Recognition of Natural Scenery

摘要

It is important to recognize natural scenery that can be found outdoors. Because of the complexity of outdoor scenery, it is not successful to understand each region from the natural scene, although several scientists attempted to solve the problem. A neural network with a multi-layer neural network architecture has been constructed to label the visible objects in color images of outdoor scenes. The efficiency of segmentation is also very important for recognition with neural networks. Region growing method for segmentation was implemented rather than edge detection. The input data for the training with neural network can be separated as two kinds: The first group consists of color information such as hue and saturation. The second group consists of geometric information such as region size and region position. Labeling the visible objects in natural scenery has been up to 99.4% accurate in terms of the number of pixel numbers based on the training of a neural network to recognize objects from a set of examples.