CSE-666 Programming Assignment 02

Advanced fingerprint region segmentation from hand selfie images using image processing and machine learning.

Course Details

  • Course Number: CSE 666
  • Assignment: Programming Assignment 2
  • Professor: Nalini Ratha

Team Information

  • Student: Ronak Haresh Chhatbar
  • UBName: ronakhar
  • Teammates: Ronak Haresh Chhatbar

Project Goal

The objective of this project is to segment and detect fingerprint regions from a series of hand selfie images using advanced image processing and machine learning techniques.

Project Description

This project involves several stages of development, from initial data collection to final testing of the detection algorithm:

  1. Data Collection: Curated a personal dataset of hand images with different orientations and distances.
  2. Annotation: Used LabelMe and YoloMark to annotate hand images, highlighting visible fingerprint regions.
  3. Detection: Trained a YOLOv4-tiny object detection model using Darknet to identify fingerprint regions within images.
  4. Validation: Validated the algorithm against annotated images to assess performance, achieving a mean Average Precision (mAP) of 76.71%.
  5. Testing: Tested the algorithm on an unseen dataset, demonstrating its effectiveness with a 100% mAP.
  6. Reporting: Drafted a detailed report discussing the algorithm’s design, methodology, and performance outcomes.

Model Development and Results

  • The YOLOv4-tiny model balances speed and accuracy effectively.
  • Trained on a diverse and augmented dataset for robust detection.
  • Achieved high precision, recall, and mAP scores, indicating reliable fingertip detection.

Future Scope

Potential improvements include:

  • Expanding the dataset for enhanced robustness.
  • Utilizing a more complex network like full YOLOv4 for better accuracy.
  • Optimizing the model for specific hardware for real-time applications.

References

References to open-source tools and libraries used in this project:

Conclusion

The Fingertip Detection Project demonstrates a successful approach to detecting fingerprint regions in hand selfie images, offering a foundation for future biometric identification systems.