Skip to Main Content
This paper presents a new method for improving region segmentation in sequences of images when temporal and spatial prior context is available. The proposed technique uses elementary classifiers on infra-red, polarimetic and video data to obtain a coarse segmentation per-pixel. Contextual information is exploited in a Bayesian formulation to smooth the segmentation between frames. This is a general framework and significantly enhances segmentation from the classifiers alone. The method is demonstrated by classifying images of a rural scene into 3 positive classes: sky, vegetation and road, and one class of all other unlabelled data. Priors for the probabilistic smoothing in this scene are learned from ground-truth images. It is shown that an overall improvement of around 10% is achieved. Individual classes are improved by up to 30%.
Date of Conference: 13-18 June 2010