A generalized regression neural network for logo recognition
Kathleen Zyga, R. Price, Bryan M. Williams · 2002
One of the primary concerns of document analysis systems is logo or trademark recognition, but few solutions proposed to date can deal with the problem of successfully classifying a logo that has been distorted in scale or rotation. We propose the use of a two-stage method applying a generalised regression neural network to provide the necessary flexibility to cope with these variations. A novel method of tiling which increases classification accuracy is also presented. The issues of scale and rotation are discussed in relation to the network's interpolation capability, as well as several other points effecting overall accuracy.