Semantics extraction from multimedia data

• Main Author: Wang, Xiao Meng
• Other Authors: Mao Kezhi
• Format: Final Year Project
• Language: English
• Published: 2016
• Subjects: DRNTU::Engineering
• Online Access: http://hdl.handle.net/10356/67931
• Institution: Nanyang Technological University

Driven by the significant advancements in Computer vision technology and relevant applications such as Unmanned Aerial Vehicle (UAV) and robotics, scene classification has become a hot and challenging problem in nowadays. Scene classification means categorizing images into different categories according to the physical or semantic properties. Most of the existing scene classification algorithms are based on the low level features and have achieved success in their own field. However, due to the large uncertainties in the scenes used for classification, it is very hard to achieve a high accuracy for a scene classification technique. In this thesis, a new scene classification technique is proposed based on a combination of image and audio features. 3 stages are involved in this new technique. The first stage is image feature extraction, the main feature used is Scale Invariant Feature Transform (SIFT); the second stage is audio feature extraction, the feature used here is the Mel-Frequency Cepstrum Coefficients (MFCCs). The third stage is Fusion and Classification, which is combining the features extracted from both image and audio and classify the scenes using Support Vector Machine (SVM). Several experiments have been done to test the performance of this new technique. Based on the test results, the final accuracy on classifying the different sports activities can reach up to 85%.