A DEEP LEARNING MODEL-BASED FEATURE EXTRACTION METHOD FOR ARCHITECTURAL SPACE AND COLOR CONNECTION IN INTERIOR DESIGN

In architectural interior design, color is one of the important design elements. Through the reasonable combination of various color elements, it can effectively improve the interior environment and create an atmosphere that meets the preferences and needs of users. And with the continuous development of social economy, the application of color in interior design is becoming more and more widespread. Using different colors in interior design to harmonize not only can relieve people’s visual fatigue, but also can bring people a pleasant mood. Different colors have different meanings, therefore, the use of color in interior design should be more flexible and color matching should be more innovative. The warm and cold, near and far, expansion and contraction of color make the color space the most dynamic key element in design. The grasp of color and scale of architectural space and the flexible use of color will directly affect the quality of architectural space design. Color can strengthen the form of interior space or destroy its form. In order to accurately grasp the connection between architectural space and color in interior design, this paper proposes a deep learning model-based feature extraction method for the connection between architectural space and color in interior design. First, we construct a product color sentiment imagery dataset; then, we build a model for generating architectural interior space layout and color design schemes based on the product color sentiment imagery dataset and conditional deep convolutional generation adversarial network, and innovatively generate product color design schemes. This algorithm can better balance the chromaticity, saturation, and clarity of images. When determining the similarity of indoor space colors, depth features are superior to point-to-point pixel distance and aesthetic features of indoor space colors. Finally, the effectiveness and applicability of the proposed method are verified in relevant experiments. © 2024 SCPE.