Abstract

The spatial layout of images plays a critical role in natural scene anal- ysis. Despite previous work, e.g., spatial pyramid matching, how to design op- timal spatial layout for scene classi fication remains an open problem due to the large variations of scene categories. This paper presents a novel image repre- sentation method, with the objective to characterize the image layout by vari- ous patterns, in the form of randomized spatial partition (RSP). The RSP-based image representation makes it possible to mine the most descriptive image lay- out pattern for each category of scenes, and then combine them by training a discriminative classi fier, i.e., the proposed ORSP classi fier. Besides RSP image representation, another powerful classi fier, called the BRSP classi fier, is also pro- posed. By weighting and boosting a sequence of various partition patterns, the BRSP classi fier is more robust to the intra-class variations hence leads to a more accurate classi fication. Both RSP-based classi fiers are tested on three publicly available scene datasets. The experimental results highlight the effectiveness of the proposed methods.