KR-20260062335-A - APPARATUS FOR PROVIDING DEEPFAKE DETECTION INFORMATION, METHOD FOR PROVIDING DEEPFAKE DETECTION INFORMATION, AND MEDIUM FOR STORING A PROGRAM PROVIDING DEEPFAKE DETECTION INFORMATION

KR20260062335AKR 20260062335 AKR20260062335 AKR 20260062335AKR-20260062335-A

Abstract

One disclosed embodiment provides a method for providing deepfake detection information, comprising: acquiring a file to be analyzed from a client system; extracting a user face region included in the file to be analyzed using a face detection model; performing a deepfake detection analysis on the user face region using at least one deepfake detection model; and providing information on the analysis result of the deepfake detection analysis to the client system.

Inventors

김기홍
최민준
김건주

Assignees

주식회사 샌즈랩

Dates

Publication Date: 20260507
Application Date: 20241029

Claims (11)

A step of obtaining a file to be analyzed from a client system; A step of extracting a user face region included in the analysis target file using a face detection model; A step of performing a deepfake discrimination analysis on the user face region using at least one deepfake discrimination model; and A step of providing the analysis result information of the deepfake discrimination analysis to the above client system; including, A method for providing deepfake detection information.
In Article 1, The above extraction step is, A step of calculating at least one of the face region coordinates of the user face region and the face region coordinates of the face region included in the user face region using the above face detection model; including, A method for providing deepfake detection information.
In Article 1, The steps performed above are, A step of inputting the user face region into each of the at least one deepfake discrimination models, and calculating a fake image probability for the user face region from each of the at least one deepfake discrimination models; including, A method for providing deepfake detection information.
In Article 1, The steps performed above are, A step of calculating feature values for each of a plurality of coordinates included in the user face region using a deepfake interpretation model; A step of determining a plurality of attention coordinates among the plurality of coordinates such that the feature value is greater than or equal to a first threshold value; A step of mapping the plurality of attention coordinates above to a face region area included in the user face region; and A step of calculating deepfake interpretation information including a pixel value of at least one attention coordinate among the plurality of attention coordinates, wherein the distance value from the coordinate of the face region is smaller than a second threshold value; including, A method for providing deepfake detection information.
In Article 1, The steps provided above are, A step of providing analysis result information comprising at least one of a fake image probability for the analysis target file calculated by each of the at least one deepfake discrimination model, a heatmap image based on the pixel value of at least one attention coordinate of the user face region, and deepfake discrimination result explanation information in natural language form generated using a natural language model; including, A method for providing deepfake detection information.
A storage device for storing data; and In-memory storing a set of library engines related to software; and A processor that executes the above software; including, The above processor is, Acquire the file to be analyzed from the client system, and Using a face detection model, extract the user face region included in the above analysis target file, and Perform deepfake discrimination analysis on the user face region using at least one deepfake discrimination model, and Providing the analysis result information of the deepfake discrimination analysis to the above client system A device that provides deepfake detection information.
In Article 6, The above processor is, Calculating at least one of the face region coordinates of the user face region and the face region coordinates of the face region included in the user face region using the above face detection model, A device that provides deepfake detection information.
In Article 6, The above processor is, Inputting the user face region into each of the at least one deepfake discrimination models, and calculating a fake image probability for the user face region from each of the at least one deepfake discrimination models, A device that provides deepfake detection information.
In Article 6, The above processor is, Using a deepfake interpretation model, feature values for each of the multiple coordinates included in the user face region are calculated, and Determining a plurality of attention coordinates among the above plurality of coordinates in which the feature value is greater than or equal to a first threshold, and The above multiple attention coordinates are mapped to a face region area included in the user face region, and Calculating deepfake interpretation information including the pixel value of at least one attention coordinate among the plurality of attention coordinates, the distance value from the coordinate of the face region is smaller than a second threshold value. A device that provides deepfake detection information.
In Article 6, The above processor is, Providing analysis result information comprising at least one of a fake image probability for the analysis target file calculated by each of the at least one deepfake discrimination model, a heatmap image based on the pixel value of at least one attention coordinate of the user face region, and deepfake discrimination result explanation information in natural language form generated using a natural language model. A device that provides deepfake detection information.
Acquire the file to be analyzed from the client system, and Using a face detection model, extract the user face region included in the above analysis target file, and Perform deepfake discrimination analysis on the user face region using at least one deepfake discrimination model, and A storage medium storing computer-executable software that performs the step of providing analysis result information of the deepfake discrimination analysis to the client system.

