Combining the YOLOv5 and Grabcut Algorithms for Fashion Color Analysis of Clothing

The analysis and prediction of apparel fashion colors are very important for the production and sales activities of the apparel industry. In the field of fashion color analysis of apparel images, existing image algorithms have problems such as poor segmentation effects in complex backgrounds and poo...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:2022 5th World Conference on Mechanical Engineering and Intelligent Manufacturing (WCMEIM) s. 1126 - 1129
Hlavní autoři: Liu, Fenghua, Liu, Zhaoqi, Liu, Weiguang, Zhao, Hongsheng
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 18.11.2022
Témata:
Approximation algorithms
Clustering algorithms
component
fashion color
GrabCut algorithm
Histograms
Image color analysis
Image segmentation
KMeans clustering algorithm
primary color extraction
Video surveillance
YOLOv5 algorithm
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:The analysis and prediction of apparel fashion colors are very important for the production and sales activities of the apparel industry. In the field of fashion color analysis of apparel images, existing image algorithms have problems such as poor segmentation effects in complex backgrounds and poor data real-time. In this paper, the YOLOv5 algorithm is applied to garment detection, the histogram equalization is used to enhance the image of garment pictures, the KMeans clustering algorithm is used to get the approximate foreground area of the garment, the GrabCut algorithm is used to segment the image with the processed pictures to get the final foreground color area of the garment, and then the KMeans clustering algorithm is used to get the main color of the garment. thus analyzing the pattern between colors. The study of fashionable colors of clothing in video surveillance scenes has higher real-time data volumes, larger data capacities, and faster analysis speeds than the current research methods.
DOI:10.1109/WCMEIM56910.2022.10021426