The Design And Implementation Of Image Retrival System Based On Deep Hashing

With the explosive growth of image and text data in the Internet,there is an increasing demand for the retrieval of multimedia data.Retrieval systems can help people quickly retrieve the multimedia data of interest,such as picture,text and so on.The existing retrieval system can meet the content-based image retrieval and the text-based retrieval requirements.However,It's difficult to meet the retrieval and storage requirements of the current massive image data.Hash methods are widely used by researchers to solve the problem of large-scale image retrieval because of their advantages of reducing the dimension of image features and improving the response speed.According to the way to extract features,hash method can be divided into two methods:the artificial based hash and the deeplearning based hash which also called deep hashing.Besides,there is a semantic gap between texts and images.The image retrieval system in the Internet faces great challenges.In this paper,we construct a triplet based deep convolution neural hashing algorithm to meet the content-based image retrieval requirement.The algorithm extracts image features while learns the image hash coding.Cross-modal deep hash algorithm can cross the semantic gap between texts and images so that images and texts can be retrieved from each other.this paper proposes a triple-based cross-modal hash model to meet the text-based retrieval requirement.These algorithms' loss functions take into account the regular items that reduce the quantization error,the ranking loss based on triplets and the classification error loss in Softmax classifier.Through experiment and analysis,we find that these two algorithms have better effect on the accuracy of retrieval than other hashing methods.This paper designs and implements an image retrieval system based on deep hashing.The image retrieval system can meet the large-scale multimedia data retrival requirments.This paper tests the performance of the system from three aspects:index storage space,retrieval effect evaluation and retrieval speed.The results show that our image retrieval system based on deep hashing has the characteristics of fast retrieval speed,small index space and high retrieval accuracy.This paper also designs a serious test cases which the test results are in line with the expected results.