Scene Classification via Hierarchical Semantic Blockes Vote Model

The contributions of image blocks to the holistic scene semantic classification are further exploited in this paper. An image is subdivided into non-overlapping regular grid of blocks hierarchically, 2 × 2 blocks at the first level and 3 × 3 blocks at the second level. For each level, “bag-of-features” strategy is deployed to predict the scene category of each block. Then the holistic scene category of an image can be recognized through a vote model based on the semantic categories of blocks at all levels in this image. Classification performance is compared to five state of the art approaches using their own datasets and testing protocols. In all cases, the proposed model achieves equal or superior results. Source codes are available by email.