Clothing Identification based on Fused Key Points

In order to solve the problem of low accuracy of multi-category clothing recognition, an algorithm combining global and local features of clothing is proposed, which is aimed at solving the decrease of clothing recognition rate caused by changes of perspective and attitude and complexity of images. Based on the deep convolutional neural network, the algorithm locates the key points of clothing, excavates the key points image blocks, fuses the low-level visual information of the key points image blocks and the high-level semantic information of the whole image for clothing classification. This paper conducts experiments on the Deepfashion dataset, uses keras for data enhancement, and compares it with HOG+SVM image recognition algorithm and deep learning algorithms such as Alexnet and FashionNet. The experimental results show that the proposed algorithm can effectively improve the recognition rate of clothing.