README.md

Edge Detection

Sobel edge detection algorithms implemented in Python for grayscale images.

This project intends to serve as a model for future Stochastic Computing applications.

Progress

• Deterministic
  • Software
  • Hardware
• Stochastic
  • Software
    • LFSR
    • Module
    • Wrapper
  • Hardware

Quick-start

All commands consider a Linux environment

Deterministic implementation

cd ./sw-sim
python interactiveDetSobel.py
src, edges = detectAndShow('320px-1000_years_Old_Thanjavur_Brihadeeshwara_Temple_View_at_Sunrise.jpg')

This will calculate an edge image in a few seconds and plot the result in a new window. It also returns the source image as a numpy array (src) and the 8-bit edges array. Expected image:

Stochastic implementation

cd ./sw-sim
python interactiveStochSobel.py
src, edges = detectAndShow('320px-1000_years_Old_Thanjavur_Brihadeeshwara_Temple_View_at_Sunrise.jpg')

This version takes more time than the deterministic version since Python types and code have not been optimized to do the bit-wise operations that Stochastic Computer takes advantage of. Expected image:

Parallel processing with Ray

If available, you may opt to use

python interactiveRayStochSobel.py

instead. This version defaults to starting 8 threads to process images, which is about 8 times faster than the single-threaded version on modern CPUs (which already come with 8 threads).

Remarks:

• Deterministic implementation:
  • Works as OpenCV's Sobel example without image blurring
• Image database for testing from "CURE-OR: Challenging Unreal and Real Environment for Object Recognition", IEEE Dataport, 2019. [Online]. Available: http://dx.doi.org/10.21227/h4fr-h268. Accessed: Nov. 11, 2019.
• Sample image file 320px-1000_years_Old_Thanjavur_Brihadeeshwara_Temple_View_at_Sunrise.jpg from Wikimedia Commons. Author: KARTY JazZ.

Dependencies

Python dependencies:

• Python >= 3.9
• Libraries:
  • opencv-python
  • matplotlib
  • numpy
  • scipy
  • bitarray
  • ray (recommended)
    • for parallel processing in simulation
    • currently, only supported on Linux and MacOS
  • wheel (recommended for bitarray support) Hardware simulation dependencies:
• Icarus Verilog 10.1
• make (for ease of executing multiple commands)

This project currently uses stochastic circuits derived from ones synthesized with scsynth/STRAUSS

Installation recommendation:

• Newer versions of Python 3 (like 3.8.x) come with pip preinstalled. PyPi/pip is a simple package manager for Python (normally aliased as pip3)
• For this project:

pip3 install --user wheel numpy scipy matplotlib opencv-python bitarray 'ray[default]'