Zero-shot Learning Based On Deep Learning And Common Space

With the development of human society,image recognition becomes more and more important in human's life.In recent years,with the progress of information technology,the recognition algorithm has been more and more powerful.However,modern pattern recognition systems are often limited to insufficient train set due to the growth number of instance category.Using the data from other sources is a remedy,such as text data,to train visual models and predict unknown class labels.This method is called Zero-shot Learning.Based on the analysis of the current mainstream Zero-shot Learning algorithms,three new Zero-shot classification models are proposed in this paper:1.A Zero-shot Learning algorithm based on deep common space is proposed.With Deep Learning technology,the algorithm combines image feature extraction modal and semantic feature extraction modal.Also,it establishes an end-to-end deep image feature extraction and common space embedding model.Therefore,the algorithm can train the parameters of the image mode and the semantic modal simultaneously.2.A Zero-shot Learning algorithm based on structured deep common space is proposed.The algorithm noted that Zero-shot Learning problem is due to unconnected of the image train set and test set in class space.So the output features are considered as structured objects in this algorithm.With the introducing of nonlinear mapping framework by the structured embedding,test images can match better semantic features according to a compatibility function.Therefore,the algorithm has better generalization ability.3.A Zero-shot learning algorithm based on bidirectional deep common space is proposed.The algorithm focuses on the internal structure of visual and semantic features.Visual modal learns the latent common space from the visual features of supervised data with the maintaining of the internal structure.Semantic modal can learn a mapping to common space under the internal structure between the known class and the unknown class.Therefore,the algorithm can effectively preserve the prior knowledge of visual and semantic.Comparison experiments are performed on the three public databases(AwA,CUB and A-Pascal/A-Yahoo),and verify the feasibility and effectiveness of the proposed algorithms.