Rapid Scene Categorization Using Novel Gist Model
Xianglin Meng, Zhengzhi Wang · 2010
Scene recognition poses great challenges due to large intra-class variations. We present a novel visual descriptor to build scene gist.It is an extended version of census transform histogram.The proposed gist model is more robust and has better generalizability.It is a holistic scene-centered representation that bypasses the segmentation and the processing of individual objects or regions.We experimentally demonstrate that the gist model outperforms state-of-the-art methods in scene categorization tasks. Also,it is computationally efficient and consistent with rapid scene categorization ability of humans.