Hdri image, dataset generation and model training method and image sensor

CN122120636APending Publication Date: 2026-05-29FEILING MICRO (SHANGHAI) ELECTRONIC TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
FEILING MICRO (SHANGHAI) ELECTRONIC TECHNOLOGY CO LTD
Filing Date
2024-11-27
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

Existing HDR generation techniques are difficult to process quickly in real-time scenes. Multiple exposure fusion methods have high requirements for scene stability and are computationally complex, while PWL compression methods require a lot of computation and resource consumption.

Method used

By acquiring PWL compressed data in the Raw domain, a deep learning model is used to directly output the initial HDR image in the Raw domain, skipping the traditional decompression process and restoring the image details lost due to compression.

Benefits of technology

It enables the efficient generation of high-quality HDR images on resource-constrained devices, avoiding the resource consumption of high bit-width data computation, storage and transmission, and improving real-time processing capabilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122120636A_ABST
    Figure CN122120636A_ABST
Patent Text Reader

Abstract

The application is suitable for the image technical field, and provides an HDR image generation method, a data set generation method, a model training method and an image sensor. The image generation method comprises the following steps: acquiring PWL compressed data in a Raw domain; inputting the PWL compressed data into a deep learning model to obtain an initial HDR image in the Raw domain output by the deep learning model. The application can effectively restore image details lost due to compression, and avoids resource consumption caused by high-bit-width data calculation, storage and transmission caused by PWL data decompression.
Need to check novelty before this filing date? Find Prior Art