This paper presents an object segmentation technique that builds on the success of Seeded-Region Growing (SRG) segmentation. SRG methods are typically initialized by a single point or patch in the image that represents the object of interest. Unlike previous approaches which utilize patches of the object of interest to obtain first and second-order characteristics, the author explores the potential of higher-order textural statistical descriptors. The proposed unsupervised approach relies on both the homogeneous and heterogeneous textural characteristics of the selected object region to iteratively expand the boundary to encompass the full object. In addition, the research proposes a dynamic selection criterion for determining segmentation parameters based on patch neighborhood features. The presented experiments are conducted in unconstrained environments wherein a textural description of the object of interest is extracted and the proposed algorithm automatically segments it from the background and other captured objects in the scene. The approach is evaluated using various subsets of the PASCAL Visual Object Classes (VOC) challenge imagery. Through quantitative metrics and analysis, the proposed algorithmic framework outperforms state-of-the-art methods for segmenting objects with non-homogeneous textural descriptors from complex real-world environments.