Dynamic spectral residual superpixels

We consider the problem of segmenting an image into superpixels in context $k$-means clustering, which we wish to decompose local, homogeneous regions corresponding underlying objects. Our novel approach builds upon widely used Simple Linear Iterative Clustering (SLIC), and incorporate a measure objects' structure based on spectral residual image. Based this combination, propose modified initialisation scheme search metric, helps keeps fine-details. This combination leads better adherence object boundaries, while preventing unnecessary segmentation large, uniform areas, remaining computationally tractable comparison other methods. demonstrate through numerical visual experiments that our outperforms state-of-the-art techniques.