Rotation Invariant Local Derivative Patterns for Image Retrieval and Indexing

Vijay Shrinath Patil and Pramod Jagan Deore

Feature extraction is the most prominent step in image retrieval and indexing. This paper proposes a novel feature using rotation invariant local derivative pattern for content based image retrieval. The proposed method describes the new feature by encoding the direction and magnitude of the reference pixel with its neighboring pixels. For defining the direction patterns, first order derivatives along the 0, 45, and 90 degrees are considered. These patterns are then further encoded into unique patterns using third order derivatives. The magnitude patterns along horizontal, vertical, diagonal and anti-diagonal directions are computed. The generated direction and magnitude patterns are converted into rotational invariant features and concatenated with each other to form Rotational Invariant Local Derivative Patterns (RILDPs). The RILDPs are then fused with Edge Directed Histogram for effective retrieval of the images. The performance of the proposed approach is evaluated on three different databases. The average precision of the proposed system is improved from 72% to 80%, 70% to 84% and 83% to 90 % as compared to local binary pattern on Wang database, Brodatz database and MIT VisTex database respectively. The experimental evaluation shows that the proposed approach outperforms the other existing approaches.

Volume 11 | 04-Special Issue

Pages: 1548-1558