采集多人重叠图像中高精度分割方法研究

Research on Precision Segmentation Method for Multi-person Dverlapping Image Acquisition

研究随机采集多人重叠图像的分割问题,提高多人重叠图像分割完整性。随机采集的图像受到人员不可控的影响,在采集过程中很容易出现人员的重叠。传统的多人重叠图像分割,评估采用主观评估的方法,需要对图像质量进行评估后,再优化分割图像效果,加入了固定的前提条件,对随机采集图像的分割受到限制。提出一种基于计算机图形学的随机采集多人重叠图像的分割方法,通过计算机视觉识别技术,对多人重叠图像进行分割的奇异值特征提取,解决随机性的问题,提取多人重叠图像主要奇异值特征,进行图像质量评估,优化后期的评估效果。仿真结果表明,采用计算机图形学识别方法分割提取图像特征像点进行多人重叠图像质量评估,整体效果可以得到大幅提高。

This study focuses on the segmentation of multi-person overlapping image acquired randomly to improve the integrity of the multi-user overlapping image segmentation. Randomly collected images are affected by uncontrollable persons, and in the acquisition process it is prone to generate multi-person overlapping. This study proposesd an image segmentation method for random collection of multi-person overlapping image based on computer graphics. Computer visual recognition technology was used to extract the singular value feature of multi-person over- lapping image segmentation, solving the problem of randomness. The main singular value feature of multi-person overlapping image was extracted to evaluate image quality and optimize the assessment effectiveness in late stage. Simulation results show that using computer graphics method to extract the feature point of image segmentation and assess multi-person overlapping images quality, the overall effect can be greatly improved.