Research On Image Retrieval Based On Fusion Feature Of AlexNet

Deep Learning originated in artificial neural network, it is a branch of machine learning. It has a Deep Learning process through multi-layer complex structures with complex data processing layer in the abstraction level modeling. Deep model itself is composed of some non-linear modules. Deep Learning's advantages are the use of an efficient unsupervised or supervised feature selection algorithm instead of doing the process by human. Content image retrieval problem is to search large databases by digital images. Content-based image retrieval, namely, search for image by image content. Content-based retrieval of visual information is an application for computer vision technology on image retrieval problem. With the rapid development of GPU and CUDA programming model, it is a hot topic of current research by using high-speed parallel GPU computing power and Deep Learning models for image retrieval.In order to apply the Deep Learning model to image retrieval better, the traditional image retrieval technology and the image retrieval technology in the Deep Learning field are introduced in this thesis. Research the image feature extraction capability of the deep model and the performance of the fusion deep feature in image retrieval. Combined with the characteristics of the Deep Learning model and the basic principles of improving the image retrieval effectiveness, experiments are carried out to improve the image retrieval results by applying the deep features. The hierarchical feature will be integrated to improve the effect of image retrieval. The Deep Learning framework Caffe and AlexNet model is used to extract the image features with the public image data sets: Inria Holidays, Oxford Buildings, Paris and UKBench to conduct the retrieval experiments to verify the effectiveness of the method and study the influence of different data sets. PCA algorithm is used to reduce the dimension of features and study the retrieval effect of the reduced features. The experimental results show that the fusion feature of AlexNet model can improve the effect of image retrieval. Using PCA algorithm to compress the fusion feature can effectively improve the retrieval efficiency without losing too much precision.