基于统计量模板的半调图像特征提取与分类

Feature Extraction and Classification of Halftone Image Based on Statistics Template

为了实现误差分散半调图像的分类,提出了基于统计量模板的半调图像特征提取与分类方法。利用像素对的概念和统计量模板的特征描述方法,提出了基于分块的特征提取算法。提出了类特征矩阵概念;通过建立误差目标函数和利用梯度下降法来求取最优类特征矩阵,以描述半调图像的类别;探讨了最优类特征矩阵的特性。实验中,与其他类似方法进行了分类性能比较,探讨了参数对分类性能的影响,分析了特征提取算法的时间复杂度。大量实验比较和分析表明,提出的方法是有效的。

Feature extraction and classification method was presented based on statistics template for classifying half- tone images produced by various error diffusion methods. Statistics template was described as the descriptor of texture feature of halftone image according to the definition of pixel pairs, and a feature extraction method was presented based on image patches. The ideas of class feature matrices was proposed acting as the descriptor of category and then the op- timization problem was formulized by establishing error object function and utilizing gradient descent method to seek the optimal class feature matrices. The characteristics of class feature matrices were discussed by experiments. In experi- ments, the performance comparisons of our method with two similar methods were conducted. The influences of parame- ter on classification performance were also discussed and time complexity of feature extraction algorithm was analysed. Experimental results demonstrate that the proposed method is effective.