores various data. The storage unit 103 includes at least one mass storage medium such as a volatile memory, a non-volatile memory, and / or a hard disk. The storage unit 103 temporarily stores various data and programs using the volatile memory, and stores system settings and system programs (boot programs, etc.) using the nonvolatile memory. The storage unit 103 stores various images, various log data, and various programs using a mass storage medium such as a hard disk.

The images stored in the storage unit 103 are used for constructing or generating an image capcha. These images may be images that have been determined to be correct or incorrect for a particular problem, or images that have been collected through the Internet but have not yet been determined as correct or incorrect for a particular problem. Images may be included in the image cap. The programs stored in the storage unit 103 include a program for providing a function according to the use of the server, a program for generating image capcha, image management, and log data management. The program related to image captcha generation is called, for example, when an account is created, and is configured so that an image captcha problem can be transmitted to the connection target device 200 according to the call.

The control unit 105 controls each block of the image captcha providing server 100. The control unit 105 includes an execution unit capable of executing the code of the program. The control unit 105 is preferably configured to load the code of the program stored in the storage unit 103 and control each block according to the program. The control unit 105 is configured to include so-called one or more CPUs, MPUs, and the like.

The control unit 105 can transmit or receive various data to or from the connection target apparatus 200 in cooperation with the interface unit 101. [ The control unit 105 can perform the function assigned by the image captcha providing server 100 and can perform the function according to a request from the connection target device 200. [ In this process, the control unit 105 needs to determine whether or not the connection target device 200 is a person. For example, when creating a new account, the control unit 105 generates an image capcha to determine whether the connection target apparatus 200 is a person, and outputs the generated image capcha through the interface unit 101. The control unit 105 receives the selection of the connection target apparatus 200 for the image captcha through the interface unit 101, determines whether the connection target apparatus 200 is a person in accordance with the received selection, And provides the account to the connection target apparatus 200.

The control flow performed by the control unit 105 will be described in detail with reference to FIGS. 3 and 4. FIG.

The connection unit 107 transmits and receives data between the blocks. The connection unit 107 is configured to transmit data of another block to the control unit 105 under the control of the control unit 105 or to transmit data generated by the control unit 105 to another block. The connection portion 107 is constituted by, for example, a parallel bus.

Each block of the image captcha providing server 100 of FIG. 2 may be embedded in a single device or may be embedded in a plurality of devices. Even the control unit 105 may be embedded in a plurality of instruments. According to the structure of the image providing server 100, the shape of the connecting part 107 may be different.

3 is a diagram illustrating an exemplary control flow for determining a behavior pattern to be used in providing an image capcha. The control flow of FIG. 3 is performed by the image captcha providing server 100. The control unit 105 of the image captcha providing server 100 preferably loads the program stored in the storage unit 103 and controls the interface unit 101 and the storage unit 103 according to the code of the program, Flow is performed. Here, the control and data flow interlocked among the respective blocks will be described in detail. The determined behavior pattern is used to provide subsequent image captcha.

First, the image captcha providing server 100 constructs a problem of seeded images and image captcha by intervention of an administrator of the server 100 (S101). For example, an administrator or the like generates a problem of image captcha, collects images to be used as seeds through on-line / off-line, etc., and stores them in the storage unit 103. If the seed images are correct images or incorrect images . In step S101, whether the right or wrong answer of each seed image is determined and classified through semantic analysis of each seed image by involvement by a person, that is, an administrator.

The control unit 105 of the image captcha providing server 100 configures an image captcha according to a request to create a (user) account of the connection target device 200 through the interface unit 101 (S103). The constructed image cap difference further comprises a problem comprising a plurality of images and being a selection criterion for a plurality of images. The plurality of images are preferably selected from the sorted seed images.

First, all of the plurality of images of the image captcha according to the present invention are configured such that identification of the correct answer or wrong answer image is impossible. That is, the configured image cap difference is a difference between the images of the group (hereinafter referred to as " first group ") used for determining whether or not the problem of the connection target apparatus 200 has passed, Images (hereinafter also referred to as " second group "). The image of the second group is hereinafter also referred to as a " honeypot image ".

