论文部分内容阅读
人脸检测的至关重要的一个步骤是人眼识别,现在比较流行的人眼检测方法是基于灰度图像的方法和基于红外图像的方法。根据灰度图像进行人脸识别主要有以下三种方法模板匹配法、基于外观的方法和基于特征的方法。基于外观的方法,采用2分类的方法,图像分为眼睛和非眼睛部分,首先需要一个有效的分类器,比如支持向量或者神经网络,分类器进行大量的训练,然后将图像检测出人眼。本文主要讲述基于外观的方法,分类器采用卷积神经网络,设计一个级联的CNN结构检测图像中的人眼。
A crucial step in face detection is human eye recognition. Nowadays, the most popular eye detection methods are based on grayscale images and infrared images. According to the grayscale image face recognition mainly has the following three methods: template matching method, appearance-based method and feature-based method. Based on the appearance of the method, the use of 2 classification method, the image is divided into eye and non-eye parts, first of all need an effective classifier, such as support vector or neural network, the classifier to conduct a lot of training, and then detect the human eye. This paper mainly focuses on the appearance-based method. The classifier uses convolutional neural network to design a cascaded CNN structure to detect human eyes.