基于人脸图像的BMI预测算法研究

BMI是目前国际上常用的衡量人体胖瘦程度以及是否健康的一个标准。通常情况下,BMI是由个体的身高和体重计算得到的。目前,国外的研究人员提出了基于人脸图像预测BMI的算法,通过构建面部特征与BMI之间的关联集合,利用SVR回归模型进行BMI预测工作。该算法在实验室实验环境下表现良好,但在日常生活应用环境下仍有较大的预测误差。为了提高BMI预测算法在日常生活应用环境下的预测精度,提出面部区域面积比(RAR)、嘴颌宽度比(MJWR)和颊宽高度比(CWHR)这三种新的面部特征用于补充改进BMI预测算法,同时使用神经网络拟合代替SVR回归进行BMI预测实验。实验结果表明,在日常生活应用环境下,改进的BMI预测算法使得预测结果更加精确,BMI预测的平均绝对误差(MAE)降低了0.7。

BMI PREDICTION ALGORITHM BASED ON FACE IMAGES

BMI is a standard commonly used in the world to measure the obesity and health of the human body. Usually,the BMI is calculated from the height and weight of the individual. Nowadays,researchers have proposed the algorithm of the BMI prediction based on face images. They built the collection set between facial features and BMI,and predicted the BMI by SVR regression model. The algorithm performs well in the laboratory test environment,but there is still a large error when it is used in daily application. For improving prediction accuracy of the algorithm in daily application,we proposed three new facial features added to the algorithm. They are the region-to-area ratio(RAR),the mouth-to-jaw-width ratio(MJWR) and the cheek width-to-height ratio(CWHR). In addition,a neural network fitting model was used in BMI prediction instead of SVR regression. The experimental results show that in the daily life application environment,the improved BMI prediction algorithm makes the prediction results more accurate,and the MAE of BMI prediction is reduced by 0.7.