Research On Hash Learning And Its Application

With the development of Internet technology,data presents an explosive growth,which marks that we have entered the era of big data.Hash learning algorithm can map data with high clutter,large order of magnitude and high feature dimension into compact binary hash code.As a research hotspot in the field of machine learning,hash learning has been widely used in many fields.This paper mainly discusses the application of hash learning in image content integrity authentication and motion capture data retrieval.The main research contents and contributions are as follows:(1)This paper reviews the research status and application fields of various typical hash learning.(2)The existing methods of image content integrity authentication generally focus on the design of various new image feature extraction algorithms.However,regardless of the type of feature descriptors,their original feature expression ability is always limited.To solve this problem,this paper proposes an image content integrity authentication algorithm based on hash learning.This paper first designs a weighted large margin for manipulation classification(WLMMC)algorithm to achieve the Classification of image malicious and non-malicious operations.Then this paper adopts supervised personalized quantization(SPQ)to distinguish the influence of each dimension on classification performance.This strategy can generate hash codes with different bits for different types of images.Experimental results show that the combination of WLMMC algorithm and SPQ algorithm can effectively improve the performance of existing image authentication algorithms.(3)Most of the traditional motion feature extraction methods adopt manual design features and have high computational complexity.To solve this problem,this paper proposes a motion capture data retrieval algorithm based on depth hash learning.In the phase of feature extraction,this paper improves a hierarchical independently recurrent neural network(Ind RNN)model,implements the semantic characteristics of high quality human behavior information extraction.The model can retain not only the time information of motion capture data,but also the spatial information of motion capture data.At the same time,this paper combines deep learning and hash learning to realize the rapid retrieval of similar sequences in large-scale motion capture sequences.Experimental results show that the proposed algorithm can not only effectively improve the retrieval efficiency and accuracy of motion capture data,but also shorten the retrieval time.