基于类高斯模型的显微图像固点噪声定位算法

Fixed Noise Locating Algorithm Based on Quasi-Gaussian Model for Microscope Image

针对显微图像固点噪声的特点和技术要求,本文提出类高斯模型和相应的固点噪声检测方法。首先,利用显微图像序列的时间相关性,计算类高斯模型中图像序列相同位置像素彩色分量的均值和方差;然后,根据彩色分量均值图像用最大类间方差法计算彩色分量的自适应阈值,用均分法计算彩色分量方差的自适应阈值;再者,根据阈值对图像序列中的每个像素进行判定,由此获取潜在固点噪声;最后,通过连续迭代确定最终的固点噪声及其相应位置。实验结果表明,在不同背景与对象亮度基础上,该方法能稳定地检测出图像序列中的固点噪声。

An adaptive algorithm based on quasi-Gaussian model is proposed for locating fixed noise in microscope image. Firstly, by considering temporal correlation of microscope image sequence, mean and variance of each color component for pixels at the same location in image sequence are computed. Secondly, the maximizing inter-class variance （OTSU） method is used to obtain three adaptive mean thresholds of color components. Meanwhile, three adaptive variance thresholds are acquired by averaging minimum variance and maximum variance. Then, the thresholds are utilized to check each pixel in the image sequence to detect the candidates of fixed noise. After several iterations to eliminate false candidates, fixed noises are finally located. Experimental results show that the proposed algorithm can stably detect fixed noise in microscope image sequences even though the background and object luminance change a lot.