Kim Yong-Ho
Pukyong National University, SouthKorea
Title: Automatic key-frame extraction techniques using ERP responses to Real-time Video Watching: Testing N400 and P600 hypotheses
Biography
Biography: Kim Yong-Ho
Abstract
Theoretical Background: In the linguistics literature on verbal semantic integration, the N400 effect refers to the fact that unexpected words will cause very low level of negative potentials in brainwave measures around 400ms(milli-seconds) after repeated presented settings and the P600 effect refers to the fact the unexpected words will cause very high level of positive potentials around 600ms.
Methods: Research literature on the video summarization indicates importance of the development of an un-obtrusive method of gathering external information of video users (Z. Lu and K. Grauman,, 2013; W. Ren and Y. Zhu,, 2008; A. Porselvi and S. Gunasundari, 2013). The analysis of event-related potentials (ERP) is such a method which extracts only a result of reaction with respect to certain stimuli from the brain waves.
Findings and Implications: We observed greater maximum potentials at the left prefrontal cortex (FP1, t = 6.930, p = 0.013), the left, right, and middle central lobes (C3, t = 4.670, p = 0.039; Cz, t = 9.818, p = 0.004; C4, t = 10.549 , p = 0.003), and the right and middle frontal-central lobes (FC4, t = 7.370, p = 0.011; FCz, t = 6.541, p = 0.016) of brain wave responses to topic-relevant shots. The right parietal and right temporal-parietal lobes (P8, t = 4.483, p =0.043; TP8, t = 5.326, p = 0.028). It is indisputable to further attempt this sort of ERP analysis of the EEG data during continuous viewing session using topic-relevance ratings from still image testing. Still, the surprisingly large effect of N400 and P600 at prefrontal lobe are asking for further refinement in the future experimental design.
Importance: We developed a method to import time code of video presentation to the EEG data with the topic-relevant information from ratings of topic-relevance for still-image captured from visual shots which were included in the videos. SNR(signal-to-noise ratio) of ERP analysis for the visual shots are about 12.2 well fit in the rage of 10-14 as professional consultant recommended for SNR.