The fractal analysis of images is a common technique to study natural processes. Its application to landscape photographs, however, requires pre-processing which can produce widely different results. Using two categories of landscape scenes with different levels of vegetation, five types of image segmentation were compared and the resulting structures analysed with the box-counting method. The fractal dimensions estimated were compared and found to characterize either the naturalness or the complexity of the landscape scenes. This demonstrates the potential of fractal analysis for landscape perception studies, and offers a set of reliable methods to estimate the fractal dimensions of landscape photographs.