The paper presents a method for classifying real-world scenes into natural and manmade categories based on depth by analyzing local texture information using a texture unit matrix. It employs both K-nearest neighbor (KNN) and self-organizing map (SOM) classifiers for supervised and unsupervised learning, which provide efficient online classification with low computational complexity. The study details the significance of image depth on texture and establishes a framework for classifying images while discussing the potential for real-time applications.