Self-adaptation in Image and Video Retrieval - Multimedia Database Retrieval: Technology and Applications - page 97

Fig. 3.18 A comparison of the retrieval performance, showing 16 top matches of video shorts;
( a ) obtained by KFVI and ( b ) obtained by TFM. The query is shown in the top-left corner, and
the retrieval videos are ranked according to the descending order of the similarity scores, from left
to right and top to bottom. From the result, TFM allows similarity matching based on the video
contents, whereas KFVI emphasizes the content of the key-frames, resulting in the similar color of
the key frames
precision increased by more than 11 %, and is particularly significant in the top 10
to 16 retrievals. Secondly, the network is stabilized very quickly. Thirdly, allowing
many iterations degraded the performance slightly. Finally and most significantly,
the results were achieved by simply allowing activation flow, with no user input.