Image-to-Netlist Conversion Pipeline

This repository contains the implementation of an Image-to-Netlist Conversion pipeline. The tool extracts circuit layouts from image representations and generates a corresponding netlist. This is a proof of concept that can be improved and implemented in circuit analysis softwares. It can be beneficial for students and professionals in electronics and hardware design domains.

Table of Contents

• Introduction
• Features
• Installation
• Acknowledgements

Introduction

As PCB designs and circuit layouts become increasingly intricate, it becomes essential to have tools that can aid in design and verification. The Image-to-Netlist Conversion pipeline takes an image of a circuit layout and translates it into a functional netlist, helping streamline the design process and reverse-engineer existing designs.

Features

1. Image Pre-processing:

• Remove the values to increase the clarity of input images for better conversion.

2. Component Detection:

• Detects components like resistors, inductors, current sources, and voltage sources.

3. Routing Extraction:

• Extracts routing information and connections between components.

Feature: Device Orientation
Path Checking: Using the midpoints of the bounding box, the function checks connectivity paths both vertically and horizontally in the image using the check_path function. Orientation Decision: Based on the results from path checks and certain assumptions, the function determines if the device is vertically or horizontally oriented.
Feature: Blocked Directions
Region Extraction: Using given x and y check points, the function extracts small regions around these points in all four directions: left, right, up, and down. RGB Averaging: It calculates the average RGB values for each of the extracted regions to determine their color characteristics. Path Blockage Check: For each direction, the function checks if the average color values exceed a threshold (200 for each channel in this case) to decide whether that direction is blocked or not.
Feature: Correct Path
Find the Correct Path: The function finds the row with the smallest average RGB values and adjusts the check_pt accordingly, either directly or based on the most frequent row with the smallest value.
Feature: Find Nodes
Traverse through all path: The loop starts with each device and continues as long as the pixel color is either black (shade of black) or red. If a different color pixel is encountered, the loop breaks. Find all the nodes: For each pixel check if the certain coordinate is in the node_list, if so add it to the node_device array with corresponding coordinate and the device.

4. Netlist Generation:

• Outputs a SPICE-compatible netlist for simulations.

Installation

For a detailed installation guide, please refer to this repository or watch our installation tutorial on YouTube.

Acknowledgements

• Nicholas Renotte: For his invaluable insights on training and setting up custom object detection model.
• OpenCV: For image processing tasks.
• TensorFlow: For providing the deep learning framework used in various components of this project.
• EasyOCR: For optical character recognition capabilities employed in this project.
• Network Analysis and Synthesis: Circuits Illustrated, authored by M.E. Van Valkenburg. This book was a pivotal resource for sourcing circuit images used in training the model. I express my sincere gratitude for the comprehensive compilation of circuit diagrams and the contribution it has made to my research.