Mixture of Visual Features For Content-Based Image Retrieval

Abstract

Image retrieval system is a computer system for
browsing, searching and retrieve images from a huge data set
of digital images. Content-based image retrieval (CBIR) is, the
problem of searching for digital images in large datasets. CBIR
retrieves similar images from huge image dataset based on image
features. Content-based means that the searching will evaluate
the actual contents of the image. The content of image might
refer colors, shapes, textures, or any other information that can
be derived from the image itself. There are two major approaches
to content-based image retrieval using local image descriptors.
One of them is descriptor by descriptor matching and the another
one is based on comparison of global image representation that
describes the set of local descriptors of each image. Image
representation is one of the key issues for large-scale CBIR.
Using MPEG-7 descriptors and local descriptors more number of
features are extracted from given query image and to reduce the
feature size principle component analysis(PCA) is used. These
features are embedded and aggregated into a compact vector to
avoid indexing each feature individually. In the embedding step,
each local descriptor is mapped into a high dimensional vector.
The aggregation step integrates all the embedded vectors of an
image into a single vector which obtains a compact representation
for image retrieval. Subsequently, k-means algorithm is used,
clusters are formed and images are trained in addition cosine
similarity measure is used to increase the retrieval performance.
In re-ranking step, Euclidean distance is used to find similarity
and provide efficient searching.