<![CDATA[The face is an important object in the biometric identification system. However, low image quality due to uneven lighting, noise, and variations in facial expressions can interfere with the accuracy of the recognition system. The study investigated the use of Convolutional Artificial Neural Network (CNN) combined with hybrid screening techniques to improve image quality, thereby improving the accuracy of facial recognition systems. Filters used include weight mean filtering, median filtering, Contrast-Limited Adaptive Histogram Equalization and gaussian filtering, wavelet filtering. The pre-processed image was then trained using image denoising measurements of the Structural Similarity Index, Mean Squared Error, and Peak Signal to Noise Ratio. The main objective of this study is to evaluate the best filtration combination to produce high accuracy in face classification. The datasets used were 55 classes and 100 images per class. The inceptionV3 architecture model is used for classifications with a number of epochs of 10. Evaluation was carried out on a facial data set with an 80%:20% scheme. The results of the experiment showed that the hybrid method produced the best performance with 94.5% validation accuracy, 94.2% precision, and 94.6% recall, an increase of +1.4% compared to baseline. The (original) baseline itself recorded 93.1% validation accuracy, 92.8% precision, and 93.2% recall. In addition, the loss graph shows that the pre-process model has faster and more stable convergence than the non-pre-processing model. These results confirm that the application of preprocessing, especially the hybrid approach, is able to improve the accuracy and stability of the model in image classification tasks.]]>
Copyrights © 2025
