search query: @supervisor Kaski, Samuel / total: 58
reference: 15 / 58
Author: | Nallamothu, Kranthi Kumar |
Title: | Decoding of dynamic visual scenes from gaze features |
Publication type: | Master's thesis |
Publication year: | 2012 |
Pages: | xi + 93 Language: eng |
Department/School: | Tietotekniikan laitos |
Main subject: | Informaatiotekniikka (T-61) |
Supervisor: | Kaski, Samuel |
Instructor: | Ramkumar, Pavan ; Pannasch, Sebastian |
OEVS: | Electronic archive copy is available via Aalto Thesis Database.
Instructions Reading digital theses in the closed network of the Aalto University Harald Herlin Learning CentreIn the closed network of Learning Centre you can read digital and digitized theses not available in the open network. The Learning Centre contact details and opening hours: https://learningcentre.aalto.fi/en/harald-herlin-learning-centre/ You can read theses on the Learning Centre customer computers, which are available on all floors.
Logging on to the customer computers
Opening a thesis
Reading the thesis
Printing the thesis
|
Location: | P1 Ark Aalto 6905 | Archive |
Keywords: | eye tracking gaze features dynamic scenes feature selection eye movement pattern classification filter method wrapper methods RFE RMANOVA SVM naive Bayes |
Abstract (eng): | This thesis aims to examine the similarity across participants' eye-gaze behaviour and explore eye-gaze features that can effectively classify different dynamic visual scenes. Eye-gaze behaviour of 17 naíve participants was recorded while viewing 62 different dynamic visual scenes grouped into 4 categories (artificial patterns, naturalistic sceneries, sports, and directed movies). From the eye gaze, 15 features were extracted for each video clip viewed by participants. Three classification approaches (SVM, Adaboost SVM, and Naíve Bayes) were used to classify the eye-gaze data. The analysis showed that eye-gaze patterns carry characteristic features that can be used to categorize the 4 video categories. Subsequently, a filter approach (RMANOVA), and a wrapper approach (RFE) were used for feature selection to identify a subset of features that can increase the classification accuracy. The filter approach (RMANOVA) results suggest that 10 out of 15 features were statistically significant across different video categories while wrapper approach (RFE) revealed 12 features when combined with a naíve Bayes classifier as the induction algorithm in RFE. The classification accuracy with all 15 features was 54.96+-0.93% (mean +- S.E.M.). The average accuracy increased to 58.96+-1.68% when the subset of features given by filter approach was considered but showed no recognizable change for subset of features selected by wrapper approach (54.84+-0.96%). The results suggest that eye-movement features allow for a classification of different dynamic visual scenes. The subset of eye-gaze features selected by filter approach gives higher classification accuracy over wrapper approach. |
ED: | 2012-11-06 |
INSSI record number: 45386
+ add basket
INSSI