Research On Image Retrievalalgorithm Based On Feature Fusion

With the advent of the era of big data,more and more image information is pouring into people’s daily life.Although the technical level of image processing,image recognition,image retrieval and other research has been continuously improved,good results have been achieved in many fields,but there are still many shortcomings that need to be optimized.Therefore,on the basis of content-based image retrieval,this paper analyzes and studies the characteristics of images,and proposes to reconstruct the image features in image retrieval to solve the problems of multi-feature fusion.image retrieval method to improve the accuracy of image retrieval.The main contents of this paper are as follows:1.In order to solve the problem that affects the similarity calculation in multi-feature fusion retrieval,image feature reconstruction in image retrieval is proposed.First,through experimental research,it is found that the length of the feature vector and the correlation between the components of the feature vector in the multi-feature fusion retrieval are the factors that affect the similarity calculation.Then,for this problem,using principal component analysis,the image feature vector extracted by a given feature extraction method is analyzed,and the corresponding weight matrix is obtained.Finally,using the weight matrix and the corresponding feature extraction method,the image features in any image database can be reconstructed,and the image retrieval task can be well completed.The proposed feature reconstruction method,on the one hand,effectively solves the multi-feature fusion problem in image retrieval,and on the other hand analyzes the image feature extraction method using statistical methods,which has certain theoretical and practical significance.2.In order to improve the effect of multi-feature fusion method in image retrieval system,an image retrieval algorithm based on multi-scale texture feature fusion is proposed.First,use Gaussian pyramid to perform multi-scale feature extraction on the image to construct a multi-scale image.After that,for images of different scales,the grayscale co-occurrence matrix and the grayscale run-length matrix are used to extract texture features.Then,the extracted texture features are integrated,enhanced,optimized and fused to obtain fused texture features that can be used for image retrieval,and then image retrieval is performed.The image retrieval experiment is carried out on the database Corel-1k,and the accuracy and retrieval rate are used as the evaluation indicators of the image retrieval performance.