Recognition of Logo Images using Invariants Defined from Higher-order Spectra

This paper presents a system for the automatic recognition of logo images extracted from electronic document images. It uses feature extraction by use of higher-order spectra and classification by means of a Nearest Neighbour classifier. Results demonstrate the robustness of the technique for logo recognition where the images suffer from imaging defects, varying degrees of Gaussian, speckle and salt and pepper noise as well as slight variations in logo structure and inclusion of background artefacts. The technique has been applied to two independent databases and achieved recognition accuracies as high as 99.6%.

[1]  C. L. Nikias,et al.  Higher-order spectra analysis : a nonlinear signal processing framework , 1993 .

[2]  Henry S. Baird,et al.  The skew angle of printed documents , 1995 .

[3]  Vinod Chandran,et al.  Pattern Recognition Using Invariants Defined From Higher Order Spectra- One Dimensional Inputs , 1993, IEEE Trans. Signal Process..

[4]  Giovanni Soda,et al.  Logo Recognition by Recursive Neural Networks , 1997, GREC.

[5]  Henry S. Baird,et al.  Document image defect models , 1995 .

[6]  Tieniu Tan,et al.  A general algorithm for document skew angle estimation , 1997, Proceedings of International Conference on Image Processing.

[7]  B. Boashash,et al.  Pattern recognition using invariants defined from higher order spectra: 2-D image inputs , 1997, IEEE Trans. Image Process..

[8]  Francesca Cesarini,et al.  A neural-based architecture for spot-noisy logo recognition , 1997, Proceedings of the Fourth International Conference on Document Analysis and Recognition.