KR-20260065158-A - System and method for constructing a database for copyright protection of damaged images of facilities for AI learning, and a recording medium recording a computer-readable program for executing the method

KR20260065158AKR 20260065158 AKR20260065158 AKR 20260065158AKR-20260065158-A

Abstract

A database construction system for protecting the copyright of facility damage images used for AI training includes an image information storage unit, a watermarking unit, a noise image generation unit, and a dataset construction unit. The image information storage unit stores original images of facility damage and copyright information regarding said original images; the watermarking unit watermarks the original images; the noise image generation unit generates noise images by adding noise to the original images to degrade their quality; and the dataset construction unit constructs a public dataset using the noise images when the quality of the noise images is below a preset standard. With this configuration, by watermarking facility damage images and adding noise, image copyrights can be effectively protected to reduce the risk of unauthorized use and illegal copying of data, thereby protecting the rights of data owners and promoting the distribution and utilization of data.

Inventors

김동규
이철희
김동구

Assignees

한국건설기술연구원

Dates

Publication Date: 20260508
Application Date: 20241101

Claims (13)

An image information storage unit that stores an original image of facility damage and copyright information regarding the original image; A watermarking unit that watermarks the above original image; A noise image generation unit that generates a noise image by adding noise to the original image to degrade the quality of the original image; and A database construction system for copyright protection of facility damage images for AI learning, characterized by including a dataset construction unit that constructs a public dataset using the noise image when the quality of the noise image is below a preset standard.
In claim 1, A database construction system for copyright protection of images of damaged facilities for AI learning, characterized in that the above-mentioned watermarking is performed in a form that is not identifiable by the user's visual inspection.
In claim 2, A database construction system for copyright protection of facility damage images for AI learning, characterized in that the above noise is set to degrade the learning results performed by the artificial intelligence using the above original image.
In claim 3, A database construction system for copyright protection of facility damage images for AI learning, characterized in that the intensity of the above noise is set by preset parameters.
In claim 4, A database construction system for copyright protection of facility damage images for AI learning, characterized by further including a learning result output unit that learns the original image and the noise image by artificial intelligence to produce learning results, respectively.
In claim 5, A database construction system for copyright protection of AI learning facility damage images, characterized by further including a learning degradation determination unit that determines the degree of degradation of the learning result using the learning result calculated from the original image and the learning result calculated from the noise image.
In claim 6, A database construction system for copyright protection of facility damage images for AI learning, characterized by further including a learning result output unit that outputs the learning result performed by the above noise image to a user.
In claim 7, the watermarking portion is, A meta-information insertion unit that encrypts copyright information of the original image and inserts the encrypted information into the meta-information of the image; and A database construction system for copyright protection of damaged facility images for AI learning, characterized by including a meta watermark unit comprising an encryption key storage unit that stores the key of the above-mentioned encrypted information.
In claim 8, the watermarking portion is, A database construction system for copyright protection of facility damage images for AI learning, characterized by further including a pixel watermark unit that inserts copyright information into the LSB of a pixel of the original image.
In claim 9, the watermarking portion is, Alpha channel adding unit for adding an alpha channel to the RGB mode of the original image above; and A database construction system for copyright protection of damaged facility images for AI learning, characterized by further including an image watermark unit comprising an alpha channel input unit for inputting copyright information into the alpha channel.
In claim 9, A database construction system for copyright protection of facility damage images for AI learning, characterized in that the above dataset construction unit further constructs an original image dataset using the above original image.
As a method for constructing a database for copyright protection performed by a database construction system for copyright protection, An image information storage step for storing an original image of facility damage and copyright information regarding the original image; A watermarking step for watermarking the above original image; A noise image generation step for generating a noise image by adding noise to the original image to degrade the quality of the original image; and A method for constructing a database for copyright protection of facility damage images for AI training, characterized by including a dataset construction step of constructing a public dataset using the noise image when the quality of the noise image is below a preset standard.
A recording medium storing a computer-readable program for executing the method of claim 12.

Description

System and method for constructing a database for copyright protection of damaged images of facilities for AI learning, and a recording medium recording a computer-readable program for executing the method The present invention relates to technology for constructing a database of facility damage images, and more specifically, to a data construction system and method for copyright protection of facility condition damage images for AI learning. Recently, advancements in artificial intelligence, particularly deep learning technology, are driving innovation in various fields such as image recognition, classification, and generation. In particular, in the field of image-based facility maintenance, technologies that utilize deep learning to automatically detect and diagnose defects in facilities are being actively developed. The performance of such deep learning-based facility maintenance technology depends heavily on the quantity and quality of training data. To build high-quality training data, it is important to secure high-resolution images and resolve data copyright protection issues, but interest in the relevant field is still low. Meanwhile, most images in the field of facility maintenance suffer from issues such as low resolution, noise, and motion blur due to limitations in the shooting environment. These low-quality images lead to performance degradation in deep learning models, making accurate defect detection and diagnosis difficult. Even if high-quality image datasets for training deep learning models are built by overcoming these problems and investing significant time and money, there is a high risk of unauthorized use and illegal copying due to a lack of clear regulations or technical measures for the protection of image data copyrights, which infringes upon the rights of data owners and acts as a factor hindering data distribution and utilization. FIG. 1 is a schematic block diagram of a database construction system for copyright protection of facility damage images for AI learning according to an embodiment of the present invention. FIG. 2 is a schematic flowchart of a database construction system for copyright protection of facility damage images for AI learning according to one embodiment of the present invention. FIGS. 3 and FIGS. 4 are drawings illustrating examples of an explicit watermark and an implicit hidden watermark, respectively. Figure 5 is a diagram illustrating the AES test results. FIG. 6 is a drawing illustrating an example of watermarking using LSB. FIG. 7 is a drawing illustrating an example of copyright concealment and output using a pixel watermark. FIG. 8 is a diagram illustrating an example of the output of the alpha channel of an image. FIG. 9 is a diagram illustrating a hostile attack technique. FIG. 10 is a drawing showing an example of an image with noise added by different weights. Figure 11 is an exemplary configuration diagram of a cloud-based database framework. Hereinafter, preferred embodiments of the present invention will be described with reference to the attached drawings. FIG. 1 is a schematic block diagram of a database construction system for copyright protection of facility damage images for AI learning according to one embodiment of the present invention, and FIG. 2 is a schematic flowchart of a database construction system for copyright protection of facility damage images for AI learning according to one embodiment of the present invention. In FIG. 1, a database construction system for copyright protection of images of damaged facilities for AI learning includes an image information storage unit (110), a watermarking unit (120), a noise image generation unit (130), a dataset construction unit (140), a learning result output unit (150), a learning degradation determination unit (160), and a learning result output unit (170). Additionally, the watermarking unit (120) further includes a meta watermark unit (122), a pixel watermark unit (124), and an image watermark unit (126), the metawork unit (122) further includes a meta information insertion unit (122-1) and an encryption key storage unit (122-2), and the image watermark unit (126) further includes an alpha channel addition unit (126-1) and an alpha channel input unit (126-2). The image information storage unit (110) stores the original image of the facility damage and copyright information regarding the original image, and the watermarking unit (120) watermarks the original image. In this case, the watermarking can be performed in a form that is not identifiable by the user's eyes. With this configuration, since the user cannot know whether watermarking has been performed, a user without usage rights cannot evade copyright by editing that part, etc. This configuration generates a hidden watermark on an image for AI training so that only authorized users can decrypt the image. A hidden watermark is a watermark that is not displayed on the image but is encrypted and hidden within the image data, meanin