.png){kind=icon}
返回
High-Performance Hardware/Software Co-Design for Text Extraction Using Optimized Gamma Correction in Video Decoding
.png){kind=icon}
PRE
AI
DOI：10.1002/tee.70192.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
This paper introduces an advanced Hardware/Software (HW/SW) co-design that makes it possible to extract text from decompressed video sequences in real-time, even when the videos are encoded using video coding. The design tackles the tricky computational challenges of video decoding and text extraction by embedding the Gamma Correction Method (GCM) directly into the video decoder. Furthermore, we develop an upgraded version of the GCM, known as the Optimized GCM, to enhance performance and reduce computational demands. Specialized hardware accelerators handle the computation-heavy parts of the optimized GCM. We then integrate these hardware modules as co-processors alongside the ARM Cortex-A53 processor via an AXI-stream interface. Finally, we tested the entire HW/SW system on the ZCU102 embedded kit. The proposed HW/SW co-design for the optimized GCM slashes processing time by a staggering 96% compared to a software-only implementation of the same approach. Even better, this time-saving boost doesn't sacrifice accuracy, as both methods achieve identical F-Measure values for text extraction. These findings showcase just how effective this HW/SW design is for handling the heavy computational workload of real-time text extraction from videos. They also highlight the potential of combining hardware accelerators with optimized algorithms to improve video processing systems in embedded environments. (c) 2025 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.
Keyword:
real-time text extraction
video decoder
embedded system
HW/SW co-design
FPGA
computer vision
期刊
IF:
1.1
论文数:
251
被引数:
1.8K
机构
引用论文
Scene video text tracking based on hybrid deep text detection and layout constraint基于混合深度文本检测和布局约束的场景视频文本跟踪.png)
NEUROCOMPUTING
NEUROCOMPUTINGIF6.5
PREAI
分享收藏
查看更多内容