LEARNING SHORT BINARY CODES FOR LARGE-SCALE IMAGE

RETRIEVAL

Abstract:
Large-scale visual information retrieval has become an active research area in this
big data era. Recently, hashing/binary coding algorithms prove to be effective for
scalable retrieval applications. Most existing hashing methods require relatively long
binary codes (i.e., over hundreds of bits, sometimes even thousands of bits) to achieve
reasonable retrieval accuracies. However, for some realistic and unique applications,
such as on wearable or mobile devices, only short binary codes can be used for efficient
image retrieval due to the limitation of computational resources or bandwidth on these
devices. In this paper, we propose a novel unsupervised hashing approach called min-
cost ranking (MCR) specifically for learning powerful short binary codes (i.e., usually
the code length shorter than 100 b) for scalable image retrieval tasks. By exploring the
discriminative ability of each dimension of data, MCR can generate one bit binary code
for each dimension and simultaneously rank the discriminative separability of each bit
according to the proposed cost function. Only top-ranked bits with minimum cost-values
are then selected and grouped together to compose the final salient binary codes.
Extensive experimental results on large-scale retrieval demonstrate that MCR can
achieve comparative performance as the state-of-the-art hashing algorithms but with
significantly shorter codes, leading to much faster large-scale retrieval.

Proposed System :
We propose a novel unsupervised hashing method called Min-
cost Ranking (MCR) to learn very short
binary codes (i.e., usually the code length shorter than 100 bits) for large-
scale retrieval tasks. To better explore the
discriminative properties for effective binary coding, a two stage learning procedure
is introduced in our MCR. In the first stage, we aim to learn one bit binary code for each
dimensionof the original features. Considering the intrinsic locality consistency
between dimensions, we associate each dimension
with its neighboring dimensions to form a linear subspace.Further to assess the
discriminative ability, each dimension is optimized by minimizing the cost function under
the maximum margin criterion in their corresponding subspace and finally
transformed to one bit via the obtained mapping vector.