| Since the 21 st century,with the popularity of the Internet,the rapid development of multimedia technology has been promoted,and massive amounts of image and video data have been produced.Image data has become an important part of Internet data.How to quickly retrieve useful information in a complex and large number of image databases is an urgent problem to be solved.Nowadays,due to the increasing amount of image data on the Internet and the increasing dimensionality of data,the traditional image retrieval methods based on tree indexes cannot meet the requirements.Image retrieval based on hash algorithms have been paid attention by researchers at home and abroad because of its fast calculation speed,low memory and good retrieval performance.The main research content of this thesis is image retrieval based on hash algorithm.Hash method has a good ability to reduce the dimension of data,so it can effectively and quickly retrieve useful information for large image databases.For image retrieval based on hash algorithm,this paper first introduces and comprehensively analyzes the existing hashing algorithm,then improves it and applies it to image retrieval.Firstly,for some hash algorithms,each dimension of the projection data is treated equally in the data processing,and ignore the information between the various dimensions of the data,so that the similarity between the data can not be preserved well.A weighted hash algorithm(WABQ)is proposed to improve the ABQ algorithm by combining the adaptive binary quantization(ABQ)method.After the data is reduced by PCA,calculate the variance of each dimension of the data after dimensionality reduction and form a weight matrix to balance the variance of each dimension of the original data after mapping the dimension data and to ensure that the amount of information for each dimension is approximately the same.Experimental results show that the proposed method improves the mean average precision(mAP)and recall rate to some extent,and also improves the training time of the data set compared with the original algorithm.Secondly,because it is difficult to establish the image retrieval index when extracting the local features from the image using the hash algorithm for image retrieval,the global features are extracted from the image,and each image is represented by a global feature when the improved algorithm is applied to image retrieval.Moreover,for some image retrieval methods based on hash algorithm,the color image is often change tothe grayscale image,and then the global features are extracted from the image,thereby ignoring the color information of the color image.This paper proposes an image retrieval method that combines color features and weighted hash method based on improved ABQ algorithm.The method can calculate the color distance and Hamming distance of each picture in the query image and image database separately,and the last retrieved distance is the weighted sum of the color distance and the Hamming distance.Experiments show that the retrieval performance is improved compared with the WABQ method.Finally,the thesis summarizes the work of this paper,and points out the future research direction according to the shortcomings of the proposed method. |
