Reducing the Teknomo-Fernandez pixel processing through Scale Invariant Feature Transform (SIFT) descriptor matching

• Main Author: FLAVIER, JAVIER TEODORO
• Format: text
• Published: Archīum Ateneo 2018
• Subjects: Image processing > Digital techniques; Image stabilization; Image processing > Mathematical models; Image analysis > Mathematical models; Computer algorithms.
• Online Access: https://archium.ateneo.edu/theses-dissertations/173; http://rizalls.lib.admu.edu.ph/#section=resource&resourceid=1535876971&currentIndex=0&view=fullDetailsDetailsTab
• Institution: Ateneo De Manila University

The number of pixels processed by the Teknomo-Fernandez (TF3,4) algorithm was reduced through Scale-Invariant Feature Transform (SIFT) feature detection and matching (TF-SIFT). Two (2) TF-SIFT variants were proposed, TF3,4-SIFT3 and TF3,4-SIFT0. Both used TF-based SIFT matching, mask generation, and background image generation using the generated mask. The TF3,4-SIFT3 performed matching every three (3) frames for all levels, while TF3,4-SIFT0 performed it on the descriptors of the initial 81 frames. The performances were evaluated on the Wallflower and BMC datasets. TF3,4-SIFT3 and TF3,4-SIFT0 reduced a maximum of 28.54% of pixels processed compared to TF3,4. TF3,4-SIFT0 matched less features as the TF level processing of feature matching increased. Both variations yielded accuracies comparable with TF3,4 and ranked 24th out of 33 previously proposed techniques. Incorporating the generated mask on the background subtraction operation decreased accuracy. Both variants had an increased processing time caused by the time it takes for the SIFT feature detection and matching steps. Preprocessing videos with illumination changes using equalization techniques increased the features matched, resulting in further reduction in the number of pixels processed. It is possible to reduce the number of pixels processed by the TF algorithm by incorporating SIFT descriptor matching in the expense of an increase in runtime.