| Query type | Description (segments and descriptors) | Execution time (sec) |
| Text-based semantic query | Keyframe (keyword) | 0.125 |
| Text-based semantic query | Moving Region (keyword) | 0.125 |
| Text-based semantic query | Keyframe (keyword), Moving Region (keyword) |
0.188 |
| Color query | Keyframe (CSD) | 0.141 |
| Texture query | Keyframe (HTD) | 0.125 |
| Color + Texture query | Keyframe (CSD+HTD) | 0.172 |
| Shape query | Moving Region (RSD) | 0.156 |
| Spatial query | Text-based, 2 Still Regions | 0.172 |
| Spatial query | Text-based, 2 Moving Regions | 0.187 |
| Spatial query | Sketch-based, 2 Moving Regions | 0.187 |
| Composite query in Figure 3.1 | Keyframe (DCD+keyword), Moving Region (CSD+RSD+keyword) |
0.438 |
| Composite query in Figure 3.2 | 2 Still Regions (CSD+EHD), Moving Region (keyword) |
0.391 |
The query execution time is proportional to the number of subqueries (number of video segments and descriptors in the query), database size (number of video segments in the database), the sizes of the descriptors and the complexity of the matching algorithm (distance metric). Queries involving low-level descriptors take longer to execute compared to text-based queries since the distance computation between the low-level descriptors are computationally more expensive.
The multi-threaded query processing architecture provides some degree of parallelism and shortens the query execution times when the subqueries are executed in separate threads. For instance, a Keyframe query with CSD takes 0.141 seconds and a Keyframe query with HTD takes 0.125 seconds to execute, while a Keyframe query with CSD and HTD descriptors takes 0.172 seconds to execute, which is less than the serial execution times of CSD and HTD queries (0.266 seconds). This is also demonstrated in the two composite queries in the table.