MSc Thesis Defense: Ayşe Ceren Şahin, A SYSTEMATIC APPROACH TO HANDLE MISSING PIXEL DATA IN DATA ANALYTICS PROCESSES, Date & Time: 22 July, 2026 – 2:00 PM, Place: FENS L065

A SYSTEMATIC APPROACH TO HANDLE MISSING PIXEL DATA IN DATA ANALYTICS PROCESSES

Ayşe Ceren Şahin

Data Science, MSc Thesis, 2026

Thesis Jury

Prof. Kemal Kılıç (Thesis Advisor)

 Prof. Erchan Aptoula

  Asst. Prof. Faran Ahmed

Date & Time: 22nd of July 2026 – 2:00 PM

Place: FENS L065

Keywords :  Image Restoration, Denoising, Zero-Shot Segmentation, Image Inpainting, Adaptive BM3D

Abstract

Data preprocessing is a fundamental prerequisite for effective image analysis. Missing pixel data usually caused by acquisition errors, transmission noise, or storage corruption leads to a significant loss of structural information and visual quality. In real-life applications, the underlying degradation process is rarely known a priori. To address this challenge, this paper presents a comprehensive evaluation of various restoration techniques across five distinct missing pixel degradation patterns.We propose a framework that analyzes and enhances traditional restoration methods by integrating the Frangi vesselness filter alongside zero-shot segmentation models, the Segment Anything Model (SAM) and ClipSEG. Experimental results on benchmark datasets (CBSD68, Dunhuang Murals, McMaster, and Kodak24) demonstrate that our boosted approaches outperform traditional inpainting and BM3D, yielding an improvement of 2 to 5 dB in average Peak Signal-to-Noise Ratio (PSNR). Finally, we introduce an end-to-end framework featuring a ResNet-based classifier that automatically detects the corruption type and routes the input image to the optimal restoration pipeline.