Development of a Novel Fast Rotation Angle Detection Algorithm using a Quasi-Rotation Invariant Feature Based on Sobel Edge

Abstract

In this paper, we proposed a fast algorithm to detect a rotated angle of a High Definition (HD) image that features an overall framed image without using a multiple iteration like the trigonometric function. The well-known method, the Coordinate Rotation Digital Computer (CORDIC) involves a simple shift-addition iterative procedure to perform rotation angle detection, which uses between two points only, causing an inefficient operation to process a certain image. In our algorithm, Sobel edge is used as a pre-process to simplify the information on the image in a gray scale form. Then, a binary conversion of the extracted image in a 1×n set of points that only depend on an angle of distribution on the same radius from the center of the image in an extreme line of the circular boundary. The set of features of the original and the rotated image, the rotation angle is evaluated for comparison. The detectable angle is limited only to an angle below 9 degrees in the side of its accuracy, but the execution time is about 11 times much faster in comparison to the method of rotation matrix based on CORDIC. It was simulated using Matlab R2012a and the testing environment was based on Intel Core i5 3.3GHz CPU.