Image-text matching method for clothing hierarchical feature representation and association learning

• Main Authors: PENG WEI, JIANG AIPING, LIU LIJUN, FU XIAODONG, LIU LI
• Format: Patent
• Language: Chinese; English
• Published: 14.04.2023
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention relates to an image-text matching method for clothing hierarchical feature representation and association learning, and belongs to the field of computer vision and artificial intelligence. The method comprises the following steps: firstly, constructing a hierarchical feature representation module comprising a CNN, a Faster-RCNN, a cascaded pyramid network and an RNN by combining a self-defined clothing structure, respectively extracting global, style and structural features of an input clothing image and description, subject and label features of an input clothing text, and performing hierarchical representation; then, in combination with quadruple comparison and correlation reasoning, correlation learning is carried out on three-layer features of global and description, style and subject, structure and label of the clothing image and text, and three relation vectors of global, style and structure label are obtained respectively; finally, similarity is calculated based on the three relation vect