Find in Library

In response to the high cost and low efficiency in traditional interior space design, this research applies image feature extraction algorithms to interior space design to improve design efficiency. This method first extracts image features from two-dimensional home images to construct an image feature library for home items. Principal component analysis is combined to reduce the dimensionality of vectors in the library. Then a matching recommendation model is constructed. Next, binary encoding is applied to the feature layout of layout scene. Hot unique encoding technology is used to map it into vector representations and abstract vector extraction using word embedding algorithms. Finally, a layout network model is constructed using a neural network model to achieve indoor space layout. According to the results, the proposed recommendation model improved the average recommendation precision by 66.7% compared with traditional recommendation algorithms. When the principal component analysis algorithm reduced the vector dimension to 128 dimensions, the precision of the designed model improved by 17.1% compared with the Collaborative Filtering Recommendation (CF). When the number of similar home items was 4, the model training effect was better, and the recommended precision of the model was 61.3%. The learning rate determined the convergence effect of the layout model. The best convergence effect was achieved when the learning rate was 0.1. Therefore, the proposed method can quickly and accurately complete intelligent design work, which has practical significance for the field of computer vision.