We briefly outline the process of sharing and retrieving images in DISCOVIR
(DIStributed COntent-based Visual Information Retrieval). First, each peer is
responsible to perform feature extraction, for example, color, texture, shape, and
so on, on its shared images in each peer using the DISCOVIR client program.
With this program, each peer maintains its local index of feature vectors of its
image collection. When a peer, the requester, initiates a query by giving an example
image and a particular feature extraction method, it performs feature extraction on
the example image and sends the feature vector, contained in a query message, to
all its connecting peers. Consequently, other peers compare this query to their
feature vector index. Based on a distance measure, they find a set of similar images
and return results back to the requester. Likewise, these peers will propagate the
query to their connecting peers and this process continues to query other peers in a
large portion of the network.
DISCOVIR uses a plug-in architecture to support different feature extraction methods.
User may choose to download a plug-in in the format of compiled Java bytecode if
they want to perform CBIR based on that particular feature extraction method. For a
screen capture of the DISCOVIR client program, which can be downloaded from the
site [MIPLAB]. In the following, we describe the overall architecture of DISCOVIR,
which operates on the current Gnutella network, and the modification of query messages.
Gnutella Message Modification : The DISCOVIR system is compatible with the
Gnutella (v0.4) protocol [Gnutella]. In order to support the image query functionalities
mentioned, two types of messages are added. They are:
--ImageQuery—Special type of Query message. It carries the name of the feature
extraction method and the feature vector of the query image.
--ImageQueryHit—Special type of QueryHit message. It responds to the Image
Query message. It contains the location, file name and size of similar images retrieved,
and their similarity measure to the query. In addition, the location information of
corresponding thumbnails is added for the purpose of previewing the result set in a
faster speed .
(DIStributed COntent-based Visual Information Retrieval). First, each peer is
responsible to perform feature extraction, for example, color, texture, shape, and
so on, on its shared images in each peer using the DISCOVIR client program.
With this program, each peer maintains its local index of feature vectors of its
image collection. When a peer, the requester, initiates a query by giving an example
image and a particular feature extraction method, it performs feature extraction on
the example image and sends the feature vector, contained in a query message, to
all its connecting peers. Consequently, other peers compare this query to their
feature vector index. Based on a distance measure, they find a set of similar images
and return results back to the requester. Likewise, these peers will propagate the
query to their connecting peers and this process continues to query other peers in a
large portion of the network.
DISCOVIR uses a plug-in architecture to support different feature extraction methods.
User may choose to download a plug-in in the format of compiled Java bytecode if
they want to perform CBIR based on that particular feature extraction method. For a
screen capture of the DISCOVIR client program, which can be downloaded from the
site [MIPLAB]. In the following, we describe the overall architecture of DISCOVIR,
which operates on the current Gnutella network, and the modification of query messages.
Gnutella Message Modification : The DISCOVIR system is compatible with the
Gnutella (v0.4) protocol [Gnutella]. In order to support the image query functionalities
mentioned, two types of messages are added. They are:
--ImageQuery—Special type of Query message. It carries the name of the feature
extraction method and the feature vector of the query image.
--ImageQueryHit—Special type of QueryHit message. It responds to the Image
Query message. It contains the location, file name and size of similar images retrieved,
and their similarity measure to the query. In addition, the location information of
corresponding thumbnails is added for the purpose of previewing the result set in a
faster speed .
No comments:
Post a Comment