Texture Image Retrieval Using Systematic CBIR Technique
Sana Qammar, Department of Computer Science, University of Engineering and Technology, Taxila, Pakistan.
Muhammad Javed Iqbal, Department of Computer Science, University of Engineering and Technology, Taxila, Pakistan.
Corresponding Author:
Muhammad Javed Iqbal(javed.iqbal@uettaxila.edu.pk)
Abstract:
Texture features are essential for many computer vision applications, such as object recognition, segmentation, and retrieval. Local binary pattern (LBP) and local neighborhood difference pattern (LNDP|) are effective texture descriptors for distinguishing different textures. While LBP captures the overall texture structure, LNDP can detect more subtle texture variations. However, they have limitations, such as LNDP’s lack of rotational in-variance and LBP’s limited ability to capture fine texture details. Locally encoded transform feature histogram (LETRIST) improves upon LBP+LNDP by combining them with local gradient information, resulting in a more robust and accurate texture descriptor. This paper proposes a fusion approach called LETRIST+LBP+LNDP for content-based image retrieval systems, which significantly improves accuracy on widely used datasets such as Corel-10, STex, and Brodatz. The system combines LETRIST’s rotation-invariant feature with LBP+LNDP’s texture and local pattern features. Performance analysis shows the achieved robust and discriminative retrieval results of proposed method with comparison of existing efficient methods, in term of Mean Average Precision (MAP) values. In comparison of accuracy of our proposed systems results are higher than other state of the art descriptors.
Keywords:
CBIR; Texture Feature; LETRIST; Rotation-Invariant; Fusion Approach