| Summary: | 碩士 === 國立臺灣科技大學 === 資訊工程系 === 103 === Detecting straight lines, circles, and ellipses from digital images is important and has a long history in pattern recognition and image processing community. In this paper, a novel fast unified method is proposed to efficiently detect the three shapes simultaneously from the images. In the proposed unified method, we first propose the arc segment-lookahead approach to filter the useful segments from the input image, then these filtered arc segments are allocated into the line bucket or curve bucket. The arc segments of each bucket are sorted based on their lengths. The proposed method selects the arc segment with the longest length as the seed to determine the resultant shape, leading to clear computation-saving and robustness merits. In order to speed up the circle/ellipse detection time, an efficient discrimination strategy is proposed to decide the detection order: detecting circle first or ellipse first. In addition, an array data structure is presented to be as the voting platform in the shape detection. Based on some synthesized and real images, experimental results demonstrated the computation-saving merit of the proposed method for detecting the above three shapes, while preserving the similar detection accuracy when compared with some related state-of-the-art methods.
|
|---|
