Home > Published Issues > 2015 > Volume 3, No. 1, June 2015 >

CE Video Summarization Using Relational Motion Histogram Descriptor

Mohamed Maher Ben Ismail and Ouiem Bchir
College of Computer and Information Sciences, Computer Science department, King Saud University, Riyadh, KSA

Abstract—We propose a capsule endoscopy summarization system relying on two main components. The first one consists of a semi-supervised Clustering and Local Scale Learning (SS-LSL) algorithm which is used to group video frames into prototypical clusters that summarize the video scene. The second component of the system relies on a novel Relational Motion Histogram (RMH) descriptor that is designed to represent local motion distribution between two contiguous frames. The main idea is to identify "highlight" frames which contain typical variations within the frame collection. These variations are due to different pathologies, small tumors and other subtle abnormalities of the small intestine, etc. The proposed video summarization system is trained, field-tested, evaluated, and compared through a large-scale cross-validation experiment.

Index Terms—capsule endoscopy, semi-supervised clustering, relational clustering, motion descriptor

Cite: Mohamed Maher Ben Ismail and Ouiem Bchir, "CE Video Summarization Using Relational Motion Histogram Descriptor," Journal of Image and Graphics, Vol. 3, No. 1, pp. 34-39, June 2015. doi: 10.18178/joig.3.1.34-39