Interest Points guided Convolution Neural Network for Furniture Styles Classification

Furniture style describes the distinctive appearance characteristics of furniture with more details. It plays an important role in realistic interior decoration. However, at present, image search technology often uses text for retrieval, which requires a lot of manual labeling, and often presents the problem of text difficulty. In this paper, a convolution neural network model based on interest points (CNN-IP) is proposed to realize automatic furniture style classification. Firstly, the interest points of the furniture images are acquired by combining the features of color, brightness and contour. Moreover, interest factors can be obtained by the combination calculation. Secondly, the training images are trained by CNN to construct the feature maps. Finally, we combine the interest factors and the feature maps to form a salient feature map for furniture style image classification. Compared with the current popular image classification methods, the experimental results show the superiority and comprehensiveness of CNN-IP.