Y. Kusachi, A. Suzuki, K. Arakawa, and T. Yasuno (Japan)
Object recognition, principal component analysis, occlusion, complex background, specular reflection
We propose an object recognition method for developing an application that returns information about images when they are captured with a cellular phone. Our appearance-based method is able to model many objects and overcomes the effects of pattern deficit and noise addiction, which are problems in previous methods. Models are generated by applying a subspace method for the feature vectors, where a co-variance matrix for a large number of images, added to the original feature vectors with all considerable local noise, is calculable using only original feature vectors. Experimental results showed that our method achieved high recognition rates for severely damaged images with occlusion, specular reflection, and complex background. Our method was applied to information retrieval systems for cellular phones and achieved high recognition rates in real environments.
Important Links:
Go Back
