Clothes classification based on deep belief network

In this paper, faced with the diversity and difference of clothes, we propose a novel method combined deep belief network and softmax classifier to achieve the classification of clothes. First of all, we preprocess the angle and scaling of various styles of clothes images that are collected through web crawler, then we do something to convert the image into the corresponding input format. In addition, the DBN is trained layer-by-layer with the input of all image pixels and the top level of the network is a classification hyperplane. At last, we apply the DBN that has been trained completely to classify the strange clothes. Compared with traditional classifier softmax and deep structure deepid1, our algorithm has higher classification accuracy rate.