Finally, A RSI database consists of 50 images is built
to test the proposed algorithm. Buildings in all the 50
images are labeled manually in advance. Table II lists the
false alarm rate and missing rate of the three algorithms
for the RSI database. As our analysis, the algorithm
presented in reference [8] detects only the rectangular
candidate regions, many building in other shapes are
missed by this algorithm. Hence this algorithm shows
higher missing rate. However, the false alarm rate of it is
comparatively lower. In contrast, the algorithm in
reference [11] depends more on the local color.
Consequently its missing rate is lower but the false alarm
rate is evidently higher than other algorithms. As a
comparison, our algorithm works better than the two
algorithms in both the false alarm rate and the missing
rate. It proves that the our model is obviously has better
performance in precision and robustness.