基于相似度计算方法的人脸图分割

Face Image Segmentation Based on Similarity Calculation Method

结合监督和非监督学习模型实现了人脸图像的自动分割。图像分割中一般常用监督学习模型,这种学习模型受不同分类标志图像数据的限制,结合非监督学习模型,可以有效解决资源有限条件下的图像自动分割。该文首先用监督学习模型SVM(支持向量机)进行人脸图像分割,这种方法需要人工找出SVM中不同类别的支持向量,在实验中发现同一张图像进行多次实验分割效果不一致,存在较大的人为误差。为了实现支持向量的自动选取,该文用非监督学习模型K-means得到人脸图的不同类别作为SVM的支持向量,提出了相似度计算方法。并对输入的支持向量数据进行优化,提高了SVM的运算效率。实验结果表明,通过这种方法得到的SVM学习模型的分类结果和图像分割误差指标明显比单独使用非监督学习模型K-means方法好。

The supervised and unsupervised learning models are combined to realize automatic segmentation of face images. Supervised learning models are generally commonly used in image segmentation, which is limited by the image data of different classification signs. Combined with the unsupervised learning model, it can effectively solve the automatic image segmentation under limited resources. Firstly, we use the supervised learning model SVM(support vector machine) to segment the face image. This method requires manual identification of the support vectors of different categories in the SVM. In the experiment, it is found that the same image is divided into multiple experiments with inconsistent results and large human error. In order to realize the automatic selection of support vectors, we use the unsupervised learning model K-means to obtain different types of face images as the support vectors of SVM,and propose a similarity calculation method. The input support vector data is optimized to improve the operation efficiency of SVM. The experiment shows that the classification results and image segmentation error indicators of the SVM learning model obtained by this method are significantly better than using the unsupervised learning model K-means method alone.