Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

EDCSSM: Edge Detection with Convolutional State Space Model

Qinghui Hong, Haoyou Jiang, Pingdan Xiao, Sichun Du, Tao Li

TL;DR

This work introduces EDCSSM, a convolution-based discrete-time state-space edge detector that learns image edge information through a two-operator SAIM framework and refines results with a Wind Erosion post-processing step. It couples this method with a memristor crossbar accelerator to achieve real-time performance (≥30 FPS on 5K images), while delivering thin, continuous edges and strong noise suppression across multiple datasets. The approach benefits from minimal down-sampling, full-scale texture feature handling, and hardware acceleration, though it faces challenges in complex scenes and post-processing parameterization. Overall, EDCSSM offers a new, efficient pathway for high-precision edge detection suitable for resource-constrained devices and real-time applications.

Abstract

Edge detection in images is the foundation of many complex tasks in computer graphics. Due to the feature loss caused by multi-layer convolution and pooling architectures, learning-based edge detection models often produce thick edges and struggle to detect the edges of small objects in images. Inspired by state space models, this paper presents an edge detection algorithm which effectively addresses the aforementioned issues. The presented algorithm obtains state space variables of the image from dual-input channels with minimal down-sampling processes and utilizes these state variables for real-time learning and memorization of image text. Additionally, to achieve precise edges while filtering out false edges, a post-processing algorithm called wind erosion has been designed to handle the binary edge map. To further enhance the processing speed of the algorithm, we have designed parallel computing circuits for the most computationally intensive parts of presented algorithm, significantly improving computational speed and efficiency. Experimental results demonstrate that the proposed algorithm achieves precise thin edge localization and exhibits noise suppression capabilities across various types of images. With the parallel computing circuits, the algorithm to achieve processing speeds exceeds 30 FPS on 5K images.

EDCSSM: Edge Detection with Convolutional State Space Model

TL;DR

This work introduces EDCSSM, a convolution-based discrete-time state-space edge detector that learns image edge information through a two-operator SAIM framework and refines results with a Wind Erosion post-processing step. It couples this method with a memristor crossbar accelerator to achieve real-time performance (≥30 FPS on 5K images), while delivering thin, continuous edges and strong noise suppression across multiple datasets. The approach benefits from minimal down-sampling, full-scale texture feature handling, and hardware acceleration, though it faces challenges in complex scenes and post-processing parameterization. Overall, EDCSSM offers a new, efficient pathway for high-precision edge detection suitable for resource-constrained devices and real-time applications.

Abstract

Edge detection in images is the foundation of many complex tasks in computer graphics. Due to the feature loss caused by multi-layer convolution and pooling architectures, learning-based edge detection models often produce thick edges and struggle to detect the edges of small objects in images. Inspired by state space models, this paper presents an edge detection algorithm which effectively addresses the aforementioned issues. The presented algorithm obtains state space variables of the image from dual-input channels with minimal down-sampling processes and utilizes these state variables for real-time learning and memorization of image text. Additionally, to achieve precise edges while filtering out false edges, a post-processing algorithm called wind erosion has been designed to handle the binary edge map. To further enhance the processing speed of the algorithm, we have designed parallel computing circuits for the most computationally intensive parts of presented algorithm, significantly improving computational speed and efficiency. Experimental results demonstrate that the proposed algorithm achieves precise thin edge localization and exhibits noise suppression capabilities across various types of images. With the parallel computing circuits, the algorithm to achieve processing speeds exceeds 30 FPS on 5K images.
Paper Structure (15 sections, 7 equations, 8 figures, 10 tables, 1 algorithm)

This paper contains 15 sections, 7 equations, 8 figures, 10 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Edge Detection
  4. Discrete-time State Space Model
  5. Accelerating Strategy for Edge Detection
  6. The State Space Model for Edge Detection
  7. Circuits Accelerator for EDCSSM
  8. Experiments on Edge Detection
  9. Experiments Datasets
  10. Performance Metrics
  11. Implementation Details
  12. Comparison with State-of-the-art
  13. Ablation Study
  14. Simulation of Crossbar Circuits Accelerator
  15. Conclusion

Figures (8)

  • Figure 1: Examples of edge detection. Our method, EDCSSM, learns and extracts precise edges at all scales in images through state space variables. (a, e): Input images from BSDS500 ref_dataset_BSDS500. (b, f): Detected edges by EDCSSM. (c, d, g, h): Zoomed-in patches.
  • Figure 2: The framework of the proposed EDCSSM.
  • Figure 3: Parallel analog computation circuits for SAIM
  • Figure 4: Crossbar circuits for one-step convulution
  • Figure 5: Performance Comparisons on BSDS500 dataset with previous state-of-the-arts.
  • ...and 3 more figures