Research On Hash Coding Method For Large-scale Image Retrieval

As the basis of human vision,image is an objective reflection of natural scenery,so it is the most commonly information carrier in human social activities.With the rapid development of the internet technology,the advent of the era of big data,and the popularity of image collection tools,there are more and more images in life,and people's requirements for large-scale image retrieval are also increasing.It's a challenge work to retrieve on such vast amount of images,since the number of images is large,and the feature dimension is high,and the search time need to be reduced.This makes contentbased image retrieval face new challenges in new situations.The brute-force search can get good accuracy,but it's extremely time-consuming and low memory usage efficiency.The tree-based method shows promising performance when processing low-dimensional data,however,the performance of tree indexing drastically degrades for highdimensional data.Therefore,a hash-based image retrieval method has been proposed.The hash image retrieval method encodes the features of the image as a hash code and uses the Hamming distance to calculate the similarity of the image.The hashing method can greatly reduce the memory consumption of the computer and the retrieval response time,so that it can be better used for large-scale image retrieval.This thesis has done a lot of research on hash coding methods for large-scale image retrieval.The main work and contributions of the thesis are summarized as follows:On the one hand,in order to reduce the loss of information during hash image retrieval,an image retrieval algorithm with minimum loss hash is designed.Firstly,the original high-dimension data is reduced by combining principal component analysis with Laplacian feature mapping,and then the hash function is obtained by minimizing the feature dimension reduction and the binary quantization loss function.Then,the original data matrix is converted into a hash code,and the similarity of the image is obtained by calculating the Hamming distance between the hash codes.The proposed method is compared with related methods on four datasets respectively.The experimental results show that the proposed method improves the retrieval performance.On the other hand,in order to solve the problems that single-bit quantization can't protect the feature's neighbor structure and the Manhattan quantification is extremely time-consuming,a double-bit efficient quantized method using bit operation is designed.This method uses a new encoding method,adding position information to binary numbers without having to use conversion decimal operations.This algorithm designs a finely tuned Hamming distance metric method for distance calculation to increase the calculation speed.It is to make full use of computer's efficient bit computing capability in distance calculations.The algorithm is compared with the existing quantification methods on the four datasets.The experimental results show that the quantification method does improve the retrieval performance.