Latent fingerprint quality assessment plays an essential role in forensic analysis, directly impacting the reliability of biometric evidence. However, challenges such as low contrast, noise, and complex backgrounds often hinder effective evaluation. This study introduces a two-stage pipeline that automates quality scoring of latent fingerprints using deep learning. In the first stage, a robust preprocessing framework is applied—comprising Signal-to-Noise Ratio (SNR) filtering, image resizing, contrast enhancement via CLAHE, and saliency-based segmentation—to enhance fingerprint clarity. The refined images are then passed through the NFIQ 2.1 tool to generate quality scores, effectively transforming unlabeled fingerprint data into labeled training samples. In the second stage, a convolutional neural network (CNN) is trained in a supervised regression setting using these auto-labeled pairs. Experimental evaluation on a forensic-grade dataset demonstrates that the trained model achieves improvements in MAE, MSE, RMSE, and R \(^{2}\) metrics. The proposed approach offers a scalable and reliable solution for automated fingerprint quality assessment, reducing dependence on manual labeling while aligning with forensic standards.