Research Of Large Scale Automatic Image Annotation Based On Deep Learning

Automatic image annotation is one of the most important challenges in computer vision,which is critical to many real-world researches and applications.Automatic image annotation is one of the most important challenges in computer vision,which is critical to many real-world researches and applications.Considering the weakness of traditional methods for large-scale data processing,in most cases,it does not meet the reality of large data processing.But the deep learning model has a unique advantage over large-scale data processing.So many researchers tend to use deep learning model to solve the problem of large scale automatic image annotation.A variety of deep learning models for various processing environments have been proposed in succession.However,there are still some problems,such as the construction of the model and the identification of the label.In this dissertation,aiming at the problem of automatic annotation,a detailed analysis and in-depth study of a variety of deep learning models is presented.The automatic image annotation methods based on these models and our own design models are proposed,and its effectiveness and efficiency are verified through experiments.The main contents are as follows:1.The research background and current situation of image annotation technology are expounded.The image annotation theory and key technology based on deep learning model are deeply studied.Many classical annotation algorithms are described in detail,including traditional methods and deep learning based methods.And the characteristics of these algorithms are analyzed.2.Firstly,considering the existing image data,especially the network images,most of the labels of themselves are inaccurate or imprecise.Dissertation proposes a Multitask Voting(MV)method,which can improve the accuracy of original annotation to a certain extent,thereby enhancing the training effect of the model.Secondly,the MV method can also achieve the adaptive label,whereas most existing methods pre-specify the number of tags to be selected.Additionally,based on convolutional neural network,a large scale image annotation model MVAIACNN isconstructed.Finally,we evaluate the performance with experiments on the MIRFlickr25K and NUS-WIDE datasets,and compare with other methods,demonstrating the effectiveness of the MVAIACNN.3.The lack of learning ability for a single model and the inability to fully consider the relevance between labels,an automatic image annotation method based on two deep learning models is proposed in this dissertation.First,in order to ensure the effectiveness of the two models,dissertation chooses Faster R-CNN as one of the models,because it is the best deep learning way of target detection based on region.In addition,due to the accumulation of the research group in the research of automatic image annotation model,and the good results obtained by the previous AIACNN model,so take it as another model of cooperative training.Secondly,inspired by collaborative training,a cooperative training algorithm is proposed to make full use of the relevance between labels.Finally,through experimental comparison and verification,this method achieves good results.