Assessing the Effectiveness of ICM Method for Privacy Protection (ICM手法のプライバシー保護における有効性の検証)

Recent advances in image recognition technology have created new opportunities for applying these techniques. Asimagerecognition models increasingly rely on image data, the demand for efficient image compression methods tailored for recognition tasks has grown. This area of research is known as Image Coding for Machines (ICM). In this paper, we explore the effectiveness of ICM methods in protecting privacy. With rising concerns about privacy, there is a need for image transmission and storage methods that prevent individual identification. By comparing facial features between the original images and those compressed using the ICM method, we demonstrate that the ICM approach makes it more difficult to identify individuals in the images.

In Picture Coding Symposium Japan 2024