Unlike the known image capcha, the image cap difference according to the present invention includes at least the images of the first group and the images of the second group so that the identity of each image can be hidden even if passed. The honeycomb image of the image capcha according to the present invention does not participate in the status of correct answer or wrong answer. The honeypot image may be an already misclassified image or a correct image. Even if the connection target device 200 passes the image capcha problem, the honeypot image can not determine the correct answer or wrong answer of each of the images. Regardless of the selection of the honeypot image, it is impossible to determine the identity of the images and the problem consuming attack can be prevented because the passage of the image is determined by the image of some of the plurality of images (the first group of images).

That is, as the honeypot image is included in the image captcha, the robot, which is the connection target device 200, can pass the problem even with the wrong selection of the honeypot image among the selected images. The honeypot image can be either correct or wrong, but does not affect the passing of the problem. The robot can then make the same selection for the image (previously the honeypot image but may be subsequently selected as the first group of images) by correct or incorrectly mistaking the honeypot image, .

Because the honeypot image disturbs attempts to infer the identity of the image, and because the choice does not affect the outcome, it prevents the pattern distortion as discussed below and is used to determine the identity of the newly collected image.

After the configuration of the image capcha, the control unit 105 outputs the image captcha to the connection target apparatus 200 via the interface unit 101 (S105). The output image cap difference includes the problem and the plurality of images, and the plurality of images includes the first group image (s) and the second group (honeypot) image (s).

5 is a diagram showing an example of an image captcha that is output to the connection target apparatus 200 and displayed on the connection target apparatus 200. [ As can be seen from FIG. 5, the connection target apparatus 200 receives the image captcha through the Internet network and displays the received image captcha. The user of the connection target apparatus 200 or the robot of the connection target apparatus 200 selects a plurality of images included in the image capcha according to the problem and transmits data representing the image selection to the image captcha providing server 100 through the Internet network send.

The control unit 105 receives data indicating image selection through the interface unit 101 (S107). The control unit 105 determines whether the image captcha problem has passed according to the reception of the image selection. In accordance with the determination, the control unit 105 creates an account and provides an account to the connection target device 200 or refuses to create an account.

Further, the data representing the image selection is stored (S109) by the control unit 105 in the storage unit 103 as log data. The data stored in the log data includes data indicating a problem, an identifier of a plurality of images, whether or not to select an image of each identifier, an IP (Internet Protocol) address of the connection target device 200, . Log data is collected by IP address. The IP address is an exemplary identifier capable of identifying a specific connection target device 200. The specific access target device 200 can be connected to the image-captcha providing server 100 several times as necessary, and accordingly, a plurality of image captcha output and reception histories can be stored in the log data. The pass-through data can be stored for each image capcha output, from which the pass probability can also be stored as log data.

FIG. 6 shows a problem passing method of an image capcha of the conventional method, and FIG. 7 shows a problem passing method of an image capcha according to the present invention.

As can be seen in FIG. 6, the existing problem of passing through the known image capcha is passed unless all correct answers are selected and all the wrong answers are selected. If it is passed, the identity of all the outputted images is exposed to the connection target apparatus 200.

On the other hand, as can be seen in FIG. 7, the images of the first group are used for passing the problem with the selection of correct or incorrect images and the images of the second group (honeypot) are not used for determining the problem. Therefore, even if the honeycomb image is selected incorrectly, the connection target apparatus 200 can pass through and the identity of the images can not be exposed as a result of the passage. The number of images of the first group is, for example, 14 or more. The first group of 14 images can be prevented from accidental passage by a robot or a random algorithm.

Steps S103 to S109 are repeatedly performed. The images of the image capcha outputted through steps S103 to S109 may be different from the images of the previous image capcha. Also, images of the first group of previous image captcha may be output as a honeypot image at a subsequent output, and conversely, a honeypot image may be output as an image of a first group at a subsequent output. When sufficient log data is stored in the storage unit 103, the control unit 105 determines a behavior pattern of a specific type of connection target apparatus 200 or a person using the log data (S111).

The determination of behavior patterns, in particular the determination of a person's behavior pattern, makes it possible to add collected images. If the particular access target device 200 passes the problem and is determined to be the same as a person's behavior pattern, the identity of the collected unknown image can be determined and the collected image can then be utilized as a new image in the image capcha .

