Research On Image Retrieval Based On The Extreme Learning Machine

The key and the difficulty lies in the application of the technology is effective image feature selection and extraction,and using the characteristics data to implement image retrieval.But the traditional image retrieval method based on shape and color and texture is lacking the ability to retrieve because of the single selected characteristics.So this article proposed an image classification retrieval method what is extreme learning machine retrieval method based on differential evolution.On the basis of the traditional image retrieval methods.The method according to different types of images,select and extract a number of different image features for image retrieval tasks.First,this paper research the problem of interpolation and filtering in image preprocessing,and through this kind of pretreatment operation improved the quality of the original images.Then respectively to research the same type images and different types images of the improved extreme learning machine image retrieval problem.According to previous problem,this paper choose the HU invariant moment features,to build the improved extreme learning machine’s characteristic input,and in turn,experiments on VW logo image,Buick logo image and Peugeot logo image retrieval problem in the logo image library.In the experiments,the improved method of the logo image retrieval accuracy is as high as 95%,far higher than ordinary extreme learning machine retrieval accuracy of about 73%.Experiment results show that using the differential evolution method,which can better realize the extreme learning machine input and hidden layer weights for reasonable initialization of the offset value,thus improves the image retrieval performance of the traditional extreme learning machine.In order to improve retrieval accuracy,this paper proposes to add distance rate as a new feature,this retrieval algorithm makes the image retrieval accuracy up to 98%.Then take the image library of car brand images,flower images and buildings images etc.As research objects to research multi-type image retrieval problem.On the basis of HU moment invariant features.Article puts forward that design the biggest H component ratio characteristics of the target image,LBP texture features as a learning machine input,retrieval experiments.Similarly,the improved extreme learning machine algorithm is superior to the traditional algorithm of retrieval results have been achieved.Through the above research,the differential evolution algorithm is introduced into the extreme learning machine,can effectively enhance the robustness of the network;Through the reasonable selection of input characteristics,realize the extreme learning machine multiple feature fusion retrieval.In the same type image and multi-type image retrieval,the retrieval system can achieve higher retrieval accuracy.