Research On Image Attribute Learning With Ontology Fused Lasso

With the big-data era approaching, the large-scale web images bring out a great challenge.to image understanding and retrieval. Great progress has been made partly thanks to the adoption of techniques from machine learning and the development of better probabilistic representations. But the traditional pure statistical learning methods always ignore prior knowledge. Extended from them, we propose to augment the statistical learning methods with ontology.Due to the describable or human-nameable nature of visual attributes, the attribute-based methods have been receiving much attention in recent years in many applications. We propose an ontology-guided fused lasso model and apply this model for attribute feature selection and attribute learning. Our method first finds the semantic correlation with the ontology-guided attribute space(The WordNet space in this paper) and constructs a graph model of inter-attribute with a path-based method, then the graph model is combined with graph-guided fused lasso and the new model encourages highly correlated attributes to share a common set of relevant low-level features and transfer the learned common structure from the source image set to the target image set.The hierarchy of ImageNet is exploited to define the image attributes and a dataset from ImageNet including over 30,000 images is collected. The SIFT(Scale-Invariant Feature Transform) features is extracted and the Bo VW(Bag-of-Visual-Words) model is exploited. Multiple classifiers such as kNN( k Nearest Neighbors) and SVM(Support Vector Machine) are used to verify the result. We find that the proposed model has lower time complexity comparing with traditional methods. The experimental results show that this method can both improve the accuracy and accelerate the algorithm convergence. Hence, our method is more conducive to attribute learning comparing with the traditional methods.