Description

Apparatus for providing deepfake detection information, method for providing deepfake detection information, and storage medium for storing a program providing deepfake detection information The disclosed embodiments relate to a deepfake detection information providing device, a deepfake detection information providing method, and a storage medium storing a program for providing deepfake detection information. Deepfake is a compound word of deep learning and fake, referring to human image synthesis technology based on artificial intelligence. Recently, damage caused by manipulated images, video footage, or other digital representations generated by the sophisticated algorithms of this deepfake technology has been rapidly increasing. For example, deepfake videos are created to distribute fake news, and deepfake videos are also used by fraudsters or phishing criminals. In addition, sex crimes such as the distribution of fake pornography through deepfake images are also on the rise, becoming a social issue. These deepfake technologies are evolving rapidly and have reached a state where synthesized images are so realistic that they are difficult to distinguish with the human eye, and such sophistication is often difficult to detect using existing image or video analysis techniques. Furthermore, various deepfake detection technologies may reach incorrect conclusions if their detection accuracy is low or their false positive rate is high. In other words, because the accuracy of deepfake detection algorithms is not perfect, there may be instances where real images are misidentified or deepfakes are missed. Since this can have a serious impact on individuals or society, it is becoming increasingly important to enhance the reliability of the technology by presenting detection accuracy and the basis for interpreting the findings. Therefore, while interest in such deepfake detection is increasing, actual research on this is insufficient. FIG. 1 is a drawing disclosing another embodiment of a deepfake detection information providing device according to an embodiment. FIG. 2 is a drawing disclosing an example of a user interface for a deepfake analysis request according to an embodiment. FIG. 3 is a drawing disclosing an example of a process for providing deepfake identification information according to an embodiment. FIG. 4 is a drawing disclosing an example of a face detection model according to a first embodiment. FIG. 5 is a drawing disclosing an example of a face detection model according to a second embodiment. FIG. 6 is a drawing disclosing an example of a deepfake detection model according to a first embodiment. FIG. 7 is a drawing disclosing an example of an inception module of a deepfake detection model according to a first embodiment. FIG. 8 is a drawing disclosing an example of a deepfake detection model according to a second embodiment. FIG. 9 is a drawing disclosing an example of a deepfake detection model according to a third embodiment. FIG. 10 is a drawing disclosing an example of a deepfake detection model according to a fourth embodiment. FIG. 11 is a drawing disclosing an embodiment of a natural language model according to an embodiment. FIG. 12 is a drawing disclosing an example of analysis result information for a deepfake according to an embodiment. FIG. 13 is a drawing disclosing a flowchart illustrating an embodiment of a method for providing deepfake detection information according to an embodiment. FIG. 1 discloses another embodiment of a deepfake detection information providing device according to an embodiment. The deepfake determination information providing device (1000) of this drawing can acquire a file to be analyzed, determine whether it is a deepfake, and provide a determination result. In one embodiment, the deepfake discrimination information providing device (1000) may include a WEB or API (1001), a storage unit (1002), a face detection unit (1003), a deepfake discrimination unit (1004), a deepfake interpretation unit (1005), an analysis result providing unit (1006), and a storage (1007). A WEB or API (1001) can obtain a file to be analyzed from a client system (2000). In one embodiment, the file to be analyzed may include at least one of an image, a video, text, and voice. Here, the video may consist of a plurality of images corresponding to frames of the video. In one embodiment, a user's file to be analyzed uploaded through a user interface for a deepfake analysis request can be obtained. The storage unit (1002) can store the acquired target file for analysis. In one embodiment, the storage unit (1002) can store the target file for analysis transmitted via WEB or API (1001) in a queue. In one embodiment, the storage (1007) can store the target file for analysis transmitted via WEB or API (1001). The face detection unit (1003) can extract a user face region included in the file to be analyzed. In one embodiment, the face detection unit (1003) can extract a user face region corres