摘要人脸识别技术在近年得到了飞速的发展,然而,人脸识别的准确率至今仍是 一个难以解决的问题,其中的一个重要因素就是输入系统中的某些人脸图像品质 较差,导致人脸识别系统受到影响,因此,在将采集的人脸图像输入系统之前, 对图像品质所进行的评估就显得尤为重要。本文介绍了目前主流的人脸图像质量 评价方案,并对这些方案进行整理与改进,提出一种新的人脸图像品质评估算法, 对人脸的尺寸、姿态、对称性、对比度、亮度均进行了评估,给出了其 C#+EmguCV 的实现,并进行了一系列实验,最终结果证明,通过本文算法对图像进行评估后, 可有效避免品质较差的图像进入识别阶段,从而使人脸识别的准确率得到提高。 68525
毕业论文关键词 人脸图像 品质 客观评价
Title The Design and Implementation of a Face Image Evaluating Algorithm
Abstract
A great progress has been made in the field of face recognition these years, however, the accuracy of face recognition system remains a problem for both developers and users.One important factor among them is some face image with low quality haven’t been checked before they are inputted. So it is crucial to evaluate the image before enrolling them. This paper collects the popular solutions to evaluate face images, trys to combine them and make an improvement, to come up with an algorithm of Face Image Evaluating, implement it in C# and EmguCV. This paper also do a series of experiments, turns out that after being evaluated with this algorithm, the image with low-quality won’t be in the identification stages, and that will improve the accuracy of face recognition.
Keywords Face Image, Quality, Objective Evaluation
1 引言 1
1.1 研究背景 1
1.2 研究现状 1
1.3 本文结构 2
2 人脸品质评估概述 4
2.1 主观评价和客观评价 4
2.1.1 主观评价 4
2.1.2 客观评价 4
2.2 影响客观评价的常见因素 5
2.2.1 光照 5
2.2.2 佩戴饰物 5
2.2.3 表情及姿态 6
2.2.4 人脸的大小(人脸与采集设备的距离) 7
2.3 ISO/IEC CD 19794-5 标准 8
2.4 OpenCV 及 EmguCV 概述 9
2.4.1 OpenCV 9
2.4.2 Emgu CV