Image Segmentation and Restoration Using Switching State-Space Model and Variational Bayesian Method.
Image Segmentation and Restoration Using Switching State-Space Model and Variational Bayesian Method.
复制标题
使用切换状态空间模型和变分贝叶斯方法进行图像分割和恢复。
DOI：
10.1143/jpsj.81.094802
复制
发表时间：
2012
期刊：
影响因子：
--
通讯作者：
Seiji Miyoshi
中科院分区：
文献类型：
--
作者：
Ryota Hasegawa;Ken Takiyama;Masato Okada;Seiji Miyoshi
我们推导出一个确定性的算法，恢复和分割的图像使用切换状态空间模型和变分贝叶斯方法。该算法在估计超参数的同时，通过贝叶斯推理来推断原始图像和潜在变量。图像的平滑度被认为取决于区域。这里，平滑度表示原始图像中的每个区域是平滑还是粗糙的程度。该算法的新奇在于它能够估计控制原始图像每个区域平滑度的超参数。同时，还估计了控制观测或传输过程中噪声的超参数。通过实验，使用人工图像和自然图像退化的高斯噪声，我们表明，派生的算法具有潜在的能力，使恢复和分割，只有一个嘈杂的图像。
We derive a deterministic algorithm that restores and segments an image using a switching state-space model and a variational Bayesian method. This algorithm estimates hyperparameters as well as infers the original image and latent variables by Bayesian inference. The smoothness of an image is considered to depend on the region. Here, the smoothness indicates what degree each region in the original image is smooth or rough. The novelty of the proposed algorithm is its ability to estimate hyperparameters that control the smoothness of each region of the original image. The hyperparameter that controls noise added in the observation or transmission process is also estimated. Through experiments using artificial images and a natural image degraded by Gaussian noise, we show that the derived algorithm has the potential ability to enable restoration and segmentation from only one noisy image.