Rotated Image Matching and Correction Method Based on Improved Circular Projection

The key of rotated image matching is to determine the rotation invariant features. Accordingly, the circular projection algorithm is proposed. This paper presents an improved algorithm based on circular projection. Through improving the projection algorithm of radius and increasing the projection algorithm of rotation angle, this algorithm has the ability to match and rotation correction. Meanwhile, the algorithm for arbitrary rotation angle is effective, it is robust to the rotation and the noise, and it has high efficiency.

The Method of Character Recognition Based on Projection Transformation Combined with LS-SVM

A method of character feature extraction based on circular projection transformation is proposed. First, it transforms the Cartesian coordinates into polar coordinates using the centroid of character image as the pole. It makes circular projection calculation and generates character feature vector by transforming 2-D character image into 1-D projection curves. Based on circular projection transformation feature extraction, the least squares support vector machine (LS-SVM) is introduced into the small character set embossed concave-convex character recognition.It adopted LS-SVM training software in the experiment for the simulation of embossed concave-convex characters’ number set. It studied on the effect of different core function and multi-classifier method, and compared with the results of neural networks, pattern matching and other identification classification methods.The experiment results show that the character feature has the scale and rotation invariance by the method of circular projection transformation extraction. Which combined with LS-SVM method has high recognition rate and more practicability in recognition small character set .