Self-adaptation in Image and Video Retrieval - Multimedia Database Retrieval: Technology and Applications - page 76

Fig. 3.7 ( a ) Initial retrieval
result for DB1 database;
( b ) retrieval result after
application of automatic
ARBFN
account that the users had to provide feedback on each of the images returned by a
query in order to obtain these results.
As found in the studies from Chap. 2 , retrieval performance can be progres-
sively improved by repeated relevance feedback from the user. The semiautomatic
approach reported here greatly reduced the number of iterations required for user
interaction. This significantly improved the overall efficiency of the system. In
the semiautomatic learning method, the retrieval system first performed automatic
retrieval for each query to adaptively improve its performance. After four iterations,
the retrieval system was then assisted by the users. Table 3.4 provides a summary
of the retrieval results, based on one round of user-controlled RF. It was observed
that the semiautomatic RF method was superior to the automatic method and the
user interaction method. The best performance was given by the semiautomatic
ARBFN at 83.41 % using WM descriptor, and 81.14 % using MHI descriptor.