Semi-Supervised Multitask Learning for Scene Recognition

摘要: Scene recognition has been widely studied to understand visual information from the level of objects and their relationships. Toward scene recognition, many methods have proposed. They, however, encounter difficulty improve accuracy, mainly due two limitations: 1) lack analysis intrinsic relationships across different scales, say, initial input its down-sampled versions 2) existence redundant features. This paper develops a semi-supervised learning mechanism reduce above limitations. To address first limitation, we propose multitask model integrate images resolutions. For second build sparse feature selection-based manifold regularization (SFSMR) select optimal preserve underlying structure data. SFSMR coordinates advantages selection regulation. Finally, link SFSMR, method Experimental results report improvements accuracy in recognition.