Information Technology Reference
In-Depth Information
On kNN Classification
and Local Feature Based Similarity Functions
Giuseppe Amato and Fabrizio Falchi
ISTI-CNR, via G. Moruzzi 1, 56124 Pisa, Italy
{giuseppe.amato,fabrizio.falchi}@isti.cnr.it
Abstract.
In this paper we consider the problem of image content recognition
and we address it by using local features and kNN based classification strategies.
Specifically, we define a number of image similarity functions relying on local
features comparing their performance when used with a kNN classifier. Further-
more, we compare the whole image similarity approach with a novel two steps
kNN based classification strategy that first assigns a label to each local feature
in the document to be classified and then uses this information to assign a label
to the whole image. We perform our experiments solving the task of recognizing
landmarks in photos.
Keywords:
Image classification, Recognition, Landmarks, Pattern recognition,
Machine learning, Local features.
1
Introduction
Image content recognition is a very important issue that is being studied by many sci-
entists worldwide. In fact, with the explosion of the digital photography, during the
last decade, the amount of digital pictures available on-line and off-line has extremely
increased. However, many of these pictures remain unannotated and are stored with
generic names on personal computers and on on-line services. Currently, there are no
tools and effective technologies to help users in searching for pictures by real content,
when they are not explicitly annotated. Therefore, it is becoming more and more diffi-
cult for users to retrieve even their own pictures.
A picture contains a lot of implicit conceptual information that is not yet possible to
exploit entirely and effectively. Automatically content based image recognition opens
up opportunities for new advanced applications. For instance, pictures themselves might
be used as queries on the web. An example in this direction is the service “Google
Goggles” [11] recently launched by Google, that allows you to obtain information about
a monument through your smartphone using this paradigm.
Note that, even if many smartphones and cameras are equipped with a GPS and a
compass, the geo-reference obtained with this is not enough to infer what the user is
actually aiming at. Content analysis of the picture is still needed to determine more
precisely the user query or the annotation to be associated with a picture. A promising
approach toward image content recognition is the use of classification techniques to
associate images with classes (labels) according to their content. For instance, if an
