SCAPEVIEWER: A Semi-Automatic System For Landscape Quality Classification And Landscape Indices Selection

A computerized system (SCAPEVIEWER) based on landscape indices and artificial intelligence techniques is proposed in order to classify landscape images into various categories according to the quality of the pictured scenery. Photographs (totally 108) from various landscapes have been clustered, from seven experts in accordance to the mean estimation of landscape quality, into distinctive (C1), typical or common (C2), and indistinctive (C3). For each of the landscape photographs, ten (10) environmental and operational indices have been selected and determined, corresponding to visibility (V), naturalness (N), relief (R), flora quantity (FC), flora quality (F), water (W), skyline (SL), visual entities (G), season (S), and viewer position (P). These landscape indices are fed to a classifier. The classifier is a feed forward Neural Network (NN) trained by a novel hybrid method combining genetic algorithms and the back-propagation algorithm with adaptive learning rate and momentum. A total classification rate of 93.5% has been achieved, using only 8 input neurons. The discarded input neurons correspond to the indices S, and P. The pilot system SCAPEVIEVER has shown the feasibility of landscape photograph classification into three categories (C1, C2, and C3) according to the landscape quality, selecting simultaneously the most robust indices, by means of a computerized system combining the knowledge of an expert with a NN based classifier.