Accurate Ellipse Extraction in Low-Quality Images
Zezhong Xu, Shibo Xu, Cheng Qian, Reinhard Klette · 2019
We propose an efficient method for extracting parameters of elliptic regions in digital images. Only one 2-dimensional accumulator array A(ρ, θ) is used for Hough voting, defined by voting angle θ and voting distance p. For each voting angle θ, the voting distance p is considered to be a stochastic variable, where voting values A(ρ, θ) define the probabilistic weights. We state how the statistical variance is related to the major axis, minor axis, and direction of an elliptic region, and also how the statistical mean is related to the centre; we provide two relationship functions in image space. After voting, a linear function and a quadratic function are fitted in the parameters space. The major axis, minor axis and direction are computed based on the coefficients of the fitted quadratic function. The centre is determined by using the coefficients of the fitted linear function. The proposed method is tested on synthetic images and real-world images. Experimental results show that the method extracts accurately parameters of elliptic regions, even in noisy and low-resolution images.