Feature extraction based on ISIFT algorithm for image retrieval
Chunli Zhai, Yanchun Shen, Chao Li · 2010
In this paper, we proposed the ISIFT algorithm, in the normalized scale space, we generate 64-dimensional (4×4×4) feature vectors in order to reduce dimension, and improve matching accuracy by bidirectional matching, then based on these characteristic points we build index using BBF algorithm to find the nearest neighbour, and complete image retrieval finally. Image retrieval based on text, color, texture and other features often faced with the false retrieval caused by rotation, scaling and stretch changes, this study can not only avoid the false retrieval, but also applies in the image retrieval from observation target or scene in different perspectives.