Image Retrieval Algorithm Research Based On Pulse Coupled Neural Network

Images acquisition and use become more and more convenient as the use of portable imaging device and internet development.People prefer more with its advantages such as informative,objective reflection of facts,convenient obtain,conducive to spread.Nowadays images has been widely used in news,medical,criminal investigation,security,engineering,geography,military and so on.How can satisfy user's demands accurately from the mass images quickly and efficiently has become a focus research.content-based image retrieval causes the majority of researchers' attention.PCNN is widely used in iamge processing with advantages of rotation,zooming and other operations.however,the results depend on the selected parameters' quality that used for PCNN.This paper pay attention to the feature extraction and classification for color image through PCNN,then convert the methods to experiment.The contents are as follows:Firstly,we proposed a novel parameters selection method based on evolutionaty learning for optimizing the patameters of PCNN to avoid parameters selecting problems.Choose some classified images from database and trained them in advance,introduced the Particle Swarm Optimization and restructed the fitness fuction to optimized the parameters used for retrieval.The retrieval results poforms well even in the untrained images.The retrieval results covinces the prosed method was better than experienced parameters on precision ratio,recall ratio and personal visual judgement.Secondly,Inspired by neuron pulse oscillation statistical characteristics in PCNN,after analyzing the PCNN neuron pulse statistical feature and the binary image density distribution feature theory,proposed the concept relative density time series to perform image features that can be effectively used for retrieval.the evaluations of subjective observations and objective indicators of several simulation experiments demonstrated that the lagorithmis robustness to image translation change,rotation angle and local density distribution,and reflects good retrieval performance compared with other retrieve methods.Thirdly,Takig limitations of the single feature in retrieval into consideration,and PCNN can not be applied to color images,this paper put forward a multi-feature retrieval algorithm base on PCNN which include the color feature and the relative density time sequence.Convert color image from RGB to IHS and extracted each component feature using chapter 3's method,then synthesize each vector.Experiments show the combination vector contains images information more effectively.Fourthly,after measuring the mean shift distance from image feature,proposed a classification method for retrieval.Calculate the centroid of the similar features,and the new centroid when a new image is added.Choose the added image's category after mesuring the mean shift distance of these two centroids.And proposed self updates new centroid to shorten the mistakes.The experiment shows the proposed method can work effectively.At last,after combining PCNN and IHS color space,relative density time sequence algorithm for feature extraction,The centroid distance classification algorithm is applied to the retrieval system.Compared with other methods,proposed method has the ability to identify the colorful images and has a relatively excellent retrieval performance.