The decision of the behavior pattern consists of two stages. The first step is to classify patterns that show common features using stored log data by clustering method and the second step is to designate one of the patterns classified without help by human as a behavior pattern . The two steps are performed by the control unit 105 using the log data stored in the storage unit 103 and preferably by using the image data of the common log data Clusters the patterns. The behavior pattern may hereinafter be referred to as a selection pattern and the behavior pattern preferably indicates at least a common selection pattern of images selected by the classified connected devices 200. [

Figure 8 shows an example for the first step. As can be seen from FIG. 8, the control unit 105 uses the IP address of the log data and the used images to display the image of the root node (FIG. 8 (A) of FIG. The four IPs on the left are the IPs chosen as the correct answer for the root node image and the three IPs on the right are the IPs chosen as incorrect. In addition, these two groups construct a tree by repeatedly (recursively) selecting images that can be divided into two groups until no more differently selected IPs exist in each group. The four groups (s3, s4, s5, s6) of the tree (b) of FIG. 8 are in such a state that they can no longer divide the tree because all the images appearing in the problem are answered the same.

The tree is used in the classification through the clustering technique. The reason for using the tree is to reduce the number of images required to discriminate a behavior pattern of a person. Learning a human behavior pattern in a tree form, it is sufficient to divide a group of people and other behavior patterns into only the image that exists in the path from the root node to the person group. For example, if the group s5 is identified as a person, it is possible to distinguish between groups of people and non-people by including an image existing in the path up to the group s5. The advantage is that the depth of the tree increases as the number of groups increases. The deeper it is, the more useful it is.

To maximize the benefits of using a tree, it is important to reduce the depth of the tree and to create a balanced tree of the root or other nodes. Accordingly, the tree according to the present invention utilizes the entropy used in the ID3 decision tree to select an image of a root node or a child node.

Basically, the known entropy formula (see http://en.wikipedia.org/wiki/ID3_algorithm)

to be. Where S represents the set of data for which the current entropy is computed, X represents the set of classes on S and p (x) represents the ratio of the number of elements in class x to the number of elements on set S. The entropy equation is iteratively (recursively) performed from the root node to the child node according to the ID3 algorithm and a particular image can be selected for each node.

The entropy equation according to the previously known equation (1) with respect to the ID3 tree needs to be modified to apply to the present invention. If image A is frequently submitted to the problem (or test) and there is log data for all IPs, but image B is rarely given to the problem, so no log data for most IPs exists, no matter how high the entropy value of image B is Even if it comes out, it does not help to divide IP that you have not experienced. Therefore, the present invention uses a transformation formula that reflects how many IPs can be divided into a reference formula showing the ability to evenly divide the two groups into an actual tree. The modified entropy equation

. Where T represents the total number of IPs in the IP group to be divided and k represents the total number of IPs that have experienced image S in the IP group to be split.

According to Equation (2), the image mapped to the root node of the tree is selected as the image having the highest entropy using the total number of IPs that have experienced the image S with respect to the number of all IPs identified through the log data. The child nodes below the root node are also mapped by selecting an image having the highest entropy in the group using the total number of IPs that have experienced a specific image S with respect to the number of all IPs in the clustered IP group for each node.

FIG. 8B shows an example of a tree constructed according to this tree structure and image mapping. As can be seen in FIG. 8 (b), the behavior pattern for each group is expressed as a tree having child nodes from the root node. Each group behavior pattern represents the selection of the first group and honeypot (second group) image (s) received via step S107 and preferably represents a clustering classification according to the selection.

After the tree is constructed, the control unit 105 designates a leaf node of a leaf node clustered according to a behavior pattern of the tree as a human behavior pattern. The control unit 105 can simply designate one specific leaf node as a behavior pattern indicating a person. The control unit 105 designates the leaf node of the IP group having the highest probability of passage or passage probability included in the log data as a behavior pattern indicating a person.

Thereafter, the control unit 105 adds an image capable of performing the same sorting function to the configured tree (S113). The control unit 105 acquires a plurality of images for each node before the leaf node of the human behavior pattern by adding an image to at least a node on the path from the root node to the leaf node of the behavior pattern representing the person.

