室外环境下红外热图像内眼角定位

针对室外环境下红外热图像中目标区域受背景过热与周围环境影响，导致目标边界模糊、噪声大等问题，提出了一种室外环境下红外热图像内眼角定位算法。该算法首先对采集的图像进行面部倾斜校正，接着采用Gentle-AdaBoost与HAAR特征相结合进行人脸、人眼粗定位，并引入几何校正对眼睛区域精确定位，最终依据内眼角区域特性提出区域精化与区域生长分割相结合的内眼角定位。在3种不同的红外热图像数据集以及自主采集不同季节的温度区间室外的数据集上进行实验。结果表明:在室外环境下，所提出的方法可有效地定位内眼角，人眼定位准确率达到98.1%，内眼角定位准确率达到97.7%。

Inner-Canthus Localization in Infrared Thermal Images in Outdoor Environments

Target areas in infrared thermal images in outdoor environments are affected by background overheating and the surrounding environment, causing fuzzy target boundaries and large noise. An inner-canthus location algorithm for infrared thermal images in outdoor environments is proposed to solve this problem. First, the algorithm corrects the facial tilt of collected images. Then, Gentle-Adaboost and Haar features are combined to perform approximate localization of human faces and eyes, and geometric correction is applied to accurately locate the eye region. Finally, based on the characteristics of the inner-canthus region, an inner-canthus location is proposed by combining region refinement and region growth segmentation. Experiments are conducted on three different infrared thermal image datasets and outdoor datasets independently collected at different temperature ranges in different seasons. The results show that the proposed method can effectively locate the inner canthus in the outdoor environment, and the accuracy for human eyes and inner-canthus can reach 98.1% and 97.7%, respectively.