Multi-scale Image Fusion

Separating spatial components lets the fusion process treat broad intensity patterns differently from localized edges and textures. Coarse components can preserve large-scale structure, while fine components retain detail that might be weakened when images are combined directly. This multiresolution organization helps the reconstructed result maintain both overall scene information and small features important for engineering inspection or analysis.

Fusion rules determine which information is emphasized at each spatial scale. Weighted averaging blends corresponding components, selection chooses information from one input, and activity measures favor components showing stronger local detail or variation. The chosen rule therefore affects the balance between smooth intensity representation, feature visibility, edge preservation, and the contribution of each source image.

Scale selection controls the level at which complementary information is compared and merged. Broad scales support the representation of large structures and intensity relationships, whereas finer scales emphasize edges, textures, and localized features. Using multiple scales is especially valuable when source images differ in exposure, wavelength, sensor characteristics, or imaging conditions, because their useful information may appear at different spatial levels.

A typical workflow begins by decomposing the input images with a multiresolution transform or pyramid. Corresponding coarse and fine components are then combined using a selected rule, such as averaging, component selection, or an activity-based decision. Finally, the fused components are reconstructed into one image, whose contrast, detail, and feature visibility can be evaluated for the intended engineering task.

Implementation requires complementary input images or image features, a method for multiscale decomposition, rules for combining corresponding components, and a reconstruction operation. Multiresolution transforms and pyramids provide the computational structure, while weighted averaging, selection, or activity measures provide the merging logic. Together, these elements determine how source information is retained in the final representation.

Engineering applications include machine vision, remote sensing, medical imaging, surveillance, and industrial inspection. The method can combine information from different sensors, exposure levels, wavelengths, or imaging conditions so that relevant structures become more visible. Improved contrast, preserved edges, and retained textures can support more reliable image analysis, feature assessment, and decision-making in these settings.