Based On Fourier Described Shape And Color Feature Trademark Retrieval

Trademark is the sign of the enterprise. Trademark image registered is the landmark for the enterprise to the modem and international.It was manual management for the trademark images database ago. The trademark image was registered by the manual retrieval when similar image did not find in the trademark database, otherwise the trademark was not registered. For the trademark image retrieval the registered trademark is directly depended on the person's estimation. With the computer development, management of trademark image database is used database management. Thus using computer to finish the trademark image retrieval is necessary for the trademark registered. At the registered, firstly the new trademark image must be input the computer and transform the digital image, then extracted the characters of image, using characters to retrieval trademark image and find similar images for man, finally man is determined the image is similar or not and give registered or not.In this paper, according to the trademark image characteristic: simple shape, single colors, regular size, I bring forward grade retrieved method for the trademark, based on Fourier described shape firstly trademark retrieval and based on colors features secondly retrieval to finish two trademark image retrieval.Firstly retrieval is based on the shape with Fourier operators to find similar trademark for the image's shape information. Before retrieval trademark must be passed preprocess, zoomed the regular size make the height and width is approximate 2 power N in scale in order to use Fast Fourier Transform, used median filter to get rid off the noise, using Robert operators to detect edges, finally using Fourier transform make the shape of spatial space to the frequency space and pick up the character and compare with the characters to get the similar images. Firstly retrieval is selected percent 20 similar trademark image to be the secondly retrieval images database.Secondly retrieval is based on the HSI model with the color features retrieval and mainly depended on the trademark's colors information. RGB color model transform the HSI color model in order to close the human vision and pick up the characters in the HSI color model to compare with each other finally using global histogram intersection distance to measure the similarity. Finally similarity above 0.7 is recalled to the user for the similar images. Two methods are used in the image retrieval is fully considered the shape and color of the trademark image to be the important.The result is satisfied for trademark image retrieval; the recall and precision is higher than using only one method; the calculation is not complex and efficiency is high.