Natural vs. manmade scene classification using statistics of straight lines

Classification of scenes along the semantic categories has received tremendous attention from researchers working in the field of computer vision. The content and the context information obtained from scenes at various levels of granularity have been used to solve the problem of classification of scenes. We propose a simple approach for classifying the scenes on the broader semantic lines of categories, which are natural and manmade (or artificial) scenes. Our approach is based on the observation that at a primitive level of visual processing of scenes, the presence of large number of straight line segments is more discriminative in deciding whether the scene is natural or manmade. We extract and encode the information about the straight line segments as a descriptor and use it to classify the scene as natural or manmade. Then, we compare our descriptor with the common descriptors like HSV (Hue, Saturation and Value) Histogram and Edge Orientation Histograms (EOH).