Trademark Image Retrieval Using Region Zernike Moments

To solve the problems on the invariance property under translation, rotation, scale and the starting point existing in the shape features of trademarks, a new trademark image retrieval method based on region Zernike moments to extract the shape feature of images is presented. First, the object region of an image is extracted according to the circumcircle of object pixels. Then, the object is plotted out based on the concentric circles to produce a series of sub-images. Furthermore, the Zernike moments of sub-images are computed, and they are constructed into a feature vector for describing the shape of the trademark image. Finally, the similarity comparison between the images is drawn according to the Euclidean distance. Experiment results demonstrate that this method has good invariant property under translation, rotation and scale, and is insensitive to noise.