| Summary: | 碩士 === 國立臺灣科技大學 === 資訊工程系 === 98 === Circle-detection is an important issue in the image processing. In
general, the Hough transform is most common method for detecting
circles from digital images. However, given an input image, the Hough
transform consumes a lot of time to perform the voting process on a 3-D
accumulator array. To improve the execution-time performance of the
Hough transform, we first present curvature-based elimination strategy to
reduce the number of edge pixels considered in the voting process. Based
on the curvatures of edge pixels, we discard those edge pixels which have
lower probability of lying on circles and only a small amount of edge
pixels will be considered in the voting process. Further, the curvature of
each edge pixel is also considered in our proposed radius estimation
scheme to estimate the radius of the circle. Based on the estimated radius
and the gradient direction, each edge pixel only considers a range of
radius rather than considers all possible centers and radiuses to perform
the voting process. Combining the proposed curvature-based
elimination strategy and radius estimation scheme, the proposed fast
circle detection algorithm is presented. Experimental results demonstrate
that the proposed algorithm has 87.67% average execution-time
improvement ratio when compared to four currently published
algorithms.
|
|---|
