为了更好地描述图像内部的复杂程度; 建立图像复杂度与各指标之间的数学模型是研究图像复杂度最关键的一步。首先从图像纹理出发; 试图建立图像复杂度与各指标之间定量、精确的数学关系描述。针对目前图像复杂度与各衡量指标之间没有明确的数学关系的特点; 文中采用灰度共生矩阵对纹理的主要特征参数进行分析; 提出了基于BP神经网络的图像复杂度评价方法; 建立了图像复杂度与各个指标之间非线性的数学评价模型。通过大量的图片对神经网络进行训练学习; 得到各指标的权重值。验证结果表明; 所建评价模型能够真实地反映图像内部的复杂程度; 获得的实验结果与人类视觉感知的结果基本一致。对于将BP神经网络应用于图像复杂度的研究具有一定的参考价值。

In order to better describe the internal complexity of image, the establishment of the mathematic model between the image complexity and each index is the key step to study the complexity of image. Firstly, starting from the image texture, we try to establish a quantitative and precise mathematical description of the relationship between the image and the complexity of various indicators. There is no clear mathematical relationship between the image complexity and the measurable indicators, so gray level co-occurrence matrix(GLCM) is used to analyze the main characteristic parameters of the texture. The image complexity evaluation method is proposed based on BP neural network. Then a nonlinear mathematical evaluation model between image complexity and each index is established. And the weight values and index are obtained by the training for the neural network and learning through numbers of pictures. The verification results show that the evaluation model is able to reflect the internal complexity of the image truly, and the experimental results obtained are consistent with human visual perception. It is of a certain reference value for the application of BP neural network to study the image complexity.