If photos are taken of plants in two treatments (top and bottom) with the same camera settings (all manual control) under the exact same lighting conditions it is possible to discern subtle differences in plant health from the numerical values of NDVI (not the colors of NDVI images).
A calibration routine using targets of known spectral reflectivity in each photo can relax the requirement that the color and intensity of light be identical for all photos.
Close-up photos of individual leaves will be best so the angle of the leaf and the angle of light can be controlled (identical in both treatments) and an average NDVI value can be computed from replicate areas of each replicate leaf on each replicate plant.
Capturing the required NIR and visible images with separate dedicated cameras will provide better results than using a single camera (one or the other channel [NIR, visible] will be cross-contaminated).
None of the above is easy to achieve.