Research And Implementation Of Face Fast Retrieval Algorithm Based On Deep Hashing

While enjoying the convenience and development of the Internet,people are constantly exporting their own texts,images and voices for the rapid development of the Internet.How to find deep image information effectively in massive data is a difficult problem we are currently facing.In the mining of effective image information data,face retrieval technology based on image retrieval is developing rapidly due to the wide application of image retrieval technology.This technology has been widely used in social,livelihood and security fields.However,while the scale of images has grown exponentially with the development of the Internet,face retrieval technology faces enormous challenges in terms of retrieval accuracy and retrieval speed.How to retrieve the most similar face images in a large-scale face data set has become an urgent problem for industry and academia.The learning to hash method is widely used in the field of image retrieval because of-its fast retrieval speed and low memory usage.At the same time,due to the rapid development of deep learning,more researchers paid more attention to deep hashing methods.However,the existing learning to hash methods are designed to common image datasets such as dog or cats.Using the existing learning to hash method to face retrieval will exist some the expressing face feature problems.Meanwhile,the existing deep hash algorithm only pays attention to the learning of the similarity relationship between the pairs of images and ignores the semantic image information.Moreover,some learning to hash methods separate the feature learning and quantization into two steps,which will lead to some quantization error,seriously affecting the accuracy of face image retrieval.In view of these problems,this thesis improves the existing hashing method and proposes a joint learning method based on multiple objective optimization functions.It designs and implements a set of secondary face retrieval method,while improving the retrieval speed.It realizes high face retrieval accuracy on face dataset.Comparing with other hash learning methods,our hashing method achieves the highest retrieval accuracy in face retrieval.The specific work of this thesis is as follows.(1)We propose a deep hashing network for face retrieval.Combining with Resnet,Inception Resnet module,we design a deep hashing network for face hashing codes learning.Compared with the traditional hash network,our deep hashing network is deeper in depth.Meanwhile,we use different scale convolution kernel in different layer to learn robust face hashing feature.(2)We propose multiple hashing loss function to learn robust hashing codes.Our loss learns face semantic information,intra-similarity and inter-similarity of images to learn robust face hashing codes.Meanwhile,we use quantization loss to reduce quantization errors.(3)Based on the Hamming search,we propose a retrieval method based on the secondary ordering of floating-point features using the closest N face images.In the case of guaranteeing the retrieval speed,we use the secondary ordering on floating-point features to improve retrieval accuracy comparing with single hashing methods.