基于矩和多分辨分析的图像识别

Image recognition based on moment and multiresolution analysis

应用矩和多分辨分析对图像进行特征提取,求取不同分辨率下的小波系数的均值、能量和方差,作为特征向量。这组特征向量是将图像的矩特征和小波特征结合形成小波矩特征,即反映了图像的全局信息,又反映了图像的局域性信息,并且具有旋转、平移和缩放不变性。该算法不但解决了图像识别中特征量随图像旋转、平移和缩放而变化的问题,而且提高了对近似物体的识别能力。最后简要介绍了仿真实验及结果,证实此算法能对飞机、舰船等目标进行有效的识别。

Multiresolution and moment are applied to exact the features of image, and the average value, energy and mean square deviation of wavelet coefficients are calculated to used as characteristics. These characteristics, which combine wavelet properties and moment properties, are invariant to translations, scaling and rotation, and embody both global and local information of image. This method effectively improves the capability of distinguishing similar images, and resolves the problems that characteristics of images will vary with translations, scaling and rotation. The experimental results are given at the end of the article, which show this algorithm can identify the aircraft and the warship objects effectively.