贝塞尔曲线插值下的聚焦形貌恢复  

Shape from Focus Based on Bezier Curve Interpolation

作　　者：张明[1] 丁华[1] 刘建成 ZHANG Ming;DING Hua;LIU Jian-cheng(Institute of Mechanical Engineering,Taiyuan University of Technology,Shanxi Taiyuan 030024,China;School of Engineering and Computer Science,University of the Pacific,Stockton,CA,95211,United States)

机构地区：[1]太原理工大学机械工程学院,山西太原030024 [2]School of Engineering and Computer Science,University of the Pacific

出　　处：《机械设计与制造》2018年第9期175-177,181,共4页Machinery Design & Manufacture

基　　金：山西省煤机重点科技攻关项目(MJ2014-05-02);山西省自然科学基金(201601D011050);山西省研究生联合培养基地人才培养项目(2016JD13);太原理工大学校基金(1205-04020202)

摘　　要：聚焦形貌恢复性能提升的关键在于噪声序列图像的预处理以及初始深度值的连续化处理。通过对图像噪声类型识别理论及逼近技术的深入研究分析,实现了一种高精度的聚焦形貌恢复算法。该方法从含噪图像灰度直方图分布特性入手,在对序列图像未知噪声进行初步判断后,决定采用中值滤波进行图像预处理以减少噪声干扰;根据聚焦评价函数曲线特性,利用参数化的三次贝塞尔-伯因斯坦多项式对初始深度值进行插值处理,通过最值搜索方法得到插值曲线的峰值位置,进而实现深度值的精确估计。对不同表面特征,不同材质的实体对象进行形貌恢复实验,结果表明提出的方法兼具恢复的高精度性及有效性。The key to improving the performance of shape from focus lies in the pretreatment of the noise image sequence and the continuity of the initial depth map.Through deeply analyzing and researching the theory of identification of image noise type and approximation methods,a high accuracy of 3D shape recovery algorithm is realized.On one hand,it took cue from the distribution characteristics of noisy image’s gray histogram,made a pre-judgment of unknown noise and applied the median filter to reduce the noise.On the other hand,according to the properties of focus measure curve,a scheme for approximation of initial depth map through cubic degree Bezier-Bernstein polynomial interpolation is introduced,after the peak position of interpolating curve is determained,a more accurate depth value is obtained.Experimental results demonstrate that the recovery precision of this method is superior to traditional methods for some real objects with different material and surface details.

关 键 词：聚焦形貌恢复 灰度直方图 中值滤波 贝赛尔曲线 深度值 

分 类 号：TH16[机械工程—机械制造及自动化] TP391.41[自动化与计算机技术—计算机应用技术]