Electronic Theses and Dissertation

Citra terdegradasi dapat menurunkan kemampuan sistem deteksi wajah dalam mengenali area wajah secara tepat. hal ini dapat terjadi karena citra terdegradasi banyak yang menghasilkan resolusi rendah, blur, serta pencahayaan yang tidak konsisten. penelitian ini menganalisis dampak penerapan normalisasi fotometrik terhadap performa deteksi wajah menggunakan haar cascade classifier pada citra terdegradasi. pengujian dilakukan dengan membandingkan hasil deteksi pada citra terdegradasi sebagai baseline dan citra setelah penerapan normalisasi fotometrik. hasil penelitian menunjukkan bahwa citra terdegradasi menghasilkan tingkat deteksi sebesar 41,79%. setelah penerapan normalisasi, homomorphic normalization menghasilkan persentase deteksi tertinggi sebesar 86,50%, diikuti gamma normalization sebesar 84,07%, dct (discrete cosine transform) normalization sebesar 77,40%, highpass normalization sebesar 75,61%, dan non-local means normalization sebesar 73,01%. berdasarkan evaluasi kinerja, homomorphic normalization menghasilkan performa deteksi terbaik dengan accuracy sebesar 83,25%, precision sebesar 96,24%, recall sebesar 86,05%, dan f1-score sebesar 90,90%. sementara itu, nlm (non-local means) normalization menghasilkan kualitas citra terbaik dengan mse sebesar 0,000008, psnr sebesar 51,27db, ssim sebesar 0,9972, dan entropy sebesar 7,5036. akan tetapi nlm normalization tidak menghasilkan tingkat deteksi tertinggi, hasil ini menunjukkan bahwa perbaikan kualitas citra tidak selalu berbanding lurus dengan performa deteksi wajah.

Degraded images can reduce the ability of face detection system to recognize facial areas accurately. This can occur because many degraded images produce low resolution, blur, and inconsistent lighting. This study analyzes the impact of applying photometric normalization on face detection performance using the Haar Cascade Classifier on degraded images. Testing was carried out by comparing the detection results on degraded images as a baseline and images after applying photometric normalization. The results showed that the degraded images produced a detection rate of 41.79%. After applying photometric normalization, Homomorphic Normalization technique produced the highest detection percentage of 86.50%, followed by Gamma Normalization at 84.07%, DCT (Discrete Cosine Transform) Normalization at 77.40%, Highpass Normalization at 75.61%, and Non-Local Means Normalization at 73.01%. Based on the performance evaluation, Homomorphic Normalization produces the best detection performance with Accuracy of 83.25%, Precision of 96.24%, Recall of 86.05%, and F1-Score of 90.90%. Meanwhile, NLM (Non-Local Means) Normalization produces the best image quality with MSE of 0.000008, PSNR of 51.27dB, SSIM of 0.9972, and Entropy of 7.5036. However, NLM Normalization does not produce the highest detection rate, these results indicate that improvements in image quality do not always correlate directly with face detection performance.