Multi-scale Contextual Image Labeling
Zhou Qua · Acta Automatica Sinica · 2014
This paper provides a novel method for image labeling by combining the local features and contextual cues in a multiple segmentation framework. Our main insight is that identifying a larger image region provides strong evidence for classifying the contained smaller ones. The proposed method weights the classification results of each image region at different levels using the Bayesian rules, which are obtained by a series of learned discriminative models based on bag of features. Multiple segmentation framework provides a robust representation,allowing a wide variety of cues to contribute to the confidence in each semantic label. Compared with previous methods, the algorithm achieves the state-of-the-art results and fastest implemental speed on the benchmark dataset.