Font Size: a A A

Research On Image Retrieval Algorithm Based On Bottom Feature Fusion

Posted on:2018-10-12Degree:MasterType:Thesis
Country:ChinaCandidate:M H ZhaiFull Text:PDF
GTID:2348330518468266Subject:Computer software and theory
Abstract/Summary:PDF Full Text Request
With the popularity of Flickr,Facebook and other social networking sites,image resources are growing at an alarming rate.How to quickly and efficiently extract the resources from massive images by the users has become a key issue.Many scholars have carried out fruitful research on this issue and got great results especially in the aspects of low level features,such as color,texture,shape,spatial hierarchy and other single underlying features or in the comprehensive feature extraction algorithm.However,the algorithm ignores the global spatial information,which only extracts some local attributes of the image and makes the retrieval inaccurate.In the comprehensive feature extraction algorithms,most of the traditional methods extract attributes directly from the original image,which leads to the large feature vector and the high computational complexity.In order to solve the above problems,perceptual characteristics and spatial characteristics of the image are proposed in this thesis,so that the underlying characteristics of the image can be integrated effectively.The underlying characteristics of the fusion can show the characteristics of the image information as much as possible.The retrieval accuracy and efficiency are improved at the same time.Specific research results are as follows:(1)In extracting the underlying feature of the image,there are usually a large number of high-dimensional features,which not only increase the computational complexity but also affect the accuracy of the underlying features to some extent.Therefore,in extracting the bottom features,we reduce the dimension by quantifying the image features.This method can improve the efficiency and accuracy of the retrieval.(2)The human visual perception mechanism is more sensitive than the spatial structure information of the image,especially when the image of the hierarchical information and angle change information are very sensitive.Based on this,an algorithm is proposed to make the hierarchical classification of the color features,change the angle of the primitive texture features,and then integrate the above two features.Among them,color histogram is used for the micro-part of the color feature to characterize the ratio of the pixels of each color to the entire image;color entropy and bit plane entropy are used separately for the macroscopic part of the image,where the bit-plane entropy is used in the stratification of the color characteristics of the most obvious first four layers,and each layer of the plane entropy is weighted;then,according to the definition of the five basic texture structure elements of the color pixel and angle of the statistical information,combined with color characteristics,to achieve image retrieval.The experimental results show that the algorithm can effectively show the spatial structure information of the image and improve the accuracy and recall rate.(3)As the local features can show more information of the details of the image,according to the color,the texture characteristics of the image and the role of expression,a new image feature descriptor is proposed,named as the multi-moment structure descriptor(MRSD).MRSD defines three structural descriptors based on the spatial structure of texture primitives.A 7-4-2-1 weighting method is used to highlight the different functions of each structural descriptor in feature expression,which can express the local features of the image more deeply.Experiments show that the HSV color model and MRSD fusion can better express the feature information of the image,which solves the problem of insufficient information of single feature expression.
Keywords/Search Tags:Search image, Bottom feature fusion, Feature dimensionality reduction, Bit-plane entropy, Multi-rectangle structure descriptors
PDF Full Text Request
Related items