ONNS: Experimental Evaluation of Human Smile Recognition Based on Image Processing Assisted Optimized Neural Network Strategy

This paper presents the experimental evaluation of human smile recognition using an image processing-assisted optimized neural network strategy (ONNS). The primary objective was to assess different neural network architectures and feature extraction techniques to improve the accuracy and real-time performance of smile detection. Several deep learning architectures, including a standard CNN, optimized CNN, BKNet, RetinaNet, and MobileNet, were evaluated on a smile recognition dataset. Geometric and texture-based feature extraction techniques such as Local Binary Patterns (LBP) and Histogram of Oriented Gradients (HOG) were employed to improve the robustness of the model. The results showed that the optimized CNN achieved the highest accuracy of 95.3%, with a precision of 93.6% and a recall of 94.1%. Furthermore, a real-time performance evaluation demonstrated that MobileNet, while achieving the fastest inference time of 10 ms (100 fps), had a slightly lower accuracy of 90.5%. The combination of geometric and texture features led to improved performance, with the combined method achieving an accuracy of 92%. These results indicate that the proposed optimized neural network strategy is effective in improving smile recognition accuracy while maintaining real-time performance.