服装廓形的识别与量化

Identification and quantification of apparel silhouette

提出一种量化与识别服装廓形的方法。通过选择合适的人脸区域作为特征区域,将人脸特征区域与头高相关联,通过Ada Boost算法检测图像中的人脸后,利用特征区域与头高的关系确定头高;根据人体比例关系,将着装人体分割为头、肩胸、胸腰、腰臀、大腿和小腿6大区段,讨论了区段高度随人体身头比变化的情况,给出了由人体身头比计算区段高度的公式;然后对各个区段提取宽度,利用区段宽度建立了A形、T形、H形、X形及O形轮廓的形态值和形态比公式,使服装廓形得以量化。结果表明该方法能够准确地识别服装廓形。

This paper proposed an approach to quantify and identify the apparel silhouettes. Firstly, a proper region in human face was selected as feature region to make relationship between this region and the head height, and after the human face detection conducted with AdaBoost method, the head height was determined in relation to face featured region. Then, according to the human body proportion the apparelled body was divided into six blocks including head, shoulder-chest, chest-waist, waist-hip, thign and crus, and the fact that the heights of blocks vary along with the body-head ratio was discussed and the formula for calculating block heights with body-head ratio advanced. Finally, the width of each block was extracted from the image, and the formulas for the shape values and percentages of A, T, H, X and O styles were established, which made the apparel silhouettes quantified. It turned out that this method can identify apparel silhouettes in an accurate way.