Modification of hough transform for object recognition using a 2-dimensional array

摘要

In this paper, a technique which transforms Ballard's 4-dimensional (4D) parameters (X, Y, S, Θ) to a 2D format {(x1, y1), (x2, y2)}, namely 2-point parameterization, is introduced. This technique leads to new voting and detection rules for object recognition using Hough transform. Due to the 2-point parameterization technique, three algorithms namely Rotation Invariant Hough Transform (RHT), Scale Invariant Hough Transform (SHT) and Rotation Scale Invariant Hough Transform (RSHT) are proposed for object recognition to handle: (1) when the scale (S) is known and (X, Y, Θ) are unknown, (2) when the orientation (Θ) is known and (X, Y, S) are unknown, and (3) when all (X, Y, S, Θ) are unknown. These proposed methods require only a 2-dimensional array for the accumulator, which greatly reduces the memory required in using Hough transform for object recognition.