An innovative implementation of Circular Hough Transform using eigenvalues of Covariance Matrix for detecting circles

In this paper, a fast and accurate algorithm for identifying circular objects in images is proposed. The presented method is a robust, fast and optimized adaption of Circular Hough Transform (CHT), Eigenvalues of Covariance Matrix and K-means clustering techniques. Results are greatly improved by implementing iterative K-means clustering algorithm and establishing an exponential growth instead of updating values in the parameter space of CHT through summation, both in runtime and quality. In fact, using the Eigenvalues of Covariance Matrix as a validating method, a well balanced compromise between the speed and accuracy of results is achieved. This method is tested on several real world images with different circular objects within them; furthermore, as shown in the experimental results, it has been proved to be noticeably robust against noise.