MCA aided geodesic active contours for image segmentation with textures

• Construction of the mathematical framework to realize morphological diversity during contour evolution.
• Proposal of a curvelet based dynamically updatable edge map in the framework of MCA.
• The multi-scale evolution strategy to obtain a gradually accurate convergence toward the edges.
• Systematic experiments for MCA-GAC.

Models of geodesic active contour (GAC) cannot usually distinguish one morphological component from another under conditions of complex textures. This paper proposes a morphological component analysis (MCA) aided GAC, namely MCA-GAC. The central effort is to segment image objects accurately and overcome obstacles from the undesired textures during the contour evolution. MCA-GAC takes advantage of the iterative property of MCA and optimal sparse representation of curvelet for edges. Segmentation is accomplished by evolving MCA-GACs through curvelet scales and MCA iterations. MCA-GAC is testified under conditions of textures and additive Gaussian white random noise. Experimental results demonstrate that MCA-GAC has competitive and practical prospects in the tasks of segmentation.