An Efficient Junction Detection Approach for Mobile-Captured Golf Scorecard Images

Mobile-captured images are usually severely degraded due to the presence of geometrical distortion, non-uniform lighting and noise, such that table detection becomes more challenging compared to that in scanned images. To improve the performance of junction detection in junction based table detection algorithms, we propose a two-step table junction detection method for mobile-captured golf scorecard images based on observations. The first step is table boundary detection, which aims to detect and extract the table boundaries, or table ruling-lines. We efficiently detect the boundaries based on observations that the table boundaries constitute the only single maximum connected component in the luminance image. The second step is junction detection, which tries to capture table junctions at the intersections of horizontal and vertical table boundaries. A novel pattern matching technique is employed to capture the junctions on the line-enhanced skeleton of table ruling-lines. To further reduce computational power, down-sampling is performed on the high resolution scorecard images before table boundary detection, and the table region is cropped out for junction detection. As there is no available application specific image set, we build one which contains an amount of mobile-captured golf scorecard images with diversity in scorecard forms and mobile brands. Experimental results show that the proposed method can successfully locate tables and detect junctions on the mobile-captured golf scorecard image set.