Table 8.

Best-performing weight combinations for each taxon and their average retrieval precisions. The taxon Retitricolpites simplex (ID = 722) has been used as an example. Each of its 24 images were used as query images, searched against the database, and the top 10 most similar images in feature space were retrieved. All 215 weight combinations were used for each image, yielding a total of 215 × 24 = 5160 queries. For each image as a query image, maximal precision was identified. There could be multiple weight combinations (n out of 215) that produced the same maximal precision for the same query image. All of these weight combinations were considered candidates. The candidates that occurred most frequently were the final candidates. In this example, there were four (#Candidates) candidate weight combinations that were identified to produce maximal precisions in nine (#Occurrence) individual query occasions. The average precisions using each of four candidates across all 24 images were calculated, and the candidate with the highest average precision was the top choice.

IDa	#Candidate	#Occurrence	Top choiceb	Avg. precision (%)
Pollen
722	4	9	0.8_0.2_0.0	50.8
767	2	17	0.4_0.6_0.2	73.1
688	7	6	0.0_0.4_0.2	63.4
570	2	24	0.0_0.6_0.2	59.1
571	21	10	1.0_0.6_0.2	74.7
552	1	16	0.2_0.8_1.0	73.1
451	8	14	0.2_0.0_0.0	61.2
511	1	9	0.2_0.0_1.0	68.9
450	1	20	1.0_0.4_0.8	67.2
254	5	16	0.2_0.0_0.0	49.6
1430	13	13	0.4_0.2_0.2	62.1
365	2	10	0.0_0.2_0.6	49.5
148	2	8	0.8_0.2_0.6	40.5
246	14	15	0.4_0.2_0.2	38.9
1014	3	11	0.2_1.0_0.2	77.1
Spore
46	3	15	0.2_0.6_0.0	89.1
44	6	11	0.2_1.0_0.00	65.0
282	2	12	0.2_0.2_0.4	78.4
45	9	8	0.2_0.4_0.2	58.3
43	5	24	0.0_0.2_0.0	91.1

^aTaxa IDs are provided in Table 1.

^bWeight combinations are of format w_c_w_s_w_t for color, shape, and texture.