Regions Labeling in Outdoor Scene Images

摘要

Outdoor scene analysis is a complex problem for both image processing and pattern recognition domains. This paper proposes an approach for labeling regions in outdoor scene images. The basic idea of this approach is to label local image regions into semantic objects such as tree, sky and road etc. There are four phases in the approach: segmentation, feature extraction, region labeling and merging. Firstly, modified Marker-Controlled Watershed (MCWS) algorithm proposes for segmented regions generation. And then, color feature extracted from segmented regions are given as input to 3-layer Artificial Neural Network (ANN) classifier for labeling. Finally, region merging is performed if the two regions are adjacent with the same color values. The proposed method is test on our real scene image dataset which are collected from our university campus.