Image-based segmentation and measurement has emerged as an affordable technology for information extraction and physical modeling with applications to automatic derivation of grain size distributions and estimation of hydraulic properties over multiple spatial scales. However, the information content of a soil image is compounded and influenced by many factors such as soil type/texture, illumination, contrast differences, and surface characteristics which make image-based segmentation and measurement a non-trivial task. In study we present a new approach for textural pattern segmentation and spatial feature measurements. This approach is embedded in a multi-scale representation. In particular, the scale at which the maximum determinant of the second moment matrix is attained is hypothesized to give information about how large an image feature is. Interestingly enough, this approach offers a simple, yet elegant approach for texture segmentation and classification in heterogeneous systems. Moreover, it serves as an effective tool for characterizing spatial features as particle sizes. Practical applications are demonstrated using images of different porous materials including soil, rocks, and building materials.