尺度相关系数融合数字半色调算法  

Scale Correlation Coefficient Fusion Digital Halftoning Algorithm

作　　者：何自芬[1] 张印辉[1] 詹肇麟[2] 

机构地区：[1]昆明理工大学机电工程学院,云南昆明650500 [2]昆明理工大学材料科学与工程学院,云南昆明650093

出　　处：《光学学报》2014年第B12期139-148,共10页Acta Optica Sinica

基　　金：国家自然科学基金（61302173,60962007）

摘　　要：混合误差测度函数的建立是利用多尺度小波系数融合建立的边界误差测度函数和改进的K-means聚类法建立的区域误差测度函数。小波域多尺度信息通过二维离散小波变换方法得到，从空间域到小波域的变换为原来的二维图像提供了一种紧凑的多尺度表征形式。利用这一紧凑的表征形式从少数幅值较大的小波系数中捕获图像的关键特征。应用基于区域的直接二值搜索方法最小化初始图像和半色调图像的误差，得到最优数字半色调图像。实验结果表明，该算法与基于模型的数字半色调（LSMB）算法比较，半色调图像均方误差值（MSEv）值降低了0.127，峰值信噪比（PSNR）值提高了12.72 dB。对原图和半色调图像进行频域变换，通过分析半色调噪声频谱图发现，半色调噪声谱仅分布在中、高频段，无很强的周期分量，且局部的频谱分量从低到高逐步增加，人眼视觉不敏感。对4种不同像素尺度进行半色调后，该方法计算时间均小于LSMB算法，从这个角度来看，该算法具有较快的收敛速度。A novel approach to improve the halftoning image quality by a mixture distortion measure function that is a combination of edge measure funciton depending on the fusion multiscale wavelet information and region measure function depending on the improved K-means clustering method. The wavelet domain multiscale characterization of the original image is gained using two-dimension discrete wavelet transform. The boundary information of the image target is fused by the wavelet coefficients of the correlation between wavelet transform layer, which increases the pixel resolution scale. The key feature of images are captured from a few large amplitude of compact wavelet coefficients in the use of this form of representation. Optical digital halftone image is gained to minimize the error of the initial and the halftone image by direct binary search （DBS） method. Experimental results show that compared with the least squares model based （LSMB） algorithm, the mean square error （MSEv） performance for the proposed algorithm is decreased to 0.127,the peak signal noise ratio （PSNR） performance for the proposed algorithm is increased to 12.72 dB. Halftone image original frequency domain transformation, half-tone noise spectrum is only distributed in the middle and high frequency band, no strong periodic component, and the local spectral components gradually increase from low to high, the human eye is not sensitive to the visual. After halftone, on four different pixel scales, the proposed method computation time is less than LSMB algorithm, from this perspective, the proposed algorithm is faster convergence.

关 键 词：图像处理 数字半色调 尺度相关 小波系数融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]