DupliCheck

Tired of manually hunting for duplicate images across your many Chrome tabs? Meet FV Duplicate Checker—the first Chrome extension that instantly identifies visual matches across your open tabs.

Comment

Inspiration

Inspired by Dr. Alexy Bhowmick’s research on Fisher Vectors and his FhVLAD/CNN-FV repositories, I explored how Fisher Vector encoding can summarize complex images into compact representations. During the Google Chrome Built-in AI Challenge 2025, I applied this idea to solve a practical problem: detecting duplicate tabs and images in Chrome.

What it does

DupliCheck scans open tabs, captures screenshots, encodes them into Fisher Vectors, and detects potential duplicates based on similarity scores. It uses Chrome’s Prompt API to generate simple explanations like “similar shapes/colors” and lets users quickly close duplicate tabs.

How we built it

  • Used TensorFlow MobileNetV2 to extract image features from CIFAR-10 images and trained a 64-component GMM with scikit-learn.
  • Captured tab screenshots with chrome.tabs.captureVisibleTab and removed browser UI noise with canvas cropping.
  • Computed Fisher Vectors in tfjs inside a worker for fast client-side processing.
  • Used chi-squared distance scoring for duplicate detection.
  • Built a lightweight MV3 Chrome extension with a popup UI, progress tracking, and duplicate management tools.

Accomplishments

  • Built a working Fisher Vector-based duplicate detection prototype in one week.
  • Successfully adapted research concepts from FhVLAD into a browser extension workflow.
  • Kept everything fully offline and privacy-friendly with no cloud uploads.
  • Integrated Chrome’s Prompt API to improve UX with AI-generated explanations.

What we learned

  • Fisher Vectors are more flexible than image hashing for detecting visually similar content.
  • Browser-based ML with tfjs is possible but requires careful memory optimization.
  • Strong hackathon prototypes focus on core functionality first.
  • Building solo reinforced the importance of balancing ambition with sustainability.

What’s next

  • Improve accuracy with AI-based content cropping and noise reduction.
  • Retrain the GMM on larger datasets like OpenImages for broader coverage.
  • Optimize performance and fully deploy as a production-ready Chrome extension.
  • Explore alternative similarity algorithms for Chromium-based browsers.

Built With

Share this project:

Updates

Submission history