Rough Sets: Theoretical Aspects of Reasoning about Data
Rough Sets: Theoretical Aspects of Reasoning about Data
| Hi-index | 0.00 |
This new method of detecting and compensating shadow is based on the general principle of the rough sets. Shadow recognition is constrained by the rough sets principle according to which the upper approximation of objects must contain non-empty lower approximation - the true class of objects in question. By imposing this constraint, the well known localization-detection trade-off is solved. In the first step the shadow is detected reliably by using a high threshold. Reliable classification (shadow detection) with the aid of a high threshold makes the lower approximation of shadow. Then, the upper approximation is constructed based on the lower approximation (reliably detected shadow) by using a low threshold. Directly using a low threshold would detect a lot of clutter and noise rather than shadow. Rough sets principle prevents this: each shadow candidate must contain the lower approximation. On the other hand, making the threshold high detects shadows reliably, but not accurately, for instance, shadow frequently begins at a low threshold. Rough sets solves this problem by tracking the upper approximation with the aid of a low threshold from the lower approximation.