Randomized Voting Hough Transform Algorithm and Its Application

Line detection using Hough transform is one of the robust image processing methods for noisy image. But Hough transform has a problem that the computation cost is very large. In order to ease this problem, many high-speed algorithms were proposed. Xu and Oja proposed RHT (Randomized Hough Transform) which reduces the computation cost by selecting the pair of the edge points at random. Independently, Kiryati proposed probabilistic Hough transform (PHT) to reduce the computation cost by voting a part of the edge points in the image to the parameter space. We proposed a new high-speed algorithm called RVHT (Randomized Voting Hough Transform) which combines RHT and PHT. And we discussed experimentally and theoretically about the computation cost and the performance of RVHT by comparing with the specially regulated RHT and the original RHT. Since the proof that RVHT is faster than RHT did not completed, we present the complete proof in this paper. And we discuss the detailed properties of the algorithm of RVHT combined the edge deletion methods, the edge selection methods and the parameter space initialization methods. We show the differences between these combinations in the experiments of RVHT line detection. The best result of RVHT is 14 times faster than RHT. We applied this RVHT to detect circles from the microscopic image of salad dressing for the taste quality control. We got the feasible results by using the algorithm of RVHT with over 3 times faster than algorithm using the normal Hough transform. And we will be able to use the Hough transform circle detection to the practical applications.