A Research Of Hashing Based Person Re-identification

With the popularity and development of Internet technology,tens of thousands of images are generated every day.If manual methods are used to identify and classify such large-scale data,it will inevitably consume huge human and financial resources and bring unnecessary waste of resources.Therefore,people efficiently use computers to help handle images for cost savings.However,due to the large size of the data,it is difficult for general algorithm to handle large-scale classification.To overcome above disadvantages,people proposes the method to increase the operation speed by using the XOR operation instead of the general addition,subtraction,multiplication and division operation,which is named hashing technology.Recent vision and learning studies show that learning compact hash codes can facilitate massive data processing with significantly reduced storage and computation.A lot of hashing methods based on hand-crafted features have been proposed,which significantly improve the speed of image retrieval without losing too much precision.In recent years,with the development of deep learning technology,it has been found that the deep features based on neural networks have better effects than the hand-crafted features of traditional methods.Therefore,learning deep hash functions has greatly improved the retrieval performance,typically under the semantic supervision.In contrast,current unsupervised deep hashing algorithms can hardly achieve satisfactory performance due to either the relaxed optimization or absence of similarity-sensitive objective.Person re-identification is one of image classifications,which uses computer vision technology to determine whether a specific pedestrian exists in an image or video sequence.Given a image of surveillance pedestrian,people can retrieve the pedestrian across the devices,which is designed to compensate for the visual limitations of the currently fixed camera,and can be combined with pedestrian detection and pedestrian tracking technology,and it can be widely applied to intelligent video surveillance and intelligent security and other fields.Person re-identification has become a hot and challenging research topic in the field of computer vision because of the following problems,like the difference between different camera devices,the combination of rigidity and flexibility of the pedestrian,and the appearance is susceptible to wearing,scale,occlusion,posture and viewing angle and so on.However,since the general person re-identification technique relies too much on semantic labels,and most of the state-of-the-art person re-identification algorithms are based on convolutional neural networks.However,the general neural network has a deeper number of layers,too many floating point calculations and the output of the final layer is too long.These situations lead to the slower speed of the current person re-identification algorithms.Neural networks rely on semantic labels,but the data with semantic labels in real world is often a small part.Therefore,how to quickly identify pedestrian without any labels has become a popular problem.A typical method is to utilize the advantages of image retrieval with hashing to complete the person re-identification task,which can greatly reduce the time complexity and space complexity of the retrieval and recognition tasks.Adopting these type of methods can achieve the real-time monitoring tasks in real world.In this work,we propose a simple yet effective unsupervised hashing framework,which alternatingly proceeds over three training modules: deep hash model training,similarity graph updating and binary code optimization.The key difference from the widelyused two-step hashing method is that the output representations of the learned deep model help update the similarity graph matrix,which is then used to improve the subsequent code optimization.In addition,for producing high-quality binary codes,we devise an effective discrete optimization algorithm which can directly handle the binary constraints with a general hashing loss.Extensive experiments validate the efficacy of the proposed method,which consistently outperforms the state-of-the-arts by large gaps.