Abstract for HONS 03/14
Computation of the Local Binary Pattern (LBP) descriptor of large scale images
Joshua Chen
Department of Computer Science and Software Engineering
University of Canterbury
Abstract
Local binary patterns (LBPs) are powerful texture descriptors that have recently found several applications in medical image analysis. Research in this field is currently directed towards parallel implementations suitable for processing large scale images. Programming tool-sets such as the Open Computing Language (OpenCL) opened up opportunities for the development of various parallel algorithms and applications for General-Purpose GPU (GPGPU), all executable across heterogeneous OpenCL compliant platforms. In this report, we give an introduction to the LBP texture descriptor and the OpenCL framework. We will also discuss the computation of LBP descriptors for high resolution tissue images of biopsy samples, and outline various implementation aspects of the algorithm in OpenCL. Experiments were conducted on consumer-grade graphical processing units (GPUs) and a central processing unit (CPU), addressing relations to more than algorithmic complexities but limits on physical resources too.