Optimal design of artistic brand patterns based on visual perception and multi-model decision making

The art brand pattern plays a pivotal role in shaping brand recognition and attractiveness. Evaluating and refining these patterns is essential to improve their overall quality. This study investigates the optimization of art brand patterns through the perspective of visual perception theory and multi-model decision-making methods. First, an image quality assessment model was developed based on the Feature Similarity Index Method (FSIM), which integrates phase consistency and contrast features to evaluate image quality. This was further extended into a color image evaluation approach, FSIM-C, by incorporating color space transformation. In addition, local spatial and frequency-domain information was extracted using a two-dimensional Log-Gabor filter to optimize the patterns. Several techniques were applied, including the Sobel operator for edge feature extraction, the Hue–Saturation–Value (HSV) model for color feature representation, and the spatial correlation index (SCI) method for texture feature analysis. The latter involved decomposing the image into sub-blocks and applying the Discrete Cosine Transform (DCT) to each block. To construct a comprehensive evaluation framework, phase consistency, edge, color, and texture features were fused into an 11-dimensional feature vector. A regression strategy was then established using a Generalized Regression Neural Network (GRNN), aligning the mapping process with the characteristics of the human visual system. Experimental validation was conducted on two benchmark datasets, the Laboratory for Image and Video Engineering (LIVE) dataset and the Categorical Subjective Image Quality (CSIQ) dataset. Three regression models were compared—GRNN, Support Vector Machine (SVM), and Random Forest (RF)—with GRNN achieving the best performance, particularly on the LIVE dataset. The proposed multi-model decision-making scheme achieved Spearman Rank Order Correlation Coefficient (SROCC), Kendall Rank Order Correlation Coefficient (KROCC), Pearson Linear Correlation Coefficient (PLCC), and Root Mean Square Error (RMSE) values of 0.9236, 0.9167, 0.7634, and 0.5372, respectively. Compared with classical approaches such as Peak Signal-to-Noise Ratio (PSNR), FSIM, and Visual Information Fidelity (VIF), the proposed framework demonstrated significant improvements in image quality assessment.