Guest Editorial: Tracking, Detection and Segmentation

Abstract

This special issue contains papers selected by representatives of the British Machine Vision Association (BMVA) describing recent advances made against major outstanding challenges in computer vision: detection; tracking and pose estimation; activity recognition; video segmentation; dense correspondence and video synthesis. Establishing a dense correspondence between images remains an open problem; recent solutions based on randomised search have yielded promising advances in accuracy and efficiency. Our first paper, “PMBP: PatchMatch Belief Propagation for Correspondence Field Estimation” presents a unified framework in which the Patch Match algorithm is first described as a variant of belief propagation, and then extended using features of the latter to yield results that are both more accurate than the original Patch Match formulation and orders of magnitude faster than previous belief propagation approaches. ‘Mixture ofTrees ProbabilisticGraphicalModel forVideo Segmentation” by Badrinarayanan, Budvytis and Cipolla, continues the theme of video analysis; proposing a new spatio-temporal representation, the Mixture of Trees for temporally coherent semantic video segmentation. This