CAJ | 학술논문

为获取质量更好的视差图,在图像分割的基础上提出一种新的立体匹配算法,首先采用均值漂移算法对图像进行分割得到不同分割区域,然后采用基于窗口的局部自适应加权方法获取初始视差图,经双向检测后得到可信区域与不可信区域,最后定义一个加入遮挡项的能量函数作为遗传算法评价函数,计算可信区域的平面模型参数,之后在能量函数中加入平滑项,为每个区域选一个使能量函数最小的平面模型,得到最终的视差图.实验数据表明,该算法可较好地处理深度不连续区域和无纹理区域,匹配精度高.
위획취질량경호적시차도,재도상분할적기출상제출일충신적입체필배산법,수선채용균치표이산법대도상진행분할득도불동분할구역,연후채용기우창구적국부자괄응가권방법획취초시시차도,경쌍향검측후득도가신구역여불가신구역,최후정의일개가입차당항적능량함수작위유전산법평개함수,계산가신구역적평면모형삼수,지후재능량함수중가입평활항,위매개구역선일개사능량함수최소적평면모형,득도최종적시차도.실험수거표명,해산법가교호지처리심도불련속구역화무문리구역,필배정도고.
In order to obtain a better quality parallax image,this paper presents a new stereo matching algorithm.First,we segment the image pairs with the mean-shift algorithm to get the different segmentation regions.Then we use the locally adaptive weighting method based on window to obtain the initial disparity map,and get the reliable regions and unreliable ones after the crosscheck process.Finally,we define the energy functional which includes a occlusion item as the evaluation functional of the genetic algorithm,and use it to calculate the plane model parameters of the reliable regions,then we add a smooth energy item to the energy functional and choose a plane model for every region which could let the region has the smallest energy.Mter that we obtain the final disparity map.The experiment results show our algorithm has better performance for processing the occluded pixels and the no texture area,and has a higher matching accuracy.