Image Processing
At LINE Yahoo, we handle an enormous volume of image and video data across a variety of services, including advertising, e-commerce, video SNS, and news media. Aiming to enhance service quality and deliver new user experiences, we are dedicated to the research and development of image and video processing technologies.
-
Advertisement Image - Image Analysis
We are researching and developing a variety of image processing technologies for analyzing advertising images. Beyond commonly used technologies like OCR (Optical Character Recognition), we are also dedicated to developing specialized techniques for analyzing advertising images, such as those for estimating the impression ads make on users and for analyzing layouts. We leverage these technologies to streamline the review process for the vast amount of advertising data submitted daily, to predict ad performance at the time of delivery, and to analyze ad designs.
-
Advertisement Image - Layout Optimization
To continuously deliver effective advertisements, it is essential to streamline the creative production process. At LY Corporation, we offer creative production services such as LINE Creative Lab. Furthermore, we are engaged in research and development aimed at further improving the advertising production experience through generative AI technologies. In particular, we are researching layout generation, which involves determining the placement of multiple elements such as images and text. By proposing the optimal layout tailored to the provided materials, we support efficient creative production.
-
E-commerce Image Retrieval
We are engaged in the research and development of similar image search technology for product images in e-commerce services. Aiming to provide new user experiences and enhance convenience, we develop and offer a system that allows users to search for visually similar or identical products from a vast product database using images as queries. The similar image search system primarily relies on two technologies: object detection and feature extraction. We are advancing the development of models that aim for robust search capabilities across various types of query images (such as product images or snapshots) and shooting scenes.
-
Video Processing - Video Classification
We are researching and developing machine learning-based video classification and filtering technologies that enable the analysis of video advertisements and the automatic verification of terms and conditions for user-submitted videos. In this effort, we are pursuing the development of robust modeling techniques that can ensure practical performance levels, even when there is an extreme shortage of labeled video data, which is a common challenge in the industry. Furthermore, we are advancing research and development in data preprocessing techniques that focus on video data characteristics to achieve efficient modeling. This modeling maintains performance while reducing temporal and spatial computational complexity compared to conventional video modeling approaches.
-
Video Processing - Video Retrieval
We are researching and developing video search technologies that enable users to quickly and accurately find the videos they want from the vast amount of content uploaded daily. In this research, we focus on two main areas: (i) enhancing video retrieval accuracy using composite features that combine visual content with audio and sound features, and (ii) developing video tagging and text generation for indexing to facilitate high-speed searches. We are also advancing the development of scene retrieval technologies that allow for the fast and precise extraction of specific scenes within videos. These research and development efforts aim to improve the user experience and the quality of search services.
-
Multi-view Image
We provide a technology that performs projection transformation on a series of images taken from multiple angles, creating images that appear as if the subject was photographed on a turntable. By implementing this technology, we enable an interactive experience on web pages, where users can virtually touch and rotate the subject. This technology is indeed utilized for the product and advertising images in the services provided by LY Corporation.
-
Super Resolution
The quality of images viewed in a browser is a critical element that significantly affects the impression of a platform. We are advancing the development of practical super-resolution technology, using our proprietary technique with diffusion models, to transform images containing various types of noise into high-quality images. Through the enhancement of images contained in web pages, we aim to provide users with a visually more beautiful information platform.
