Research Of Trademark Retrieval Based On Color And Shape Feature

In recent years, with the rapid development of Chinese market economy,the trademark requirements increased year by year.The traditional trademark retrieval method based on Classification Codes is easy to implement but has strong subjectivity,meanwhile using lots of human resources and time.At the same time,it’s accuracy and real-time capability can not meets the requirement of people.For efficient prosecution rate and query accuracy degree, on the basis of computer vision,content-based image retrieval technology provides a good way to solve the problem of trademark registration, applying digital image processing and pattern recognition technology.Concerned more and more,it has become the hot spot of solutions for the image data problems at this moment.This paper analyses the current status and development trend of trademark retrieval, introduces the principle of the trademark image retrieval based on content, designs and implements a simulation experiment environment. With the robustness of color feature extraction in space and the robustness of shape feature in the displacement transformation, a multi-feature matching algorithm which is combined color features and shape features is used. In the respect of color feature, extract color histogram as image color feature values. While in terms of shape feature extraction, here use scale invariant feature transform algorithm. It utilizes the scale selection technique to build the Gaussian scale-space images achieving invariance to scale transformation. Though the Gaussian function is the unique continuous kernel that satisfies the causality property, the discretized Gaussian kernel used in practice no longer possesses the causality property. In addition, the Gaussian derivative kernels are not well matched to the local structures in images. Finally, the computational complexity increases as the scale gets larger, which makes the key point detectors not suitable for applications that have real-time constraints. convolution algorithm of B-spline function has an efficiency solution for the specialty of the way B-spline convolved and the specialty of discrete B-spline of order zero. Computation complexity of the convolution algorithm has nothing to do with the scale of B-spline function, but only depends on the signal or image data itself. B-splines are good approximations of the Gaussian kernel, so B-spline derived scale-space inherits most of the nice properties of the Gaussian-derived scale-space.As a result, traditional Gaussian-derived scale space was substituted by B-spline derived scale-space. To reduce the influence of computation of covariance matrix caused by noise, this paper applies improved Harris operator detecting trademark’s corner as interest points. At the mean time, matching points were mostly kept. Finally, trademark feature is got by multiple color and shape feature with weighting coefficient which is used to comprised feature distance between sample and libraries, then the system return the similar image.Systematic evaluation of the proposed shape feature extraction algorithm was made in this paper. Experiments demonstrated that our algorithm is rotation-invariant and insensitive to slight scale transform. Moreover, compared with other classic algorithm, such as SIFT and SURF, the experimental results also showed that the new algorithm has good detection, localization performance and good efficiency. Meanwhile if we can combine the features of the color and shape together, adjust the weighting coefficient λ between the two to ultimately determine the value of λ, the recall ratio and precision ratio will reach a relatively higher level.