Abstract: To improve the effect of image enhancement and highlight the image subject, a new fractional-order differential image enhancement algorithm is proposed based on image characteristics division. By constructing a fractional-order differential mask operator, the different fractional orders are determined according to the results of image features segmentation, and a matrix with the orders is generated. The matrix is substituted into the mask operators to calculate with the original image. During the experiments, the original image and the image added Gaussian noise are processed respectively. In addition, the effect of enhanced images is compared in the case of different parameters. Experimental results show that the algorithm enhances the texture of the main part of the image significantly and restrains the increase of image noise in background and smooth area to a certain extent. This algorithm has a better performance on increasing image texture when the average image gradient is slightly lower than that of the traditional fractional-order differential image enhancement algorithm.

Key words: fractional-order differential, image enhancement, image characteristics division, image entropy, Peak Signal to Noise Ratio（PSNR）

摘要： 为了改善图像增强效果，突出图像主体，研究了一种基于图像特征分块的分数阶微分图像增强新算法。该算法通过构造分数阶微分掩模算子，根据图像特征分块的结果设定分数阶阶数，形成分数阶阶数矩阵，然后将其代入掩模算子，并与原图像进行运算。实验中分别对原图像和加入了高斯噪声的图像进行处理，并比较了不同参数时图像增强效果。实验结果表明，该算法能较大地增强图像主体部分的纹理，同时一定程度上抑制了背景及平滑区域图像噪声的增加。在图像平均梯度略低于传统分数阶微分增强算法的程度上，该算法对图像纹理的增强幅度更大。

关键词: 分数阶微分, 图像增强, 图像特征分块, 图像熵, 峰值信噪比