The Role of Deep Learning in Digital Image Analysis
Bachelor Thesis Defense · Fabjan Elezi · Universiteti Metropolitan Tirana · June 2026
Dual Vision Analyzer
Loading model…
Image Analysis Visualizations
How this works
Upload any image — the live classifier (MobileNet, a pretrained CNN) returns its top-5 ImageNet predictions in real time, while the Grayscale, HOG and Edge panels show the hand-engineered features a classical pipeline relies on.
My measured CIFAR-10 accuracy (Chapter 7) — same data & evaluation, both pipelines
HOG + SVM51.66%
CNN70.02%
▲ ~18-point accuracy gap
Classical HOG + SVM~51.66%on CIFAR-10
|Deep Learning CNN~70.02%on CIFAR-10
|My own measured results — Chapter 7: classical vs. deep learning image classification