Hyperlinking videos for large video collections

speaker: Ngo Chong-Wah event: Big Data Workshop, UTAR Kampar, Malaysia ** link two videos with near duplicates segments ** salient point extration using SIFT keypoints ** speed up searching for similar images *** cluster salient points into bag of visual word *** each image represented as a BoW vector (histogram of visual words) *** millions of visual words ** finding visual words in a query image *** custer visual words into vocabulary tree ** trade-off between small and large vocabulary *** small vocab - miss features *** large vocab - more misses *** hamming embedding ** query process query image -> feature extraction -> quantization -> hamming embedding -> multiple alignment -> generate rank list from vocab tree