Browsing Personal Images Using Episodic Memory Chufeng Chen School of Computing and Technology, University of Sunderland
Related works What is episodic memory Abrams et al. (1998) : Episodic memory in HCI Platt et al. (2002) : Time clustering Naaman et al. (2004) : Time and Location Classification Cooper et al. (2005) : Time and Colour Clustering
Development of Time & Location Clustering Model Time and location Clustering model Example of Data sets, and how to separate events User interface
Time and location Clustering model
Example of Data sets, and how to separate events
Example of User interface
User Centered Evaluation The hypothesis: browsing features related to episodic memory, incorporated into our time and location combination browser would improve image searching of personal collections 10 Subjects (200 photo collections) Five Browsers Time and location combination browser BR's Photo-Archiver Canon Zoom-Browser-EX Unindexed browser (WinXp image browser) Time alone (Platt, 2002)
Experimental Design Latin-Square Design Scenario Searching Tasks General Searching Tasks (4 for each subject) Specific Searching Tasks (4 for each subject) Record Searching Time for each Scenario Tasks User Satisfaction Questionnaire for each System Five Likert scale questionnaires The questionnaire had been used in Platt’s (2002) user study
Experiment Results (scenario tasks searching time) Time & location combine d BR's Photo- Archiver Canon Zoom- Browser- EX Un- indexed browser Time alone ANOVA F(4, 45) = 1. Average searching time general scenario tasks , p = Average searching time specific scenario tasks , p = Average total finish time , p =
Experiment Results (Questionnaire analysis) Time & location combined BR's Photo- Archiver Canon Zoom- Browser-EX Un- indexed browser Time alone ANOVA F(4, 45) = 1. I like this image browser , p= This browser is easy to use , p= This browser feels familiar , p= It is easy to find the photo I am looking for , p< A month from now, I would still be able to find these photos , p= I was satisfied with how the pictures were organized , p< Total , p<
System Centre Evaluation Recall and Precision 1. user and machine place the image pair in the same event; 2. user places the image pair in the same event, but the machine places them in different events; 3. user places the image pair in different events but the machine places them in the same event; 4 user and machine both place the image pair into separate events. Recall = (pairs in 1) / (pairs in 1 + pairs in 2) Precision = (pairs in 1) / (pairs in 1 + pairs in 3).
R & P Results Time and location clustering.Time Alone clustering RecallPrecisi- on F 1 measure RecallPrecisi- on F 1 measure Subject Subject Subject Subject Subject Subject Subject Subject Subject Subject Average
Findings Time and location browser significantly better than other four standard browsers in both searching time and user satisfaction Time and location combination browser had greater retrieval effectiveness than the time alone browser Factors related to human episodic memory, time and location, can be used to help users search their personal photograph collections more easily
Works So Far Develop a Location Annotation System for Personal Images (annotating by location gazetteer) Develop a Keyword Search Engine of System Annotation and User Annotation Evaluation User study: system annotation Vs. User Annotation Vs. T & L Browsing User study: system annotation Vs. User Annotation Vs. T & L Browsing Recall and Precision: System annotation Vs. User annotation Recall and Precision: System annotation Vs. User annotation
Location Annotation Data
Search Engine Example