Image object extraction with shape and edge-driven Markov random field model

For object extraction, the target object in images often cannot be extracted completely and accurately using only low-level image features, especially from cluttered, occluded and noisy images. In practice, the shape of the target object is often known in advance, and edges can be extracted directly from image, which can contribute to the object extraction task. The authors introduce shape prior and edge to Markov random field (MRF) model, propose a shape and edge-driven MRF classification model for image object extraction. To exploit the shape prior, the energy function is defined by both image features and the known shape template. Image edges are extracted and added to the energy function to permit slight shape deformation. The whole energy function is minimised by graph cuts. In addition, an alignment process is introduced to handle the affine variations between target object and shape template. The edge reduces the influence of inaccurate shape alignment because of shape deformation and makes the result smoother. The experiments show that shape and edge play irreplaceable roles for accurate object extraction.