Research On The Algorithm Of Image Retrieval Via Learning Based Hashing

Large-scale search methods are increasingly critical for many content-based visual analysis applications,among which hashing-based approximate nearest neighbor(ANN)search techniques have attracted broad interests due to their high efficiency in storage and retrieval.This thesis mainly studies the image retrieval algorithm via reconstrutive embeddings learning based hashing.Specifically,the work in this thesis includes three aspects,namely,the improvement of Binary Reconstructive Embeddings,the proposed Angular Reconstructive Embeddings,and the Collective Reconstructive Embeddings for cross-modal retrieval.Firstly,the method of Binary Reconstructive Embeddings(BRE)learns binary codes by a general and popular objective,which minimizes the reconstruction error between the Euclidean distances of original data pairs and the Hamming distances of the corresponding binary embeddings.However,the prohibitively expensive coordinate descent based optimization of the original BRE algorithm can hardly scale to large data.Inspired by the one-to-one correspondence between the Hamming distance and code inner product,in this thesis,I reformulate the objective function of BRE in a very simple form,which significantly reduces the optimization complexity.By further incorporating the code balance and uncorrelation criteria into the problem,I devise an efficient iterative optimization algorithm,which learns binary codes in an alternating manner.The experimental results show that FastBRE outperforms the original BRE algorithm as well as several recently proposed methods.Secondly,existing hashing works are commonly designed for measuring data similarity by the Euclidean distances.In this thesis,I focus on the problem of learning compact binary codes using the cosine similarity.Specifically,I proposed a novel Angular Reconstructive Embeddings(ARE)method,which aims at learning binary codes by minimizing the reconstruction error between the cosine similarities computed by original features and the resulting binary embeddings.Furthermore,I devise two efficient algorithms for optimizing proposed ARE in continuous and discrete manners,respectively.I extensively evaluate the proposed ARE on several large-scale image benchmarks.The results demonstrate that ARE outperforms several state-of-the-art methods.Lastly,this thesis studies the problem of cross-modal retrieval by hashing-based approximate nearest neighbor(ANN)search techniques.Most existing cross-modal hashing work mainly addresses the issue of multi-modal integration complexity using the same mapping and similarity calculation for data from different media types.Nonetheless,this may cause information loss during the mapping process due to overlooking the specifics of each individual modality.In this thesis,I propose a simple yet effective crossmodal hashing approach,which can simultaneously solve the heterogeneity and integration complexity of multi-modal data.To address the heterogeneity challenge,I propose to process heterogeneous types of data using different modality-specific models.Meanwhile,I unify the projections of text and image to the Hamming space into a common reconstructive embedding through rigid mathematical reformulation,which reduces the optimization complexity significantly.Comprehensive experiments show that the proposed CRE can achieve superior performance compared to the state-of-the-arts on several challenging cross-modal tasks.