中文摘要：

现有指根点提取方法多是针对有较高摆放要求的手图像，当摆放方式没有严格遵守摆放要求时，便不能正确提取指根点。针对自然摆放的手图像，本文提出一种指根点提取方法。首先确定图像中能够穿过食指、中指、无名指和小拇指的检测列位置，并做出检测列的梯度图像；然后通过提取梯度图像中4个峰值与4个谷值的坐标，定位每个手指的上、下边界点坐标，再根据每相邻两个手指问的边界点距离大小判断出这两手指的摆放状态即张开或并拢；最后根据每相邻两手指不同的摆放状态采用基于Roof边缘线跟踪方法或基于Step边缘线跟踪方法进行跟踪，其跟踪终点作为对应的指根点。方法基于灰度图像，有效回避了图像二值化及轮廓提取操作，提高算法运行速度。利用自建小型自然摆放图库进行测试，结果表明，正确提取率可达97．14％，验证了方法的可行性和有效性；方法能够避免手指张开程度产生的影响，体现了方法的优越性。

英文摘要：

Current methods for finger valley point extraction are for hand images under higher placement requirements. Once the images placed are not strictly adhered to requirements, valley points can not be correctly extracted. So an extraction method for finger valley points under natural state is proposed in this paper. Firstly, identify the detection column of the image which can cross four fingers（forefinger, middle finger, ring finger, little finger） and make out the gradient image of detection column, and then locate the finger edge points by extracting four peak points and four valley points of the gradient image;according to the distances between every two adjacent fingers, the states that whether are closed or open can be adjusted; last, according to the states, corresponding line tracing methods. Roof tracing or Step tracing, are adopted to tracing lines, and the tracing end points are the corresponding finger valley points. Besides, this method based on gray image can avoid image binarization and contour extraction operation, improving the running speed of this algorithm. A self-built image database is used to test this algorithm, and the correct extraction rate is 97. 14%,proving it feasible and effective. And this method can avoid the impact of fingers＇ opening degree, proving it superior.