Construction of Graphic Creative Feature Extraction and Recognition System based on SIFT Algorithm

The increasing demand for creative graphic design necessitates robust systems for accurate feature extraction and recognition. Traditional methods often struggle with complex transformations, prompting the integration of modern algorithms. This study proposes a system combining the SIFT algorithm for feature extraction and CNN for classification. SIFT detects keypoints and generates descriptors via Gaussian pyramids and gradient histograms. CNN optimizes matching results through convolutional layers (3×3 kernels), max-pooling, and dense layers, enhancing global feature learning. Experiments demonstrated 80% matching accuracy, outperforming edge-detection-based methods. Robustness tests revealed 95% accuracy at 0°rotation and 92% at original scaling, though performance declined under extreme noise and blur. The SIFT-CNN integration achieves high accuracy and efficiency in diverse scenarios but faces challenges in extreme conditions. Future work should optimize computational complexity for real-time applications and enhance adaptability to complex environments. This system provides a reliable foundation for graphic recognition in practical settings.