Deformable Grid in Image Recognition

摘要

Deformable template has been shown to significantly improve the performance of image recognition for difficult tasks such as character recognition, digit recognition, and trademark recognition etc. However, it is often time-consuming. Grid feature is a popular feature extraction scheme in image recognition. Here we propose deformable grid for image recognition. Unlike deform-able template where deformation is applied to image, deformation is applied to grid in deformable grid. Because the number of grid is much less than that of image, our method is very timesaving comparing to deformable template. The approximate equality of deformable template and deformable gird is also shown. We tested our method in two image recognition experiments, namely, trademark recognition and off-line Chinese character recognition. We obtained improvement in recognition rate by 6.0% in first experiment, and 5.8% in second one.