Shape-based Trademark Image Retrieval System

Trademark is a symbol of the item set up by producers and business operators,and the product of commodity economy.It is not only a sign of the enterprise and product,but also the representation of the enterprise culture and the quality of the goods.So trademark plays an important role in the market economy.Until now there have been more than two million registered trademarks in China,but its management depends on the manual Operation rather than efficient and intelligent management system.So in order to create intuitive,accurate and fast trademark image retrieval,the content-based image retrieval technology is applied to the trademark image retrieval applications.Trademark image has three characteristics: color,shape and texture feature.Shape feature is closer to the high-level semantic feature than color and texture feature.Therefore in this paper shape feature is used to describe the trademark image.The main works are as follows:(1)Edge detection is the first research emphasis of this article.It is very important for trademark shape feature extraction,directly reflects the material and spatial properties of the trademark.This article puts forward an adaptive Canny edge detection algorithm.The method obtains the trademark edge detection threshold by using twice the between-cluster variance method,solving the problem of its excessive reliance on human experience.(2)Shape feature description is another key point in the research of this paper.It is the core of the trademark shape feature extraction.On the basis of the Shape Context,this article puts forward the Inner-Angle Shape Context.This method solves the trademark Shape is affected by the change of the trademark target area position easily.(3)Design and implement the trademark image retrieval system.System has the following three function modules: trademark pretreatment,trademark feature extraction and trademark retrieval,which can retrieve any trademark and return the most similar samples from the trademark database.