Variant of Completed Robust LBP for Two-Level Probabilistic Content based Image Retrieval

Ekta Walia, Aman Pal, Deepak Pandian, and Danish Lohani

Keywords

Cooccurrence Matrix, Completed Robust LBP, Bayesian Sets, Angular Radial Transform, Average Retrieval Rate

Abstract

A two-level probabilistic content based image retrieval scheme for colored images is proposed. In the first level, the shape features computed through Color Angular Radial Transform(CART) are exploited for identifying the relevant images through the use of a Bayesian Framework. The features used in the second level are derived from Completed Robust Local Binary Pattern (CR-LBP) features. CR-LBP features are modified to capture texture information from colored images and are quantized for the computation of four-directional co-occurrence matrices to define a new color based texture descriptor. These new features (referred to as CR-LBP-Co) outperform their initial definition (implemented on colored images) in terms of retrieval accuracy. Further, they are computed only for a small set of images that have been identified to be closest to the query in the first level. The second level feature set comparison is made through a suitable similarity measure. When used in this probabilistic framework, the proposed CR-LBP-Co together with CART features achieve 51.96% retrieval accuracy in terms of Average Retrieval Rate(ARR) for Wang’s database.

Important Links:



Go Back