ICIC Express Letters ICIC International ⃝c 2019 ISSN 1881-803X Volume 13, Number 9, September 2019 pp. 867{874 θ(1) TIME COMPLEXITY PARALLEL LOCAL BINARY PATTERN FEATURE EXTRACTOR ON A GRAPHICAL PROCESSING UNIT Ashwath Rao Badanidiyoor and Gopalakrishna Kini Naravi Department of Computer Science and Engineering Manipal Institute of Technology Manipal Academy of Higher Education Manipal, Karnataka 576104, India f ashwath.rao; ng.kini
[email protected] Received February 2019; accepted May 2019 Abstract. Local Binary Pattern (LBP) feature is used widely as a texture feature in object recognition, face recognition, real-time recognitions, etc. in an image. LBP feature extraction time is crucial in many real-time applications. To accelerate feature extrac- tion time, in many previous works researchers have used CPU-GPU combination. In this work, we provide a θ(1) time complexity implementation for determining the Local Binary Pattern (LBP) features of an image. This is possible by employing the full capa- bility of a GPU. The implementation is tested on LISS medical images. Keywords: Local binary pattern, Medical image processing, Parallel algorithms, Graph- ical processing unit, CUDA 1. Introduction. Local binary pattern is a visual descriptor proposed in 1990 by Wang and He [1, 2]. Local binary pattern provides a distribution of intensity around a center pixel. It is a non-parametric visual descriptor helpful in describing a distribution around a center value. Since digital images are distributions of intensity, it is helpful in describing an image. The pattern is widely used in texture analysis, object recognition, and image description. The local binary pattern is used widely in real-time description, and analysis of objects in images and videos, owing to its computational efficiency and computational simplicity.