Matching-constrained active contours with affine-invariant shape prior

• Main Authors: Yeung, Sai-Kit, Wang, Junyan, Chan, Kap Luk
• Other Authors: School of Electrical and Electronic Engineering
• Format: Article
• Language: English
• Published: 2015
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: https://hdl.handle.net/10356/107217; http://hdl.handle.net/10220/25256; http://dx.doi.org/10.1016/j.cviu.2014.11.002
• Institution: Nanyang Technological University

In the object segmentation by active contours, an initial contour provided by user is often required. This paper extends the conventional active contour model by incorporating feature matching in the formulation for automatic object segmentation, yielding a novel matching-constrained active contour. The key to our formulation is a mathematical model of the relationship between interior feature points and object shape, called the interior-points-to-shape relation. According to this interior-points-to-shape relation, we are able to achieve the automatic object segmentation in two steps. Specifically, we are able to estimate the object boundary position given the matched interior feature points. Afterwards, we are able to further optimize the boundary position in the active contour framework. To obtain a unified optimization model for this task, we additionally formulate the matching score as a constraint to active contour model, resulting in our matching-constrained active contour. We also derive the projected-gradient descent equations to solve the constrained optimization. In the experiments, we show that our method achieves automatic object segmentation, and it clearly outperforms the related methods.