Paul Viola

1
Image Database Retrieval Overview

• Image Databases are proliferating
• Video Databases
• Catalog Databases
• Find me a bag that looks like a Gucci.
• Virtual Museums
• Find me impressionist portraits.
• We are building systems that can
• Retrieve images by example
• Can learn from feedback
• Reproduce Human perceptions
• Are very efficient

2
Image Database Retrieval Progress

• Further definition of goals (with NTT
  investigators)
• Interfacing with NTT Telop system
• Focus on 3D object recognition
• Development of Flexible Server Technology
• Fast prototyping of database servers
• Over 10 servers working on 3 different datasets
• Fast prototyping of web interfaces
• Research
• Theoretical advances have provided 72x speedup
• A much smaller set of features, stored at lower
  resolution is necessary.
• Integration of segmentation and indexing

3
Image Database 6 month goals

• Build a system using a very large database
• 50,000 images
• Explore Telop system in order to acquire labeled
  images
• 1000 images per day from CNN
• Integrate segmentation with Complex Feature
  recognition
• Learn new features using user feedback