For example, if a tree as shown in FIG. 8 (b) exists, the image can be simply mapped to a specific node using the characteristics of the tree. The image of the left child node in (b) of FIG. 8 is an image having a purpose of distinguishing (s3) from (s4). That is, all of the IPs belonging to (s3) are all picked up, and at the same time, all the IPs of (s4) do not select all the IPs commonly or all of the IPs belonging to (s3) All selected images can perform the same function. A new image may also be added to the root node as well.

Images mapped to each previous node from the root node of the tree to the leaf node representing the behavior pattern of the person are selected from the first group of images that determine whether to select the correct answer or the wrong answer, And a second group (honeypot) images. Therefore, the identification of a behavior pattern of a person is determined by the selection of the output image regardless of whether it passes or not.

A tree for discriminating a behavior pattern of a person may be stored in the storage unit 103 and then used to identify a person and add a new image.

The control flow of FIG. 3 can at least discriminate a behavior pattern of a person, and preferably a behavior pattern of a person is represented by a tree. The human behavior pattern is then used to determine the identity of the new image collected through online / offline.

4 is a diagram illustrating an exemplary flow for determining the identity of a collected image and re-determining behavior patterns. The control flow of Fig. 4 is generally performed after the execution of Fig. The control flow of FIG. 4 is performed by the image captcha providing server 100. The control unit 105 of the image captcha providing server 100 preferably loads the program stored in the storage unit 103 and controls the interface unit 101 and the storage unit 103 according to the code of the program, Flow is performed. The control and data flow associated with each block will now be described in detail. The determined behavior pattern is used to provide subsequent image captcha. In the description of the control flow of FIG. 4, a portion that overlaps with the description of FIG. 3 will be briefly described.

First, the control unit 105 collects new images periodically on-line or off-line, and stores the collected images in the storage unit 103 (S201). Step S201 may be performed independently of other steps.

The control unit 105 constructs an image captcha according to a request to create an account of the connection target device 200 through the interface unit 101 using a predetermined behavior pattern of the person in FIG. 3 (S203).

 The configured image cap difference includes a first group of images used to determine whether a problem has passed or not, and a honeypot image that does not affect the determination of whether to pass the problem. Images mapped to each node on the path from the root node to the leaf node representing the behavior pattern of the person among the plurality of images included in the image capcha are also each included. The images of the image captcha include at least one image (s) of each node on the path representing the behavior pattern of the person, and each image is designated as the image of the first group or designated as the honeypot image.

Also, the specific one or more images designated as the honeycomb images of the image capcha may be the images collected via online or offline (see step S201). The correct answer or wrong answer to the collected image is an unknown condition without human involvement. The collected image is the image before classification of the correct answer or wrong answer image. Generally, it is possible to classify by correct interpretation or wrong answer only through the interpretation of the meaning of image, and it is impossible to classify it without human involvement.

The control unit 105 outputs the image captcha including a plurality of images and problems to the connection target apparatus 200 via the interface unit 101 (S205).

The control unit 105 receives data indicating image selection from the connection target device 200 via the interface unit 101 (S207). The data representing the received image selection is stored as log data (S209) under the control of the control unit 105 in the storage unit 103 together with other data (for example, IP address, etc., see S107 and description thereof in Fig. 3) .

Thereafter, the control unit 105 determines whether the problem of the image captcha passes by the received image selection (selection of the first group of images) (S211). The images used for determining whether to pass the problem are limited to only the first group of images. Preferably, the choice of correct or incorrect answers to the first group of images, regardless of the choice of correct answer or wrong answer for the second group of images, determines whether the problem has passed.

If the problem of image captcha passes, the control unit 105 discriminates the behavior pattern of the connection target apparatus 200 (S213). For this, the controller 105 stores the behavior pattern of the image of the connection target device 200 in the storage unit 103 and compares the behavior pattern of the image of the connection target apparatus 200 with the predetermined behavior pattern of the person through FIG. It determines whether it has.

Specifically, the control unit 105 can identify a behavior pattern of a person in a tree that is stored in advance, and stores a tree of a behavior pattern stored in the storage unit 103 as a received selection for the image (s) It is possible to select a specific node by searching for a route from the root node to the child node. If the specific node is a leaf node representing a behavior pattern of a person, the control unit 105 determines that the specific node has a behavior pattern of a person.

