| Along with the in-depth application of modern information technology in our lives and the development of the new "Internet +" format,the Internet and the traditional industry are deeply integrated,e-commerce is also rapidly active,people’s life and shopping methods have been profoundly changed,online clothing shopping become a major industry,showing a trend of popularization,globalization and mobility.Due to the huge application prospects and potential benefits of the clothing industry in online shopping,a large number of clothing images have emerged in the network,making the clothing classification method widely used in the field of “searching for pictures” and clothing matching and recommendation in e-commerce.It has promoted research in the field of clothing image.In order to solve the problem that the clothing images are mostly based on the coarse-grained classification of simple styles and the classification accuracy is low,taking the fashion women’s clothing as an example,the research on the clothing image classification method based on component detection and visual features is proposed to improve the clothing classification.The goal of accuracy and precision can better meet the needs of real-world applications.To resolve the problem of the detection accuracy is not accurate enough due to the large number of shooting scenes and human body postures,a clothing image part-based detection method is proposed based on DPM model.Firstly,the gradient direction histogram is calculated and normalized,truncated and dimensionally reduced to obtain the DPM feature.Secondly,the position and root model and the component model’s response score are calculated.Finally,the optimal position of the component is estimated by the response transformation,and the detection result of the target comprehensive response score is obtained,which is better adapted to the detection of the human body parts of different human postures and perspective changes and the search for differentiated component regions.To resolve the problem of the fashion women clothing image classification is not specific enough for the professional design knowledge in the field of fashion clothing,and the classification accuracy is not high enough,a coarse-grain classification method is proposed based on visual features and LSVM.Firstly,the part-based detection was conducted to detect the unclassified images combing the training dataset of fashion women clothing,Secondly,the bottom feature is extracted on the detected images,the coarse-grained styles and attribute tables of the fashion garments are established,Finally,the coarse-grained classification results were implemented and output.To resolve the problem of the fashion women’s styles and attributes are more detailed and complicated,and the differences in attributes between different types are small,a fine-grained classification method of clothing images is proposed based on style feature descriptors.Taking fashion women’s clothing as an example,Firstly,the style feature descriptors that describe the fine-grained attributes of the clothing are defined,Secondly,the style feature descriptors are matched with the extracted four underlying features to improve the validity and accuracy of feature extraction.Finally,The Random forest and multi-class SVM supervises and learns different fashion women’s styles and attributes,realizes fine-grained classification of fashionable women’s images,and has high classification accuracy. |
