Moudgollya Rhittwikraj, Sunaniya Arun Kumar, Midya Abhishek, Chakraborty Jayasree
Department of Electronics and Instrumentation Engineering, National Institute of Technology Silchar, Silchar, Assam, 788010, India.
Department of Surgery, Memorial Sloan Kettering Cancer Center, New York, NY, 10065 USA.
Optik (Stuttg). 2022 Jun;260. doi: 10.1016/j.ijleo.2022.168980. Epub 2022 Apr 1.
Background subtraction always remains an important and challenging task for different applications. Our previous work established the effectiveness of hybrid model by exploiting the oriented patterns present in a video sequences over other statistical method. To extend this approach further, we have proposed a novel approach herein by eliminating GLCM based features with an improved local Zernike moment and color components of intensity. These features are clubbed with the orientation based features extracted from angle co-occurrence matrices (ACMs) to model the background. Furthermore the Mahalanobis distance measure is replaced by Canberra distance to categorized foreground and background pixels, which significantly reduces the computational complexity of the proposed method due to the absence of covariance matrix measure. Comparative results have shown that our proposed method is effective than other competing method on different set of video sequences.
背景减除对于不同的应用而言始终是一项重要且具有挑战性的任务。我们之前的工作通过利用视频序列中存在的定向模式，相较于其他统计方法，确立了混合模型的有效性。为了进一步扩展这种方法，我们在此提出了一种新颖的方法，即通过改进的局部泽尼克矩和强度颜色分量来消除基于灰度共生矩阵（GLCM）的特征。这些特征与从角度共现矩阵（ACM）中提取的基于方向的特征相结合来对背景进行建模。此外，用堪培拉距离取代马氏距离来对前景和背景像素进行分类，由于不存在协方差矩阵度量，这显著降低了所提方法的计算复杂度。比较结果表明，我们提出的方法在不同的视频序列集上比其他竞争方法更有效。
