摘要
针对Bayesian算法以误差分散核为先验、时空复杂度高的缺点,提出一种逆半调改进算法. 首先根据误差分散半调图的噪声特性设计去噪预处理器,然后以预处理图像为初始值,依据最大后验概率准则,采用基于矩阵运算的迭代方法估计逆半调图像. 所构造的逆半调算法与Bayesian算法相比,逆半调图像平滑且边缘清晰,时空复杂度大大降低. 仿真结果表明:N×N维图像的空间复杂度由8N2降至81N,运行时间降为原来的15%左右;采用Floyd-Steinberg 半调图,该算法的峰值信噪比(PSNR)与小波算法相当,采用Jarvis半调图,PSNR值较小波算法提高了0.3～3 dB.
To overcome Bayesian algorithm＇s shortcomings of requiring the knowledge of halftone kernel and high computational complexity and memory buffer, an improved method via maximum a posteriori was proposed. According to the characteristic of error-diffused halftone noise a denoising preprocessor was first designed to provide an initial image. Then the inverse halftoning image was obtained by updating the initial with matrix-based iteration scheme. Compared to the Bayesian algorithm, the resulting image of the proposed algorithm is smooth with sharp edges, while computational complexity and memory buffer are quite reduced. Experiments show that memory requirement is decreased from 8N2 to 81N and run time is reduced to 15% or so for an image of size N× N. The peak signal noise ratio performance for Floyd-Steinberg is almost the same as that of the wavelet algorithm, but for other kernels like Jarvis it is increased by 0.3 to 3 dB.
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
2005年第12期1340-1343,1357,共5页
Journal of Xi'an Jiaotong University
基金
陕西省自然科学基金资助项目(2001x06)
关键词
逆半调
最大后验概率
误差分散
去噪预处理器
inverse hal ftoning
maximum a posteriori
error diffusion
denoising preprocessor
