EMD vs. Euclidean distance

We had to use a smaller subset of truck images since the EMD algorithm is a time consuming computation. We reduced our database to 112 images in total. We observed that EMD and Euclidean distance measures performed almost with the same quality but Euclidean distance outperformed EMD (figure 4.13 on page ). The best classification rate for EMD was 40.9% and for Euclidean distance measures it was 50%. The highest values were achieved for 25 bins per tile. The classification rate decreased with number of bins.

Table 4.8: Distance Measures Comparison(see figure 4.13 on page )

Distance Measure and Classification Rate [%]

Number of Bins	EMD	Euclidean distance
25	40.90	50
21	34.09	43.18
17	36.36	45.45
15	43.18	47.72
13	45.45	45.45
11	40.90	47.72
10	22.73	36.36

Figure 4.13: We can see that EMD and Euclidean distance measures performed almost the same