Focused on the diversity of appearance and the complexity of configuration

laying

and occasion in human images

a coarse-to-fine method was proposed for effective human parsing.It can decompose a human image into semantic regions which consists of three phases.In the first two phases

two effective models were trained with Fast Region-based Convolutional Network(Fast R-CNN)to respectively detect human body and clothing items.In the third phase

parsing clothing items based on retrieving similar over-segmented images and morphing them into absolute image coordinates.Experiments are conducted on three public databases

and the experimental results show that proposed method has higher accuracy and promising performance.