Selecting features for neural networks to aid an iconic search through an image database
Kieron Messer · 1997
In this paper a method that facilitates an iconic query of an image/video database is presented. A query object is characterised by colour and texture properties. A feed-forward neural network is then trained on these features using the conjugate gradient-descent algorithm. The same characteristics are computed locally for the database images. The trained neural network is then used to test for the similarity between the iconically specified query and the database image descriptors. We show that by carefully selecting the set of descriptors we can significantly reduce the network size whilst not affecting the quality of results obtained.