Optimized deep learning system and method for image reconstruction
JP2026507746APending Publication Date: 2026-03-05SCIENCE COO
View PDF 0 Cites 0 Cited by
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2024-02-07
- Publication Date
- 2026-03-05
AI Technical Summary
Technical Problem
Conventional deep learning models for super-resolution image reconstruction are computationally inefficient and expensive due to the large number of residual blocks and feature maps, leading to high processing times and memory consumption.
Method used
A computationally simple deep learning-based method using optimally calculated, serially connected residual computation blocks with a Laplacian deep super-resolution model, integrating compression and convolution units, reduces the number of residual blocks and feature maps while maintaining image quality.
Benefits of technology
The method significantly reduces inference time by half and memory consumption by a factor of four to six, achieving comparable image quality with reduced complexity.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure 2026507746000001_ABST
Abstract
The deep learning network may have the following elements arranged in order: a convolution block, a first residual block (RB), a first element-wise adder, a second RB, a second element-wise adder, and an upsampling unit. The at least one processor may perform a process comprising generating a super-resolution image by processing, with the deep learning network, an image having a resolution lower than a resolution of the super-resolution image.
Need to check novelty before this filing date? Find Prior Art