图像采集系统帧序列上缺陷像素的识别

Abnormal pixel discrimination in series of image frames

给出了缺陷像素的定义。对由图像帧序列构成矩阵去除直流分量所得到的余量矩阵作主成分分解建立特征图像。对特征图像作方差分析并结合为像素所定义的Mahalanobis距离,建立起概率分布意义上的缺陷像素识别的数学模型。实验结果分析表明,在给定每帧图像所包含的像素数时,通过距离基准限及一个可调节常量的设置,该概率模型可以满足图像采集系统在不同精度要求下对包括图像采集帧上的瑕点、闪烁像素以及漂移像素等缺陷像素的识别。

The novel concept of abnormal pixels is defined. Eigen-image is developed using principal component decomposition of the residual matrix that is obtained by subtracting direct component from the original matrix of a series of image frames. The variance analysis of the eigen-images is carried out, which is combined with defining the Mahalanobis distance for pixels; and the model for abnormal pixel discrimination is established in terms of probability distribution. Analysis of the experimental results shows that the proposed model can discriminate various abnormal pixels for the image acquisition systems with different accuracy requirements effectively.