Modeling vs. Segmenting Images Using A Probabilistic Approach

Image segmentation is conventionally formulated as a pixel-labeling problem, in which "hard" decisions have to be made to partition pixels into regions. As image segmentation is usually used as a preprocessing step in many image analysis applications, the segmentation errors introduced by the "hard" decisions bring difficulties to higher-level image analysis. In this paper, we propose a "soft" image segmentation method to model the object appearance and spatial layouts in an image with an incremental mixture of probabilistic models. The proposed approach extracts "soft" regions incrementally using adaptive apertures without making any hard decisions. We show that "soft" regions not only bring more robustness than conventional "hard" regions but also enable a higher-level region-based analysis.