Homework 9 17-2011
Upcoming SlideShare
Loading in...5
×
 

Homework 9 17-2011

on

  • 205 views

 

Statistics

Views

Total Views
205
Slideshare-icon Views on SlideShare
205
Embed Views
0

Actions

Likes
0
Downloads
1
Comments
0

0 Embeds 0

No embeds

Accessibility

Categories

Upload Details

Uploaded via as Microsoft PowerPoint

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Processing…
Post Comment
Edit your comment

    Homework 9 17-2011 Homework 9 17-2011 Presentation Transcript

    • CrowdSearch: Exploiting Crowds for Accurate Real-time Image Search on Mobile Phones
      TienShengWen
      Department of Electronic Engineering
      Chung-Yuan Christian University, Taiwan
      Tingxin Yan, Vikas Kumar, Deepak Ganesan
      Department of Computer Science
      University of Massachusetts, Amherst, MA 01003
      {yan, vikas, dganesan}@cs.umass.edu
    • Outline
      • Introduction
      • System Architecture
      • Crowdsearch for Search
      • CrowdsearchAlgorithm
      • Image Search Engine
      • System Implementation
      • Experimental Evaluation
      • Conclusions
    • Introduction
      • Image search system for mobile phones
      • Real-time validation
      • Beyond Image Search
      • System Performance
      • Payment
    • System Architecture
      CrowdSearchis implemented on
      Apple iPhone and Linux servers.
      Requires three pieces of information prior to initiating search:
      (a) A image query
      (b) A query deadline
      (c) A payment mechanism
      for human validators
    • Crowdsearchfor Search
      • Amazon Mechanical Turk (AMT)
      • Constructing Validation Tasks
      • Minimizing Human Bias and Error
      • Pricing Validation Tasks
    • Crowd Search Algorithm (1/2)
      • Optimizing Delay and Cost
    • Crowd Search Algorithm (2/2)
      • Delay Prediction Model
      Case 1 - Delay for the first response:
      Case 2 - Inter-arrival delay between responses:
    • Image Search Engine (1/2)
      • The image search process contains two
      major steps:
      (1) Extracting features from a query image good features:
      Scale-Invariant Feature Transform (SIFT)
      (2) Search through database images with features of
      query image.
    • Image Search Engine (2/2)
      A SeqTree to Predict Validation Results.
      The received sequence is ‘YNY’, the two sequences
      that lead to positive results are ‘YNYNY’
      and ‘YNYY’. The probability that ‘YNYY’ occurs
      given receiving ‘YNY’ is 0.16/0.25 = 64%
    • System Implementation
      CrowdSearch Implementation Components Diagram
    • Experimental Evaluation (1/3)
      • Datasets
      • Improving Search Precision
      • Accuracy of Delay Models
    • Experimental Evaluation (2/3)
      • CrowdSearchPerformance
    • Experimental Evaluation (3/3)
      • Varying user-specified deadline
    • Conclusions
      • Multimedia search presents a unique challenge.
      Because image search system is still far from reality.
      • Humans are excellent at distinguishing images, thus human validation can greatly improve the precision of image search. However, human validation costs time and money, hence we need to dynamically optimize these parameters to design an real-time and cost-effective system.