The Application And Research Of The SPA In Trademark Image Retrieval

With the increasing of the transnational,and trans-regional growth of the global economy,the number of trade marks is increasing too in recent years.In our country, according to the statistics,the number of the application for trademark registration was 270 thousands in 2001,and this number came to a double in 2004,as much as 588 thousands.By the end the June in the year 2005,the number of the application for trademark registration had achieved as much as 2370 thousands.The greatly increase of the number of the application for trademark registration takes trouble into the trademark review work,but also a higher request.The current commercial databases retrieve information based on word-matching algorithm.They call for the images and video information with artificial word marked,and then realizing the automatic information retrieval according to the word marked.But when the amount of the image and video information is too huge,the cost of the artificial word marked will be unbearable.The method of trademark retrieval based on artificial encode cannot meet the request of the real-time and accuracy,so there is an urgent demand to establish a more accurate and efficient trademark automatic retrieval system.SPA is a kind of soft computing method been concerned by the academia in recent years.It can analyses and processes the uncertainty information effectively,and has been successfully used in area of decision-making,prediction,data integration, uncertainty reasoning,production design,networking plane,comprehensive evaluation and so on in recent years.But the SPA is less used in image retrieval,so the major research of this paper is to study SPA deeply,and use its considering certainties and uncertainties as a certain-uncertain system and making success in the area of Comprehensive Evaluation to improve the trade mark image retrieval system.And a trade mark image retrieval system based on integrated Features search is designed. The major study is as following:(1)A new image edge detection algorithm based on SPA is presented in this paper. The experiment result indicates that,compared the algorithm presented in reference ,the new algorithm detects the image edge more accurately and clearly,and it has a stronger noise suppression capability.(2)Arming at the problem that the features of the trademark still keep invariant after rotation,scaling and changing of the initial point,an algorithm of relational back in image retrieval using SPA was presented in this paper.The algorithm first extracts five features,and initialized the weight of each feature,then modifies the weight until the feedback from users is over.Finally,a comparison between the algorithm presented in this paper and the algorithm using Hu moment as the feature vectors was made.The experiment indicates that the new algorithm has a good invariant property under translation scale and rotation and strong noise resistance.(3) In order to test whether the SPA can be used in the trade mark image retrieval system,a prototype trademark image retrieval system is designed.The intermediate variables can return more related information to adding the machine learning's training samples in relevance feedback,and then enhance the efficiency of data mining.