Image Similarity Algorithm Research On Urban Scene

With the rapid rise of the mobile Internet,more and more image information resources show a rapid increase on exponential series.The frequency of the use of image search engine is also getting higher and higher.However,with the increase of image data,needs of the user are more and more diversified.Such as: how to use the Internet retrieve useful image resources;how to use the image similarity algorithm for other scenes,etc.These problems are worth to be studied at home and abroad.In order to improve the accuracy and time complexity of the algorithm,this thesis improves the algorithm.The improved image similarity algorithm is also applied to the urban scene monitoring system.The main work of this thesis is as follows:In this thesis,we firstly study the theory of SVD.We can find the relevant theoretical knowledge of SVD,which is also applied to the field of image similarity algorithm.Main research ideas: firstly,the singular value of the image is extracted,and then computing image similarity.During the experiment,extracting the singular value of the image is very time consuming,and with the increase of the image size,the time complexity of the algorithm is increased.In this thesis,we use a block strategy.First of all,the whole image is divided into blocks,and then the block of each sub image singular value isextracted,and finally to each sub image singular value is combined into a whole singular value vector,and then carrying out the similarity metric.In this thesis,we design a more detailed image segmentation strategy.Based on the singular value decomposition,the specific steps of image similarity are designed.The accuracy and reliability of the algorithm are verified by the experimental results.Experimental results show that this method can improve the time consuming of extracting image singular value and improve the overall efficiency of image similarity.Secondly,in this thesis,the basic features of the image are studied.Then the similarity of image is studied based on the color feature of the image.In the course of the study found that the traditional color histogram algorithm just described what is the quantity of a certain color in this image,and is not an accurate description of the image for each pixel in the image distribution.In order to make up for the deficiency of the color histogram algorithm,this thesis improves the image color histogram algorithm based on the texture feature of the image.In this thesis,the texture feature and color feature of the image are fused to calculate the similarity.This thesis gives the detailed algorithm design and the similarity measure formula.Finally,the two groups of experimental images are carried out,and the improved algorithm is carried out many times.Experimental results show that this method can make up the deficiency of the color histogram algorithm,so as to further improve the accuracy of the image similarity algorithm.Finally,the improved two image similarity algorithm is applied to the city scene monitoring system,and giving the overall system architecture diagram and the core module of the system are explained in detail.Based on a real scene to carry out a two set of experiments,which is to verify the accuracy and reliability of the improved algorithm.