FISC: Furniture image style classification model based on Gram transformation

Abstract

With the development of e-commerce, the types of commodities are becoming more diversified. Classification of commodities based on aesthetic attributes such as style is an important supplement to traditional classification techniques. Aiming at the problems of an unclear definition of furniture image style features, difficulty in extraction, and poor classification effect of general models, we design a furniture image classification model FISC based on Gram transformation. The FISC model is based on convolutional neural network technology, which extracts high-level content features of the image and performs Gram transformation as style features and inputs to the classifier for classification and recognition. At present, there are few public image style data sets. In this study, we build a data set of furniture image style attribute tags for the objectivity and pertinence of the experiment. The model has been fully experimentally compared, and the accuracy of the final training set and test set are 99.23% and 94% respectively, which fully verifies the superior performance of the FISC model on the task of furniture image style classification.