The control unit 105 may also generate a new account according to the determination of the behavior pattern in step S211, and notify the generated account to the connection target apparatus 200 through the interface unit 101. [

The control unit 105 classifies the collected images according to the discriminated behavior pattern of the connection target apparatus 200 (S215). Specifically, when the behavior pattern that passes the problem and the image of the image of the connection target apparatus 200 is the same as (or has) the behavior pattern of a person stored in advance in the form of a tree, the control unit 105 pre- And classifies the images output as honeypot images as correct or incorrect. That is, the control unit 105 can classify and determine the status of the correct answer or wrong answer of the collected image according to reception of the correct answer or wrong answer of the connection target device 200 of the collected image. The classification of the incorrect answer or the correct answer may be determined as the reception of the image selection (S207) over a plurality of times. Accordingly, the identity of the specific collected image is determined based on a plurality of selections from the connection target apparatus 200 determined as a human pattern a plurality of times.

Steps S203 to S215 may be repeatedly performed at the request of the connection target apparatus 200. [ The specific collected image classified in step S215 may be included in the image of the first group in the subsequent iteration and output to the connection target apparatus 200. [ Through this process, it is possible to determine the correct answer or the wrong answer to the question about the collected image without the involvement of the person, as a selection of the connection target device 200 judged as a human pattern, and add a new image to the image captcha. This makes it possible to submit problems through image analysis, but also to improve the speed of submitting problems with known image CAPTCHAs.

The storage unit 103 continuously stores the log data from the plurality of connection target apparatuses 200 in accordance with the repetition of steps S203 to S215.

The continuously stored log data can then be used to redetermine behavior patterns through learning. The recurrence of a behavior pattern can be performed when a certain period of time elapses and / or when the amount of collected log data is more than the specified amount of data.

The control unit 105 re-determines the behavior pattern (S217) by learning the accumulated log data using the log data stored in the storage unit 103 when the elapse of a predetermined period or the log data is more than the specified data amount. The method of performing the recrystallization is the same as that of S111 in Fig. 3 using Equation (2). It is possible to adapt to the changing behavior pattern of the robots as the connection target apparatus 200 through the recrystallization and determine a new behavior pattern for the various images. The determined behavior pattern includes a behavior pattern indicating a person, and a behavior pattern representing a person is expressed on a tree and expressed as a path from a root node to a specific leaf node.

It is important to use behavioral (optional) patterns that have no distortion in learning to accurately classify them as people. Log data used for learning for determining a behavior pattern in step S217 in FIG. 4, and S111 in FIG. 3, specifically, images, log data (images) in the case of passing an image cap problem. The present invention also uses the honeypot image to block the attempt to hide the pattern of the specific connection target device 200. [

The robot as the connection target apparatus 200 has a unique pattern for passing the problem. If you want to hide your own pattern, you have to act differently from your algorithm. The higher the probability of success, the more likely it is that the algorithm will behave differently from its algorithm and fail the problem (test). If you randomly select an image to hide its pattern, the probability of success with only 14 images is only 0.006%.

Because there is no honeypot image in the existing image-based CAPTCHA, there is a problem that the pattern of the robot becomes the same for all people (user) when only the case that passes the problem is used for learning. If the problem is solved by log data, the correct image is submitted as a problem. If the problem is not selected, the problem can not be passed and log data not selected is excluded. If the wrong image is selected, the problem can not be passed. Are excluded.

Finally, if only successful cases are collected, it is impossible to learn behavior patterns because all result logs for the problem are the same. The honeypot image proposed in accordance with the present invention can utilize this feature only in the case of passing the problem through the learning of the behavior pattern without affecting the problem result. Even if the problem is passed, the honeypot image that is correct but the unselected honeypot image and the wrong honeypot image may exist in the log data. Since all images stored in the storage unit 103 of the image captcha providing server 100 can be submitted to a problem as a honeypot image, if only a sufficient number of times are provided, images of log data ) It is possible to collect patterns.

