Data-driven automated identification of optimal feature-representative images in infrared thermography using statistical and morphological descriptors

Infrared thermography (IRT) is a widely used non-destructive testing technique for detecting structural features, such as subsurface. Most IRT data post-processing methodologies result in the generation of image sequences in which defect visibility varies strongly across time, frequency, or coefficient/index domains, making the selection of defect-representative images a non-trivial and critical task. Conventional evaluation metrics, such as the signal-to-noise ratio or the Tanimoto criterion, often rely on prior knowledge of defect location or defect-free reference regions, which limits their applicability for automated and unsupervised analysis. In this work, a data-driven methodology is proposed to identify images within IRT datasets that are most likely to contain and represent features, especially anomalies and defects, without requiring any prior information about their spatial position. The investigation focuses on three complementary descriptors. First, the Homogeneity Index of Mixture quantifies statistical heterogeneity through deviations of local intensity distributions from a global reference distribution. The second descriptor is a Representative Elementary Area derived from a Minkowski-functional-based adaptation of the Representative Elementary Volume concept to two-dimensional images. Building upon these approaches, a third descriptor is introduced: a geometrical-topological Total Variation Energy index based on two-dimensional Minkowski functionals, designed to enhance sensitivity to localized anomalies. The proposed framework is validated experimentally using pulse-heated IRT data acquired from a carbon fiber-reinforced polymer plate containing six artificial defects at depths between 0.135 and 0.810 mm and is further supported by one-dimensional N-layer thermal model simulations. The results demonstrate that the proposed descriptors enable robust, unbiased ranking of image sequences and provide a reliable basis for automated defect-oriented image selection in IRT.