基于张量鲁棒主成分分析的视网膜眼底图像序列变化检测

在计算机辅助诊断系统中，视网膜眼底图像序列的变化检测是一项重要且具有挑战性的任务。针对视网膜眼底图像序列采样帧少、光照干扰大、难以获得稳健的背景模型，提出了一种基于张量鲁棒主成分分析(tensor robust principal component analysis, TRPCA)的变化检测方法。该方法以TRPCA为模型，通过对序列背景扩充，再利用张量分解而获得变化区域：首先，先选择出序列中最接近正常状态的一张图像作为背景模型；然后，通过预处理将单帧背景模型扩张成多帧背景使得背景模型包含更丰富的光照变化；接着，将整个序列建模为一个3维张量体；最后，利用总变分约束背景模型和变化区域的时空连续性，通过Tucker分解分离出背景模型，获得变化区域。实验结果表明，与基于矩阵的鲁棒主成分分析(matrix robust principal component analysis, Matrix RPCA)方法，Masked-RPCA方法以及不加总变分约束的TRPCA方法相比，基于总变分约束的TRPCA方法能够更准确地分离出变化区域，对血管和光照干扰更具有鲁棒性。

Change Detection Based on Tensor Robust Principal Component Analysis for Retinal Fundus Image Serial

The change detection of retinal fundus image serial is an important and challenging task in computer-aided diagnosis systems. Major challenges for retinal fundus image serial include few sampling frames and large interference of illumination, making it difficult to obtain a robust background model. A change detection method based on tensor robust principal component analysis (TRPCA) is proposed. The method takes TRPCA as the model, expands the serial background, and uses a tensor decomposition to obtain the change region:First, an image closest to the normal state in the serial is selected as the background model. Then, the single-frame background model is expanded into multi-frame backgrounds by pre-processing so that the background model contains more abundant illumination changes. The whole serial is modeled as a three-dimensional tensor volume. Finally, the time-space continuity of the background model and the change region is constrained by the total variation, and the background model was separated using Tucker decomposition to obtain the change region. The experimental results show that compared with the matrix robust principal component analysis (Matrix RPCA), masked-RPCA, and TRPCA methods without total variation constraints, the TRPCA method with total variation constraints more accurately separated the change region and is more robust to the interference of blood vessels and illumination.