Towards Fusion Local Features to Enhance Content-Based Image Retrieval

Hanan Al-Jubouri, Harin Sellahewa, Hongbo Du
(University of Buckingham)

Advances in multimedia technology have led to a huge number of digital images captured and stored on computers. Content-Based Image Retrieval (CBIR) is an effective way to retrieve these images. However, CBIR systems face the challenge of addressing the so-called semantic gap between high level semantics and low level visual features of images. Image feature extraction and segmentation are exploited to tackle this problem. Our research investigates effective methods to extract local colour and texture features using Discrete Cosine Transform (DCT), and Local Binary Pattern (LBP), and group them into homogenous regions that may map to semantic objects in an image. Preliminary experiments using the K-means and adaptive Expectation-Maximization Gaussian Mixture Model (EM/GMM) clustering algorithms show promising results. Furthermore, the fusion of similarity scores based on DCT and LBP features to enhance image retrieval accuracy was explored with some positive results.



Poster presented at the BMVW.