Used by companies, organizations, and even individuals to promote recognition of their brand, logos can also act as a valuable means of identifying the source of a document. E-business applications can retrieve and catalog products according to their logos. Governmental agencies can easily inspect goods using smart mobile devices that use logo recognition techniques. However, because logos are two-dimensional shapes of varying complexity, the recognition process can be challenging. Although promising results have been found for clean logos, they have not been as robust for noisy logos.

Logo Recognition: Theory and Practice is the first book to focus on logo recognition, especially under noisy conditions. Beginning with an introduction to fundamental concepts and methods in pattern and shape recognition, it surveys advances in logo recognition. The authors also propose a new logo recognition system that can be used under adverse conditions such as broken lines, added noise, and occlusion.

The proposed system introduces a novel polygonal approximation, a robust indexing scheme, and a new line segment Hausdorff distance (LHD) matching method that can handle more distortion and transformation types than previous techniques. In the first stage, raw logos are transformed into normalized line segment maps. In the second stage, effective line pattern features are used to index the database to generate a moderate number of likely models. In the third stage, an improved LHD measure screens and generates the best matches. A comprehensive overview of logo recognition, the book also presents successful applications of the technology and suggests directions for future research.

Jingying Chen, Lizhe Wang, Dan Chen

Introduction Motivation Shape recognition Proposed method Objectives Assumptions and input data Book organization

Preliminary knowledge Statistics Probability Random variable Expected value Variance and deviation Covariance and correlation Moment-generating function Fourier transform Structural and syntactic pattern recognition Introduction Grammar-based passing method Graph-based matching methods Neural network Architecture Learning process Summary

Review of shape recognition techniques 2D shape recognition Shape representation Shape recognition approaches Logo recognition Statistical approach Syntactic/structural approach Neural network Hybrid approach Polygonal approximation Indexing Matching Distance measure Hausdorff distance Summary

System overview Preprocessing Polygonal approximation Indexing Matching

Polygonal approximation Feature point detection overview Dynamic two-strip algorithm The proposed method Results Comparison with other methods Summary

Logo indexing Normalization Indexing Reference angle indexing (filter 1) Line orientation indexing (filters 2 and 3) Experimental results Summary

Logo matching Hausdorff distance Modified LHD (MLHD) Experimental results Matching results Degradation analysis Results analysis with respect to the LHD and the MHD Discussion and comparison with other methods Summary

Applications Mobile visual search with GetFugu Using logo recognition for anti-phishing and Internet brand monitoring The LogoTrace library Real-time vehicle logo recognition Summary

Conclusion Book summary Contribution Future work Book conclusion References

Appendix Test images Appendix Results of feature point detection

Index