Adaptive Fast Image Retrieval On Distributed Platform

With the rise of mobile internet,the popularity of new media,the explosive growth of various photographic and video equipment,people share various pictures on social media,and the number of online images has increased dramatically.How to retrieve similar images from these huge and rich image data has always been a hot issue in the field of image research.Image feature selection and retrieval efficiency are the key issues concerned by image retrieval researchers.Selecting appropriate image features is very important to the performance of image retrieval.Convolutional Neural Networks(CNN)is composed of more complex networks and has more powerful feature expression ability than traditional feature extraction methods.At the same time,the distributed computing platform Hadoop has good stability,security,scalability and other characteristics,which can accelerate the efficiency of image retrieval.The hidden layer output extracted by convolution neural network is used as image deep feature for image retrieval,and the image retrieval task is accelerated with the help of Hadoop's powerful computing power.The main work of this thesis can be summarized as follows:1.Pre-trained AlexNet network model is used to extract image deep features.The image retrieval task is carried out by combining single-layer deep features and multi-layer network features.Then compared with the traditional manual feature based image retrieval method,the FC6 layer feature is selected to construct the image feature index.2.In order to solve the problem that the dimension of deep feature is too high in the retrieval task,which brings a long retrieval time and a large amount of computation to the image retrieval task,an adaptive index construction method is proposed in this thesis.This method can accelerate the efficiency of image retrieval by using deep feature to build index and save it in HBase.3.We implemented image retrieval task on distributed platform Hadoop.The feature index is constructed according to the image features,and the image retrieval task is completed through the feature index.Experiments show that using Hadoop platform for image retrieval tasks on large datasets can effectively improve the efficiency of image retrieval.In this thesis,experiments are carried out on four open datasets Holiday,UKbench,Oxford Building and MIRFlickr1M,including image feature extraction,index construction and image retrieval.Experiments show that the efficiency and accuracy of image retrieval tasks can be improved by establishing image feature index and using distributed platform Hadoop.