A Subjective-Contour Generation LSI System with Expandable Pixel-Parallel Architecture for Vision Systems

📋 论文概要

该论文提出了一种用于视觉系统的主观轮廓生成LSI系统，采用可扩展的像素并行架构，旨在解决低级视觉处理任务计算密集且固定、传统媒体处理器在可编程性和性能之间权衡的问题，通过专用硬件实现高效、低成本的视觉处理。

💡 主要创新点

• 提出可扩展的像素并行架构，支持大规模并行处理，适合实时视觉任务。
• 针对主观轮廓生成这一特定视觉功能设计专用LSI，相比通用处理器在性能和功耗上更优。
• 系统架构具有可扩展性，可适应不同规模的视觉处理需求。

🏷 关键词

📄 原文摘要

Media processors that can be applied to vehicle or robot vision have been developed previously [1]. However, even using the latest media processors, the ability of artificial vision is far below that of human vision. All media processors are designed considering a trade-off between programmability (versatility) and performance, but low-level visual processing tasks are computationally intensive and fixed. Therefore, designing dedicated LSIs for such tasks are useful and cost-effective. The visual processing mechanisms in the brain are roughly illustrated as shown in Fig. 28.6.1. Since each part along the visual pathway in the brain has typical functions, developing dedicated LSIs for such functions is important. Ultimately, brain-like VLSI vision systems can be realized by combining such LSIs and digital media processors. The key to achieve brain-like artificial vision systems is the interaction between low- and

👥 作者与机构

Takashi Morie, Youngjae Kim

Kyushu Institute of Technology, Kitakyushu, Japan