The honeypot image is responsible for concealing image information to all connected device 200 or users who see the problem. Unlike the existing image capcha, the problem involving the honeypot image is that the problem passing does not determine the identity of the images. Rather, it is impossible to pass the test by using the learned tree to determine the behavior pattern, quickly sorting the wrong answer image selected as the correct answer, and appearing as an incorrect image in the subsequent problem.

In order to recapitulate the behavior pattern, the control unit 105 reconstructs a tree representing a behavior pattern using a plurality of log data stored in the storage unit 103. The algorithm used for the reconstruction of the tree is the same as that of step S111 of FIG. 3 using Equation (2). The reconstructed tree replaces the existing tree constructed in Fig. The log data used for the reconstruction of the tree is preferably images when the problem is passed through step S211. According to the reorganization of the tree, the control unit 105 can designate a behavior pattern indicating a person. A behavior pattern representing a person is represented by a path on the leaf node from the root node. Images on behavior patterns representing a person may then be submitted to the first group of images or to the honeypot image.

Then, the control unit 105 adds an image capable of performing the same sorting function to the reconstructed tree (S219), as in step S113 of FIG. The control unit 105 acquires a plurality of images for each node before the leaf node of the human behavior pattern by adding an image to at least a node on the path before the leaf node of the action pattern representing the person from the root node.

Thereafter, the control unit 105 removes the error image from the mapped images at the node of the reconstructed tree (S221). For this, the controller 105 determines an error image among the mapped images of the tree and deletes the corresponding node corresponding to the determined error image from the reconstructed tree. The discrimination and deletion of the error image in the process of recalculation of the behavior pattern can also be applied in the same manner in step S113 and after in FIG.

The present invention can not always tell the result of analyzing a person's pattern because a person can discriminate a person's pattern and use the corresponding pattern for image verification, but a person can make mistakes at any time. It is necessary to find a typical tree pattern by human error and to exclude the error image.

Figure 9 shows a typical error pattern that can be automatically determined in the (re) configuration of the tree.

The two most frequent causes of errors are mistakes made by humans and ambiguous images that even people can not distinguish. 9 (a) is a real number pattern. A typical characteristic is that the IPs belonging to (s1) and (s2) represent a high probability of passage (success rate) similarly, and the root node includes a large number of images On the other hand, the right subtree shows that only a few images exist and the IP group that the image divides is overly tilted to one side (for example, left). The IPs belonging to (a) to (s2) in FIG. 9 are mostly the same in behavior pattern as (s1), but they behave differently with respect to only one image and have no other IPs behaving in the same manner. In this case, only the image (error image) of the right sub node is removed and the tree is modified by combining (s1) and (s2) into one. The error image is thus removed.

The tree of FIG. 9 (b) is a typical pattern showing a case where a problem is included in the problem that the image is ambiguous to be determined by a person. Similar to (a) in FIG. 9, (s3) and (s4) show a high level of pass probability (success rate), and the root node includes a number of images performing the same function, Corresponds to only images. Unlike Figure 9 (a), ambiguous images classify IPs evenly. 9 (b) also removes the node corresponding to the error image from the (re-constructed) tree and combines (s3) and (s4) to construct a tree from which the error image has been removed.

The two features of the above two patterns are that the number of images of the problem node is always smaller than that of the other nodes and the probability of problem passing is high. For example, the control unit 105 can remove the image and the node when the number of images of the node capable of representing the real pattern and the ambiguous image is equal to or less than the threshold, and the probability of passage of the node is higher than the probability of passage of the other node. The threshold value may be determined in advance according to the designated number or the number of images of other nodes.

The tree representing the behavior pattern after the removal of the error image is used for determining a behavior pattern indicating a person in the following steps S203 to S215. Specifically, at least one or more images of each node from a root node to a leaf node indicating a behavior pattern representing a person are included in the image captcha and output to the connection target apparatus 200. This image may be output as an image of the first group or as a honeypot image. Of course, the image of the first group and the honeypot image are not fixed in the position of the image captcha but can be variably changed, and the image of the first group and the honeypot image are outputted at different positions with each image cap discrimination outputted.

