抵抗仿射变换攻击的图像归一化方法研究

Research on The Method of Normalization Resistant to Affine Transformation

本文提出了一种基于矩的图像归一化方法。首先,通过图像分割提取出目标图像,消除了图像平移的影响;然后,通过计算图像的矩分别求出剪切变换参数β、缩放变换参数α和δ以及旋转变换参数φ。鉴于缩放变换参数α和β过小,导致归一化后的图像不适合后续处理,本文对缩放变换参数α和δ进行了调整。最后,提取出包含变换图像的最小矩形区域,并将其调整为统一大小。实验表明,本文方法有效、易操作。

This paper presents a method of image normalization based on image moments. First of all, in order to eliminate the influence of image translation, the smallest rectangle encircling the target object image is picked up by making use of image segmentation, so the background pixels are removed as many as possible; Then the shearing transformation parameter, scale transformation parameter α and δ and rotation transformation parameter φ are obtained by calculating image moments of different orders and corresponding transformations are successively applied to the original image. The scale transformation parameters α and δ are adjusted in this paper ？because they are so small that the transformed image is not suitable for post processing. At last, sub-image is obtained by choosing the smallest rectangle encircling the transformed image and the sub-image is adjusted to the sleeping size. Experimental results approve that the method presented in this paper is effective and prone to manipulate.