【作者】 杨姝；

【导师】 高立群；

【作者基本信息】 东北大学 ， 模式识别与智能系统， 2005， 博士

【摘要】 本文首先概要地介绍了足迹检验理论与技术的现状、应用和未来的发展方向，接着重点讨论了赤足足迹的结构特征、测量方法及其在足迹检验中的重要作用；然后主要介绍数字图像处理和模式识别的基本概念、基本理论、基本方法及它们在实际中的应用；最后重点讨论了对赤足足迹图像的自动处理和识别方法。 它主要包括以下五个方面： 第一，根据足迹图像的特点，提出了基于多尺度形态重构的足迹图像滤波算法。该法首先定义了一个作用于灰度图像的且不具有幂等性的连通算子，这个算子可作为多尺度滤波准则：然后用最大树结构来描述灰度图像的平面区域及其之间的相互关系，按照定义的准则实现对灰度图像的滤波。由于形态重构滤波仅通过连通算子合并平面区域和改变它们的灰度值，因此，在滤除噪声和简化图像的同时，起到保护图像边缘或轮廓的作用，尤其是多尺度形态重构滤波可以滤除不同尺度空间中的噪声，因此滤波效果更好。最后，通过对最大树的重构，输出滤波后的图像。大量实验也表明，这种方法在足迹图像的滤波中，取得非常好的效果。 第二，根据足迹图像的特点，在足迹图像被滤波的基础上，提出了基于灰度-梯度二维阈值向量区域分割的足迹边缘提取方法。该方法利用自动生成的灰度-梯度二维阈值向量对图像进行分割，具有抗噪能力强和正确分割模糊边缘像素的特点，在提高图像分割质量同时，提高了边缘提取精度。实验表明，用该方法提取的边缘定位准确、精度高，取得令人满意的效果。 第三，为了提高以灰度-梯度为模型的二维最大熵阈值法的运算速度，本文还根据shannon熵函数在等概率场下取到最大值的性质，对二维最大熵阈值法中熵函数进行了优化，得到形式简洁、意义明确的新目标函数；用该函数选取阈值只涉及到减法运算，避免了二维最大熵阈值法中的对数与乘积运算，从而提高了运算速度。理论和实验都证明该法所求阈值与二维最大熵阈值法所求阈值完全相同并有更快的运算速度。它是一种保持二维最大熵阈值法对图像分割效果不变的阈更多还原

【Abstract】 In this paper, the footprint test theory and technique’s development and application and the future developing direction is summarily introduced firstly, and the footprint’s structure characteristic and measure methods and its important action in footprint test are emphatically discussed, then the basic concept, basic theory and basic method of digital image processing and pattern recognition and its application in practice are mainly introduced, the automatic processing and recognition method of footprint are emphatically discussed at last.It mainly includes the following five aspects:First, according to the character of footprint image, the multi-scale morphological reconstructing filtering algorithm based on area condition dilatation is proposed. In this algorithm, a connected operation which acts on gray image and isn’t idempotent, this operation can be considered as multi-scale filtering rule. Then utilizing the biggest tree structure describes the plane area of gray image and their correlation, according to the defined rule to filter the gray image. Because morphological reconstructing filtering combine plane areas and change their gray value only through connected operation, so, when filtering noise and simplifying image, the edges and contours of image are protected well, especially multi-scale morphological reconstructing filtering can filter the noises in different scale space, hence the filtering effect is better. At last, through the reconstruction to the biggest tree, the filtered image is output. Lots of experiments show that this method can get very good effect in footprint image filtering.Second, according to the characteristics of footprint image, the footprint edge extracting method based on gray level-gradient 2-D threshold vector image segmentation is presented.This method enables the segmentation of images by using auto-developed gray level-gradient 2-D threshold vector ,and acquires high anti-noise更多还原