Current Issue Cover


摘 要
目的 少数民族服装色彩及样式种类繁多等因素导致少数民族服装图像识别率较低。以云南少数民族服装为例,提出一种结合人体检测和多任务学习的少数民族服装识别方法。方法 首先通过k-poselet对输入的待识别图像和少数民族服装图像集中的训练图像进行人体整体和局部检测以及关键点的预测;其次,根据检测结果,从待识别图像和训练图像中分别提取颜色直方图、HOG、LBP、SIFT以及边缘5种底层特征;然后,将自定义的少数民族服装语义属性与提取的底层特征进行匹配,采用多任务学习训练分类器模型,以学习少数民族服装的不同风格;最后实现少数民族服装图像的识别并输出识别结果。另外,由于目前缺少大型的少数民族服装数据集,本文构建了1个云南少数民族服装图像集。结果 在构建的云南少数民族服装图像集上验证了本文方法,识别精度达到82.5%至88.4%,并与单任务学习方法进行比较,本文方法识别率更高。结论 针对现有的少数民族服装识别率较低的问题,提出一种结合人体检测和多任务学习的少数民族服装识别方法,提高了少数民族服装图像识别的准确率和效率,又能较好地满足实际应用中的需求。
Human Detection and Multi-tasking Learning for Minority Clothing Recognition

Wu Shengmei,Liu Li,Fu Xiaodong,Liu Lijun,Huang Qingsong(Faculty of Information Engineering and Automation,Kunming University of Science and Technology)

Objective With the increasing number and diversity of minority clothing in the domains of multimedia, digital clothing, graphics and images, there is a growing need for automatically understanding and recognizing minority clothing images. However, most previous work used the low-level features directly for classification and recognition that lacked local feature analysis and semantic annotation of clothing. The diversity of clothing colors and styles results in low recognition accuracy of the minority clothing. Therefore, a minority clothing recognition method based on human detection and multi-task learning was proposed for Yunnan minority clothing. Method The main idea of this work is to propose the k-poselet detection method to detect minority clothing image and define the semantic attributes of minority clothing matching low-level features. Besides, multi-task learning method also applied to improve the accuracy of recognition of minority clothing images. Firstly, the k-poselet approach was employed to perform global and local human detection and key point predictions on the identifying and the training ones from the minority clothing dataset. Secondly, five kinds of low-level feature including color histogram, HOG, LBP, SIFT, and edge of the identifying image and the training ones were extracted, respectively. Then, the semantic attributes were defined to match the five low-level features, and a multi-task learning classifier model was trained to obtain the different styles of minority clothing. Finally, the recognition results of minority clothing were realized and output. Due to the lack of minority clothing dataset, we also constructed a minority clothing dataset of Yunnan, including 25 minority clothings in Yunnan, which collected mainly from online stores including Taobao, Tmall, Jingdong and other platforms, each ethnic group has 1000 maps with a total of 25,000 images. The size of each image was set to 500500 pixels, and different ethnic groups were classified and numbered. In order to facilitate the experiments, the background of image was appropriately processed, and the format of image was .jpg. Result The method of this paper is validated on the dataset of Yunnan minority clothing, the results show that the human detection method not only achieves greater precise recall rate, but also significantly outperforms DPM and the traditional poselet detection in the task of human prediction. At the same time, compared with the current detection method that using the features extracted by convolutional neural network, the experimental results are acceptable and demonstrate the effectiveness of the proposed approach. Moreover, the recognition accuracy of minority clothing images can reaches 82.5% to 88.4%. Compared with the single task learning method, the proposed method had a higher recognition rate. Conclusion Faced with a wide variety of colors and styles of minority clothing, the recognition rate of minority clothing is low, this paper proposed a minority clothing identification method based on human detection and multi-task learning, which can improve the accuracy and efficiency of minority clothing images recognition of better practical applications. The research results of this paper can be used for the digital analysis, understanding and identification of Chinese minority clothing, even provide an effective digital tool for recording, inheriting and protecting of national culture, even promote the development of tourism, economy and culture in ethnic areas. Although there are certain limitations of our method, it can provide a clear direction for future research. We only consider major attributes and limited set of training images for detection, feature extraction, and classification. The diversity of human posture and occlusion issue can affect the recognition accuracy of minority clothing images. Furthermore, our current work focuses on the dataset with 25 minorities of Yunnan. In future work, we would like to extend our method to more applications and address the limitations in the current method as described above. More investigations and analysis are required for extreme situations such as clothing accessories for minorities.