Learning from large dataset: segmentation of capsule endoscopy videos

Abstract

Reviewing video of capsule endoscopy is a tedious work that takes hours. Hence, efficient and scalable approaches are needed to automate the process of large dataset and be able to refine the model given new examples. This paper presents an incremental SVM to learn from large dataset with dynamic patterns. Our method extends the reduced convex hull concept and defines the approximate skin segments of convex hulls. Experiments were conducted using synthetic data set, real–world data sets, and CE videos. Our results demonstrated highly competitive performance that requires much less resource, which cast new light on learning with limited resource.