Gray Level Co-occurrence

Both settings determine which neighboring pixel pairs enter the analysis. Changing the distance examines relationships across a different separation, while changing the direction samples another spatial orientation. Because the resulting matrix records pair frequencies for that selected relationship, the calculated contrast, homogeneity, energy, and correlation can change with the chosen settings. Engineers can therefore tailor measurements to the image pattern being evaluated.

Contrast, homogeneity, energy, and correlation provide different quantitative views of the gray-level pair patterns recorded in an image. Considering these features together gives an engineering system more than a single visual impression of a surface or material. Their numerical values support consistent comparisons between normal and defective structures during image-based inspection and monitoring.

The matrix converts visual patterns into measurable pair-frequency information that can be processed consistently. This supports objective comparison of image regions instead of relying only on subjective visual judgment. In engineering systems, the resulting numerical descriptors help distinguish acceptable and defective structures, making them useful for automated inspection, surface-finish evaluation, and fault detection.

First, select the pixel separation and spatial direction that define the neighboring relationship of interest. Next, construct the gray-level co-occurrence matrix by recording the frequencies of the corresponding gray-level pairs. Finally, calculate descriptors such as contrast, homogeneity, energy, and correlation from those frequencies, then use the resulting measurements to compare structures or identify differences.

Engineers should document the spatial distance and direction used to identify neighboring pixels, because these choices determine the relationships represented in the matrix. They should also apply the same analysis conditions when comparing images or monitoring a process. Consistent settings make the extracted descriptors more suitable for distinguishing normal patterns from defective ones over repeated evaluations.

Applications include automated inspection, surface-finish evaluation, material characterization, and fault detection. In these settings, image-derived texture descriptors provide quantitative evidence for comparing normal and defective structures. The approach also supports image-based monitoring in manufacturing and other technical systems, where consistent measurements can help identify changes that may not be captured reliably through visual assessment alone.