Published in International Workshop on Video Processing and Quality Metrics for Consumer Electronics (VPQM), 2015
Image segmentation is a fundamental preprocessing step in multiple tasks for the recognition and detection of semantically meaningful objects. In the past decades numerous image segmentation algorithms have been proposed. However, complexities at and above 𝑂(𝑛^2) make many of these computationally very expensive, several approaches require human input or have poor boundary recall. Among the state-of-the-art algorithms are superpixel segmentations, for which fast and fully automatic approaches exist. However, often scenes have content which cannot be segmented precisely based purely on color information. Novel image and video acquisition hardware can capture not only color, but also depth and infrared information. This additional information can be used to enhance existing segmentation algorithms. We present a novel multi-channel extension to existing superpixel segmentations which makes use of this additional information in order to improve the boundary recall by more than 11% while maintaining the same oversegmentation factor compared to a purely color based segmentation.