Intelligent Eye for Landscape Image Extraction

摘要

In this paper, a semantic content extraction scheme is proposed to achieve the purpose of "the one thousand words in a picture". The whole semantic features used for the proposed method are constructed from some low-level features described in MPEG-7 standards, including Dominant Color Descriptor, Color Structure Descriptor, Homogeneous Texture Descriptor, and Edge Histogram Descriptor, and trained for some specific images of landscape. We look for high-level semantic attributes for city/landscape classification, and further classification of landscape images. The proposed system can detection some semantic content, such as the Mountain, the Coast, the Sunrise/Sunset and the Forest, of an image. The simulation results show that the proposed semantic features are helpful for the semantic content extraction system especially for the images of landscape.