Classification Similarity Learning Using Feature-Based and Distance-Based Representations: A Comparative Study

Figures & data

FIGURE 1 Feature-based and multidistance-based classification similarity learning approaches.

FIGURE 2 Scheme of the preprocessing layer in the Multidistance L_p representation.

FIGURE 3 Scheme of the preprocessing layer in the Multidistance L₁ representation.

TABLE 1 A Summary of the Three Datasets Used in the Experiments

Dataset	Size	Categories	Total Dimension	Descriptor Sizes
Small	1 508	29	50	30,10,10
Art	5 476	63	104	30,12,7,3,10,10,4,11,11,6
Corel	30 000	71	89	9,16,32,32

FIGURE 4 Average MAP vs. training set size for the Small database.

FIGURE 5 Average MAP vs. training set size for the Art database.

FIGURE 6 Average MAP vs. training set size for the Corel database.