为研究脸型与领型之间复杂的适配性视觉关系,使消费者能够根据自身脸型轮廓和比例匹配合适的领型,文章以男性脸型与男西服领型为研究对象,先对男性脸型与男西服领型进行特征提取及分类,共划分出12种脸型和20种领型.随后利用三维虚拟试衣技术建立出240个实验样本,通过主观问卷调研不同脸型与领型组合搭配的视觉效果适配度评价.最后,将脸型和领型的搭配作为模型输入值,主观视觉适配度评价作为模型输出值,并采用PSO-LSSVM算法建立脸型和领型适配度预测模型.结果表明,采用PSO-LSSVM算法的模型均方根误差为0.077 6,平均绝对误差为0.057 3;相对于PSO-RBF神经网络算法,均方根误差降低0.041 1,平均绝对误差降低0.037 6.该预测模型可为消费者线上选购、企业新品精准研发与营销、个性化定制提供一定的参考.
Personalization is on the rise in the consumer clothing market since it enables individuals to enhance their figures by selecting outfits that complement their body types.Various garment recommendation systems have been developed,and a considerable amount of research has been conducted on garment fit.In selecting clothing,the appropriate coordination between the face and collar is important,as it improves the overall appearance.However,the current recommendations of it are primarily based on subjective experience,which could not be quantified and accurately expressed.In view of this problem,developing a model that accurately predicts face/collar compatibility makes sense.The commonly used methods for prediction are neural networks and support vector machine(SVM).Nevertheless,the predictive results of many of the methods are unsatisfactory due to the limited sample size.LSSVM is one of the improved algorithms of the standard SVM,and has unique advantages in the identification and prediction of small samples and nonlinear data.Considering the significant influence of parameter combinations on prediction abilities,PSO is frequently applied to optimize the parameters instead of determining them empirically. To investigate the complex fitness visual relationship between face shape and collar shape,we first classified facial and suit collar shapes.Facial classification was based on temporal facial index,zygomatic width index,morphological facial index and jawline.And collar classification was based on the width and shape of lapels.So male facial shapes were finally divided into 12 different types,while suit lapel shapes were divided into 20 types.On this basis,240 experimental samples were created by using 3D virtual fitting technology for evaluating the visual fit of various face shapes and collar combinations in a questionnaire experiment.Then the PSO-LSSVM algorithm model was developed by analyzing the questionnaire experiment results.The combination of face and collar was taken as the input value of the model and the subjective visual evaluation as the output value.Additionally,the prediction model proposed was compared against PSO-RBF method to further verify the improvements achieved by this method. It is found that different face shapes fit different suit lapels:round faces are suitable for single-breasted men's suits with normal and narrow barge head;square faces are suitable for double-breasted men's suits with normal and wide barge head;long faces are suitable for men's suits with normal barge head;oval faces fit all collar shapes basically;and single-breasted,low-profile suits are unsuitable for almost any face shape.Based on the results of the subjective questionnaire fit score,the PSO-LSSVM prediction model and PSO-RBF neural network model for the compatibility of male face shapes and male suit collar shapes were established respectively.The result of the PSO-LSSVM algorithm and PSO-RBF shows that the root mean square error of the PSO-LSSVM is 0.077 6 and the average absolute error is 0.057 3,which reduces the root mean square error by 0.041 1 and the average absolute error by 0.037 6 compared to the latter.It indicates that the PSO-LSSVM model has a higher prediction accuracy. We use a non-contact measurement approach to classify male facial shapes into 12 categories,providing reference for machine learning based on facial classification.Meanwhile,the framework can be reference for consumers in the selection of suits.It also offers a benchmark for companies in the marketing of new products,as well as in the area of personalization.