基于数字人图像分割和识别的新框架

目的：提供一种高层知识引导分割和底层处理向知识归纳相结合的自动分割识别框架,并将该方法有效应用于彩色图像的数字化可视人体数据集（Chinese Visible Human,CVH）内部结构的自动分割和识别中.方法：首先利用分类标签在CVH2图像中提取脑干部分的图像,用Otsu获取阈值进行图像增强,将脑干轮廓作为分割的初始化轮廓,运用Level Set方法来实现精细化的分割：利用周长和质心位置建立联合判别函数,并对分割结果进行分类识别.结果：该算法准确对CVH2数据集脑干内黑质、中脑水管进行自动分割和识别,通过Amira软件三维重建分割结果获得较好的重建效果,并与手动分割结果保持了较好的一致性.结论：该框架能够实现数字人图像自动的多目标精细分割和识别,为知识引导的自动图像分割和识别提供了新的方法.

A Novel Framework Based on Chinese Visible Human for Image Segmentation and Recognition

Objective： This article offers an automatic segmentation and recognition framework, combing knowledgenducing segmentation with image processing-inducing knowledge summary techniques and effectively applies the methodology to automatic segmentation and recognition in color images of CVH. Methods： First, extract the images of the brain stem from cvh2 images using sorted tags, then, conduct an image enhancement using the Otsu method to acquire the threshold, create the outline of the brain stem as an initial outline for segmentation and accomplish fine segmentation through the level set method. Use the perimeter and centroid location to establish a union criterion function for the sorted recognition of segmentation results. Results： The algorithm was tested on a Chinese Visible Human-2. Results showed that it had the ability of automatic segmentation and recognition on the substantia nigra and mesencephalic aqueduct and acquired an excellent effect through the three-dimensional reconstruction of segmentation results with the Amira software. This is consistent with the manual segmentation results. Conclusions The framework can accomplish automatic segmentation and recognition to the multi-object of CVH images. It provided a new knowledge-based method to automatic image segmentation and recognition.