Biomedical image retrieval using microscopic configuration with local structural information

摘要

This paper focusses on the use of microscopic configuration (MiC) for discriminative information to retrieve lung cancer images. Existing local binary pattern ( LBP ) detects the local structures, such as lighting spots and edges in images, whereas the local configuration pattern ( LCP ) explores multi-channel discriminative information of both the MiC and local structures of images. Both methods are used to extract the rotation- and scale-invariant features from all lung cancer images. The performance of these methods is tested by conducting experiments on benchmark biomedical database of Lung Image Database Consortium and Image Database Resource Initiative-computer tomography. The database includes CT images with region of interest. The results show that LCP yields significant improvement in terms of average retrieval precision and average retrieval rate as compared with LBP and other state-of-the-art texture descriptors.