Generalized hough transform: Fast detection for hybrid multi-circle and multi-rectangle

The solution to the issue of hybrid multi-circle and multi-rectangle detection is to firstly detect all circles from a digital image with Multi-Rectangle. When detecting multi-circle, this approach determined each possible circle by two random sample edge pixels and the third edge pixel searched on their perpendicular bisector. As two points are randomly picked and the third point of the possible circle is searched, invalid sample decreases because of the removing of isolated noises, half-link noises and the elimination of non-co-circle points. By affirming the possible circle for true circle in minimal area, invalid computation decreases. When rectangle is detected, any possible rectangle can be obtained by tracking close curve with rule given and can be affirmed whether it is true rectangle or not by the definition of rectangle. The experimental results demonstrate that the approach can implement well hybrid multi-circle and multi-rectangle detection and the detection speed of multi-circles is faster an order of magnitude than Randomized Circle Detection (RCD) proposed by Chen&Chung (2001) on condition of multi-circle, multi-rectangle and/or random levels of noise.