Metadata generation system and metadata production method

The metadata generation system addresses the challenge of creating a database with natural-sounding descriptions for connected car images by using a processing unit and large-scale language model to analyze traffic scenes and generate descriptive metadata, improving search accuracy by including relevant objects based on traffic scenarios.

JP2026113541APending Publication Date: 2026-07-07HITACHI LTD

Patent Information

Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
HITACHI LTD
Filing Date
2026-03-30
Publication Date
2026-07-07

AI Technical Summary

Technical Problem

Conventional image caption generation systems for connected cars do not effectively create a database of vehicle images associated with natural-sounding descriptions that match user search queries, failing to consider the presence or absence of relevant objects based on traffic scenes, which hinders accurate image retrieval.

Method used

A metadata generation system that includes a processing unit generating metadata from images using a table with columns for object information and descriptive text, utilizing a large-scale language model to analyze traffic scenes and generate natural-sounding descriptions, and a necessity table to determine which objects to include or exclude based on traffic scenarios.

Benefits of technology

Enables the creation of a database with natural-sounding descriptions that improve search accuracy by including or excluding relevant objects based on traffic scenes, enhancing the matching of user search queries and improving image retrieval efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2026113541000001_ABST
    Figure 2026113541000001_ABST
Patent Text Reader

Abstract

This involves creating a database of images associated with natural-sounding descriptions that match search queries entered by humans. [Solution] The image description system 100 has a large-scale language model unit 13 that generates metadata from images for media retrieval using a recognition table 14B which includes a first column and a second column. The first column of the recognition table 14B, the column for the object to be described, is set with the names of multiple objects. The second column of the recognition table 14B, the column for the detail items, is set with information that affects the content of the description text for multiple objects.
Need to check novelty before this filing date? Find Prior Art