The image cap car according to the present invention makes it impossible to expose each image even if it passes through a problem including a honeypot image. Also, according to the present invention, it is possible to classify a behavior pattern of a robot and a behavior pattern of a person with the impossibility of correct answer or wrong answer classification according to a honeypot image. Further, the present invention can submit a new image collected in a behavior pattern of a person as a problem, determine the identity of a new image, and then utilize the newly collected image for a problem.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or scope of the invention. The present invention is not limited to the drawings.

100: Server providing image CAPTCHA
101: interface unit 103: storage unit
105: control unit 107:
200: Device to be connected

Claims (15)

(a) outputting an image capcha composed of a plurality of images including an image of a first group and an image of a second group;
(b) receiving an image selection of the output image capcha;
(c) storing log data indicating the image selection; And
(d) determining a behavior pattern with a plurality of stored log data,
Wherein the first group of images is used for passing determination of an image cap problem of a connection object,
Wherein the behavior pattern represents a selection of one or more images of the first group of images and the second group of images.
Wherein the behavior pattern is represented by a tree having one or more child nodes from a root node,
How to provide image capacity. delete The method according to claim 1,
Wherein a path from the root node to a designated leaf node is specified by a behavior pattern indicating a person, and the child nodes of the root node and the root node are mapped to an image,
How to provide image capacity. The method of claim 3,
Wherein an image mapped to a root node of the tree has a higher entropy than an image mapped to the child node and the entropy is calculated using the number of IPs to be connected and the number of IPs outputting each image,
How to provide image capacity. The method of claim 3,
(e) adding an image to be mapped to the node based on the log data,
Wherein the added image is capable of performing the same sorting function as the image mapped to the node,
How to provide image capacity. The method according to claim 1,
Wherein the step (c) is repeatedly performed after the execution of the step (d)
The image captcha providing method includes:
(f) re-determining a behavior pattern represented by a tree with a plurality of log data stored by the step (c) after the step (d); And
(g) determining an error image among the images mapped to the node of the tree and removing the node corresponding to the determined error image,
How to provide image capacity. The method according to claim 1,
Determining whether the image captcha passes a problem by selecting an image of the first group representing a correct answer or an incorrect answer; And determining a behavior pattern of a connection object by using a predetermined behavior pattern when passed,
How to provide image capacity. 8. The method of claim 7,
Wherein the second group of images further comprises a collected image prior to classification of the correct or incorrect image,
The image captcha providing method according to claim 1, further comprising the step of classifying the collected image into a correct answer image or a wrong answer image with a selection of the collected image by the connection target when the connected subject is determined to have a behavior pattern of a person,
How to provide image capacity. 9. The method of claim 8,
The step (a) is repeatedly performed,
Wherein an image classified into a plurality of images of the step (a) after the classification step is included as an image of the first group,
How to provide image capacity. 8. The method of claim 7,
Wherein the step of determining whether to pass the problem further comprises determining whether to pass the problem of the image captcha only by selection of the image of the first group regardless of correct or incorrect selection of the image of the second group.
How to provide image capacity. An interface for outputting an image captcha composed of a plurality of images including an image of a first group and an image of a second group and receiving an image selection of the output image capcha;
A storage unit for storing log data indicating the image selection; And
And a controller for determining whether to pass the problem of the image captcha according to the received image selection,
Wherein the first group of images is used for passing determination of an image cap problem of a connection object,
Wherein the control unit determines a behavior pattern with a plurality of stored log data, the behavior pattern representing a selection of one or more images of the first group of images and the second group of images,
Wherein the behavior pattern stored in the storage unit is represented by a tree having one or more child nodes from a root node,
Provide Image Capture Server. delete 12. The method of claim 11,
Wherein a path from the root node to a designated leaf node is specified as a behavior pattern representing a person,
Provide Image Capture Server. 12. The method of claim 11,
Wherein the control unit determines whether the image captcha passes the problem by selecting the image of the first group representing the correct answer or the wrong answer received through the interface unit, and when it passes the action pattern, However,
Provide Image Capture Server. 15. The method of claim 14,
Wherein the second group of images further comprises a collected image prior to classification of the correct or incorrect image,
Wherein the control unit classifies the collected image into a correct answer image or a wrong answer image by selection of the collected image by the connection target and outputs the classified classified image to the plurality of images if the connection target is determined to have a human behavior pattern And outputting, through the interface unit,
Provide Image Capture Server.

Images