Abstract:Compact and discriminative visual codebooks are preferred in many visual recognition tasks. In the literature, a number of works have taken the approach of hierarchically merging visual words of an initial large-sized codebook, but implemented this approach with different merging criteria. In this work, we propose a single probabilistic framework to unify these merging criteria, by identifying two key factors: the function used to model the class-conditional distribution and the method used to estimate the distribution parameters. More importantly, by adopting new distribution functions and/or parameter estimation methods, our framework can readily produce a spectrum of novel merging criteria. Three of them are specifically discussed in this paper. For the first criterion, we adopt the multinomial distribution with the Bayesian method; For the second criterion, we integrate the Gaussian distribution with maximum likelihood parameter estimation. For the third criterion, which shows the best merging performance, we propose a max-margin-based parameter estimation method and apply it with the multinomial distribution. Extensive experimental study is conducted to systematically analyze the performance of the above three criteria and compare them with existing ones. As demonstrated, the best criterion within our framework achieves the overall best merging performance among the compared merging criteria developed in the literature.

A Generalized Construction of Codebook Asymptotically Meeting the Welch Bound

A Generalized Construction of Asymptotically Optimal Codebooks.

A Generic Construction of Complex Codebooks Meeting the Welch Bound.

Asymptotically Optimal Codebooks in Regard to the Welch Bound with Characters

Codebooks from Almost Difference Sets

New Constructions of Codebooks Nearly Meeting the Welch Bound with Equality

Generalized asymptotically optimal codebooks

Constructions of Asymptotically Optimal Codebooks with Respect to Welch Bound and Levenshtein Bound.

Construction of Cyclotomic Codebooks Nearly Meeting the Welch Bound

Two Classes of Codebooks Nearly Meeting the Welch Bound

A new construction of asymptotically optimal codebooks

Optimal and Asymptotically Optimal Codebooks As Regards the Levenshtein Bounds.

Jacobi Sums over Galois Rings of Arbitrary Characters and Their Applications in Constructing Asymptotically Optimal Codebooks

A Method Of Optimizing Codebook Based On Codeword Use Frequency

A Generalized Probabilistic Framework for Compact Codebook Creation

Generalized Pairwise Complementary Codes with Set-Wise Uniform Interference-Free Windows

A New Class of Near-Optimal Partial Fourier Codebooks from an Almost Difference Set

Weak Superimposed Codes of Improved Asymptotic Rate and Their Randomized Construction

Bounds and Constructions for Generalized Batch Codes

Efficient Codebook Constructions of AIVF Codes.

A Tighter Upper Bound of the Expansion Factor for Universal Coding of Integers and Its Code Constructions