A region-based level-set active contour is the preferred choice for image-segmentation tasks, when the region of interest is defined by weak edges. While, the minimization of an energy functional leads to the contour’s conformation to object boundary, the energy is composed of contour’s internal and image-dependent external energies. In such a model, the overall motion (expansion/shrinkage) of contour is controlled by its area-energy, whereas the length-energy controls the contour’s elasticity. Traditionally, these two internal energies are weighted by scalar constants, which remain fixed throughout the level-set evolution. Both the internal energies are responsible for the contour’s regularization only, whereas the contour converges due to the minimization of external energies. Further, inappropriate weighting results into an inaccurate segmentation either because of premature convergence, or leakage beyond the object boundary. To address these issues, an adaptive spatially weighted region-based active contour (ASWRAC) is proposed. The weights are assigned adaptively to the internal energy terms at successive steps of evolution. This basically removes the need of weight initialization for an improved segmentation process. Additionally, the proposed weights are modeled as vectors, depending on the local regional statistics of the image. This causes the originally internal energies to also reflect external characteristics. Moreover, the time-step used in the discrete implementation of level-set evolution is also managed adaptively. This restricts the contour’s leakage beyond the region of interest. The suggested technique is tested on brain MR slices and other biomedical images from public databases. Comparison of the obtained results with state-of-the-art techniques shows its superiority in segmentation accuracy measured using the metrics: dice Similarity Coefficient, sensitivity, and specificity.
