Description

A Software System and Deep Learning Model for Image Quality Analysis

INTRODUCTION
This joint research and development project between H.I.T. and Samsung aims to deliver an integrated software solution powered by a deep learning model for automatic image quality assessment (IQA) in smartphone camera images. The project combines cutting-edge AI research with robust software engineering to enable the automated and objective comparison of Samsung’s image processing pipelines. Deliverables include a trained deep learning model, a detailed evaluation report, and a fully functional software library integrated into Samsung's image evaluation pipeline.
BACKGROUND
Sophisticated computational photography techniques increasingly determine the quality of smartphone images. Samsung has developed various image enhancement algorithms targeting different use cases and hardware configurations. However, assessing their relative performance objectively remains a challenge. Existing approaches often rely on time-consuming manual reviews or traditional metrics that poorly align with perceptual quality. This project leverages recent advances in deep learning to develop a no-reference IQA system, tightly integrated into a user-friendly software tool, which will be co-developed and validated in collaboration with Samsung.
PROJECT SCOPE
1.	Develop and evaluate a deep learning model trained using Samsung-provided labeled data
2.	Evaluate the model on test datasets and analyse application-dependent KPIs
3.	Implement a software module that integrates with Samsung’s existing evaluation pipeline
4.	Integrate and support the deployment of the tool in the Samsung environment
STUDENT REQUIREMENTS
1.	Proficiency in Python programming
2.	Hands-on experience with deep learning frameworks, especially PyTorch
3.	Understanding of deep learning concepts applied to image data (e.g., CNNs, Transformers)
4.	Suitable for a team of 2 students
DEVELOPMENT TOOLS
1.	Programming Language: Python 3.x
2.	Deep Learning Frameworks: PyTorch (primary), TensorFlow/Keras (optional)
3.	Image Processing Libraries: OpenCV, PIL, scikit-image
4.	Development Environment: JupyterLab, VSCode, PyCharm
DELIVERABLES
1.	Trained Deep Learning Model: A robust, no-reference image quality assessment model capable of evaluating image quality across various devices and processing pipelines.
2.	Evaluation Report: A detailed document covering model architecture, training procedures, validation metrics (PLCC, SRCC), and comparative results against baseline methods.
3.	Production-Ready Software Library: A modular and scalable software pipeline with a clean API for integration with Samsung's image evaluation tools.
4.	Integration Interface: A set of well-documented modules, including data ingestion, quality scoring, and result exporting, designed to plug into Samsung’s internal evaluation pipeline.
5.	Documentation and User Guide: Full technical documentation, installation instructions, and usage examples for both the model and the software system.
6.	Support for Integration and Deployment: Assistance in configuring and deploying the system within Samsung’s infrastructure, including feedback loops for refinement.

___________________________________________________________
A Proposal for a Joint H.I.T. - Samsung Israel Project
Alexander (Sasha) Apartsin, http://apartsin.faculty.ac.il/
alexanderap@hit.ac.il, 054-4663885
Image Quality Analysis via Deep Learning – Samsung

AI and Machine Learning

Description

Emphasis in project execution

The project is has cooperation with the industry and combines meeting deadlines while being creative and focused on the task
