中文摘要：

图像噪声类型识别是IC图像检测的重点和难点。针对准确获取IC图像噪声特征的要求，对噪声图像进行能量熵的推导和计算，并在此基础上形成能量熵分布特征平面，引入二维Zernike矩以对常见噪声类型进行特征值的量化及提取。建立BP网络对噪声样本特征值进行反复训练与调试，最终达到快速准确识别IC图像噪声类型的目的。通过试验计算获得了准确有效的IC图像噪声类型识别结果，并将其与其他典型图像噪声识别方法进行性能指标比较与分析，证明了新方法具有更好的准确性和可靠性。为IC图像的去噪和检测提供了理论基础与技术准备。

英文摘要：

Recognition of image noise types has become a difficult problem in IC detection. For the purpose of accurate acquiring the noise characteristics of IC image, the energy entropy of its noise image has been deduced and calculated, thus the characteristic plane of energy entropy distribution can be obtained with the introduction of two- dimensional Zernike moments by realizing the quantization and extraction of common noise type＇s characteristics as well. On the base of establishing BP neural network, it can be shown that this network is trained and adjusted repeatedly with those computed characteristic values of image noise, until the target of accurate and quick recognizing image noises was completed. Through experiments and computations a series of accurate and effective recognizing results of IC image noise types have been obtained, and their precision and reliability can be proved on the ground of the result comparison and data analysis with several other typical methods of image noise recognition in the areas of technical performances. This research provides a theoretical foundation and technical preparation for the noise removal in IC image pattern detection.