傅立叶描述子是分析和识别物体形状的重要方法之一．利用基于曲线多边形近似的连续傅立叶变换方法 计算傅立叶描述子，并通过形状的主方向消除边界起始点相位影响的方法，定义了新的具有旋转、平移和尺度不变 性的归一化傅立叶描述子．与使用离散傅立叶变换和模归一化的传统傅立叶描述子相比，新的归一化傅立叶描述 子同时保留了模与相位特性，因此能够更好地识别物体的形状．实验表明这种新的归一化傅立叶描述子比传统的 傅立叶描述子能够更加高效、准确地识别物体的形状．-Abstract Shape is a visual feature which contains intrinsic high-level semantics, and has a great application value in CBIR(Content-Based Image Retrieval) and IR(Image Recognition). There are many descr iptors for shape feature. Fourier descr iptor predigests 2-demensional image information to 1-demensional signal and be used widely. In fact, the shape of natural image is often messy and noisy. So, this paper proposes a preprocessing method which can clean the noisy shape image, and then researches and analyses the shape feature extraction with Fourier descr iptor method with an experiment.