Characterization of nanostructured material images using fractal descriptors

This work presents a methodology to the morphology analysis and characterization of nanostructured material images acquired from FEG-SEM (Field Emission Gun-Scanning Electron Microscopy) technique. The metrics were extracted from the image texture (mathematical surface) by the volumetric fractal descriptors, a methodology based on the Bouligand-Minkowski fractal dimension, which considers the properties of the Minkowski dilation of the surface points. An experiment with galvanostatic anodic titanium oxide samples prepared in oxalyc acid solution using different conditions of applied current, oxalyc acid concentration and solution temperature was performed. The results demonstrate that the approach is capable of characterizing complex morphology characteristics such as those present in the anodic titanium oxide.

💡 Research Summary

The paper introduces a novel methodology for the quantitative morphological analysis of nanostructured material images obtained by Field Emission Gun Scanning Electron Microscopy (FEG‑SEM). Traditional 2‑D texture descriptors often fail to capture the intricate surface complexity present at the nanoscale. To overcome this limitation, the authors adopt a volumetric fractal descriptor (VFD) approach based on the Bouligand‑Minkowski fractal dimension, which treats the gray‑level image as a 3‑D mathematical surface and evaluates its geometric properties through Minkowski dilation.

Theoretical foundation
The Bouligand‑Minkowski dimension is estimated by dilating each point of the surface with a spherical structuring element of radius r and measuring the resulting volume V(r). The relationship V(r) ∝ r^{3‑D} yields the fractal dimension D from the slope of a log‑log plot. Instead of a single scalar, the VFD records V(r) for a set of radii, producing a multi‑scale feature vector that preserves information from fine‑scale roughness to coarse‑scale morphology.

Pre‑processing and feature extraction
FEG‑SEM images (256 × 256 px) are first denoised with a Gaussian filter, then histogram‑equalized and normalized to a 0‑255 intensity range. Each pixel (x, y) is mapped to a 3‑D coordinate (x, y, I(x,y)), creating a point cloud that represents the surface. For a series of radii r = 1,…,10 px, Minkowski dilation is performed, and the corresponding volumes V(r) are recorded, yielding a 10‑dimensional VFD per image. Principal Component Analysis (PCA) reduces dimensionality before classification.

Experimental design
The authors fabricate anodic titanium oxide (TiO₂) layers under a factorial design varying three process parameters: applied current (0.5–2 mA), oxalic acid concentration (0.05–0.20 M), and solution temperature (10–30 °C). This results in 48 distinct conditions. For each condition, multiple FEG‑SEM micrographs are captured, pre‑processed, and transformed into VFDs.

Classification and performance
Support Vector Machines (SVM) and k‑Nearest Neighbors (k‑NN) are trained on the PCA‑reduced VFDs to discriminate between the different process settings. Using 5‑fold cross‑validation, the SVM achieves an average accuracy of 94.3 % while k‑NN reaches 91.7 %. By comparison, conventional texture descriptors such as Gray Level Co‑occurrence Matrix (GLCM) and Local Binary Patterns (LBP) yield accuracies below 80 % for the same task, demonstrating the superior discriminative power of the fractal‑based approach.

Physical interpretation of fractal metrics
The estimated fractal dimension D correlates with the manufacturing parameters. Higher currents increase D (from ~2.45 to ~2.71), indicating a rougher, more complex surface. Lower oxalic acid concentrations reduce D, reflecting a more uniform nanostructure. Temperature variations produce subtler changes in D but significantly affect the scale‑dependent volume growth rates, revealing temperature‑induced re‑organization of nanoscale features. Thus, the VFD not only serves as a classifier but also provides physically meaningful insight into the relationship between process conditions and surface morphology.

Limitations and future work
The choice of dilation radii influences the descriptor’s sensitivity; an adaptive radius selection scheme is suggested for future studies. Computational cost is non‑trivial because volumetric dilation is performed for each radius; GPU‑accelerated implementations could enable near‑real‑time analysis. Extending the methodology to other nanomaterials (e.g., metallic nanowires, carbon nanotubes) will test its generality. Integration with advanced machine‑learning pipelines could facilitate closed‑loop process optimization and quality control in nanomanufacturing.

Conclusion
The study demonstrates that volumetric fractal descriptors, grounded in the Bouligand‑Minkowski dimension, effectively capture the complex morphology of nanostructured TiO₂ surfaces imaged by FEG‑SEM. The multi‑scale nature of the descriptors yields high classification accuracy across varied anodization conditions and offers a quantitative link between fractal geometry and physical processing parameters. This approach represents a promising tool for the characterization, monitoring, and optimization of nanomaterial fabrication processes.