- 412 Downloads
The contents of multimedia resources are often not machine-interpretable due to the number of potential interpretations, limited training dataset, poor noise-signal ratio, occlusion, compression artifacts, and the lack of common sense knowledge. The low-level feature descriptors utilized in machine learning usually do not correspond directly to the meaning of the audio or video content, which can be described with high-level descriptors only. Multimedia semantics provided in the form of structured data about depicted concepts and events can significantly improve the machine-interpretability of multimedia contents. Multimedia semantics describe the actual multimedia content, such as the history and provenance data of a 3D model rather than just the vertices and edges of the 3D polygons that constitute the model, or the event shown in a video rather than the movements of objects.
Unable to display preview. Download preview PDF.