CAJ | 학술논문

采用多种深度线索获取二维视频深度，可以提高视频深度提取效果。针对运动线索，通过块匹配运动估计算法获得帧间深度信息；针对帧内线索，采用Laws滤波器提取深度特征，然后利用马尔可夫随机场模型训练获得帧内深度信息；在此基础上，提出了基于信息融合的二维视频深度提取方法。该方法首先通过计算背景和运动区域的信息熵，然后利用模式识别中的Neyman Pearson准则确定视频帧的运动区域。针对该区域，在帧内深度信息获取的基础上，融合帧间深度信息。实验结果和真实深度比较，证实了该方法的有效性。
채용다충심도선색획취이유시빈심도，가이제고시빈심도제취효과。침대운동선색，통과괴필배운동고계산법획득정간심도신식；침대정내선색，채용Laws려파기제취심도특정，연후이용마이가부수궤장모형훈련획득정내심도신식；재차기출상，제출료기우신식융합적이유시빈심도제취방법。해방법수선통과계산배경화운동구역적신식적，연후이용모식식별중적Neyman Pearson준칙학정시빈정적운동구역。침대해구역，재정내심도신식획취적기출상，융합정간심도신식。실험결과화진실심도비교，증실료해방법적유효성。
The two-dimensional video depth can be obtained by various depth cues.The in-terframe information was obtained by motion clue based on block matching motion estimation algorithm, and multidimensional features are extracted by Laws filter.The intraframe infor-mation was obtained by Markov random field model with depth features.A depth extraction method based on information fusion was proposed.The information entropy of background and motion region was calculated, the Neyman Pearson principle in pattern recognition was used to determine motion region.According to the motion region, the depth estimation in-traframe information was combined based on interframe information.The experimental results were compared with the true depth, which have proven the effectiveness of method.