M.Sc. Thesis Opportunities in Image Quality Assessment and Analysis of Imaging Sensors
We are excited to announce M.Sc. thesis opportunities in Image Quality Assessment and the Analysis of Imaging Sensors within the 3D Media Group. These projects provide graduate students with the chance to engage in cutting-edge research, gain hands-on experience with state-of-the-art imaging technologies, and contribute to advancements in image quality evaluation.
- Spectral Response Analysis of Imaging Sensors
- Subjective Image Quality Assessment
- High Dynamic Range (HDR) Texture Measurement
- Image Quality Evaluation using Large Multi-modality Models
- High Dynamic Range (HDR) Noise Measurement
These thesis projects are ideal for students looking to deepen their expertise in imaging science and expand their professional horizons. Read on for details about the topics, eligibility, and application process.
Project title: Spectral Response Analysis of Imaging Sensors
Description:
Spectral response measurements of imaging sensors help researchers and engineers to quantify how a sensor responds to different wavelengths of light. This information can be used, for example, for calibrating the sensor for accurate color reproduction and fine-tuning the sensor design for better light sensitivity.
The Centre for Immersive Visual Technologies (CIVIT) has set up a spectral response measurement system that supports the spectral response profiling of various imaging sensors. In this thesis project, we aim to extend the functionality of the system by studying how the measurement data can be used for analyzing sensor characteristics such as quantum efficiency and for designing sensor-specific white balancing algorithms.
Project tasks:
- Study the principle of operation of the sensor spectral response measurement system and the nature and meaning of output data.
- Study the methodologies for measuring sensor quantum efficiency and design a practical technique and new setup utilizing the system from task 1.
- Utilize the new setup and technique for analyzing a sensor and report the results.
- Design sensor-specific white balance and color correction algorithms using the system from task 1.
- Demonstrate the performance of the algorithms designed in task 4 on photos taken by a sensor profiled using the system from task 1.
Outcomes:
- Hands-on experience of profiling a sensor with the spectral response measurement system.
- Knowledge of how to analyze sensor spectral response data.
- Methodological understanding of measuring sensor quantum efficiency and performing sensor-specific white balance using spectral response data.
For whom: MSc students (targeting an MSc. thesis)
Prerequisites: Basics of image and signal processing and MATLAB/Python programing skills. Knowledge of optics and photonics is a plus.
Start date: As soon as possible.
End date: The work is intended for 6-8 months.
How to apply:
Send your CV and transcript of records to: Ville Pihlajamäki (ville.pihlajamaki@tuni.fi)
For more information about the position, please contact: Atanas Gotchev (atanas.gotchev@tuni.fi)
Project title: Subjective Image Quality Assessment
Description:
Image quality assessment is crucial in various fields, including sensor and display technology, image processing, and computer vision. Subjective evaluation, which relies on human perception, remains the gold standard for assessing image quality, complementing objective metrics. Recommendations such as the “ITU-R BT.500 Methodology for the subjective assessment of the quality of television images” provide guidance for structuring such evaluations.
In this thesis project, the goal is to pilot a subjective image quality assessment methodology based on the ITU-R BT.500 standard. The project will involve capturing a set of test images, designing and implementing a software tool for subjective evaluation, conducting a pilot study with participants, and analyzing the results.
Project tasks:
- Study the existing subjective image quality assessment methods and related test image datasets.
- Define a set of controlled image distortions (e.g., noise, blurring) and capture example images.
- Develop a software tool to present the images.
- Conduct a pilot study with a panel of participants who assess the shown images following a selected quality ranking methodology.
- Analyze the collected subjective rating results using statistical means.
Outcomes:
- Methodological understanding of subjective image quality assessment methods.
- Knowledge of how to prepare task– and methodology-specific image test datasets.
- Experience of implementing software for presenting images to participants and collecting their responses.
- Knowledge of how to statistically analyze subjective image rating data.
For whom: MSc students (targeting an MSc. thesis)
Prerequisites: Basics of image and signal processing. C, C#, MATLAB, Python or other software development skills. Knowledge of optics is a plus.
Start date: As soon as possible.
End date: The work is intended for 6-8 months.
How to apply:
Send your CV and transcript of records to: Ville Pihlajamäki (ville.pihlajamaki@tuni.fi)
For more information about the position, please contact: Atanas Gotchev (atanas.gotchev@tuni.fi)
Project title: High Dynamic Range (HDR) Texture Measurement
Description:
Detail preservation is one of the most important attributes of image quality for any kind of imaging system. In the context of mobile photography, the degradation of details is often quantified using the notion of “Texture Blur”, as introduced in the IEEE 1858 Standard “Camera Phone Image Quality” (CPIQ) and later expanded in the ISO/TS 19567-2:2019 standard “Texture analysis using stochastic pattern”.
The stochastic pattern in question is the celebrated “Dead Leaves”, a fractal pattern of known spectral properties that allows for an accurate determination of the Modulation Transfer Function (MTF) of the system. The MTF is then combined with a Contrast Sensitivity Function (CSF) to obtain a figure of merit of detail preservation known as “Texture Acutance”. The CSF is determined according to the reference viewing conditions in terms of viewing angle of the observer.
While the methodology as detailed above is rather generic, the standard itself is very specific in terms of color spaces in which the operations are performed. Furthermore, it relies on the notion of “well exposed” images and of exposure correction. For these and several other reasons, the current standard is ill-suited to accurately quantify detail preservation in the context of High Dynamic Range (HDR) images.
HDR refers to images represented with more than ten bits per channel per pixel encoded with a non-linear transfer function. Legacy images with fewer bits are usually referred to as Standard Dynamic Range (SDR). When using compatible displays, HDR allows for more vivid and realistic pictures than SDR, with higher contrasts (more than 1000:1) and a wider range of colors.
Several competing formats for storing HDR images are currently being supported by different manufacturers, while a standard ISO/WD 22028-5 “High dynamic range and wide colour gamut encoding for still images (HDR/WCG)” is currently under development.
Project tasks:
- Study the most common HDR formats used in consumer electronics in compliance with the current draft standard ISO/WD 22028-5 “High dynamic range and wide colour gamut encoding for still images (HDR/WCG)”.
- Develop a detail preservation measure defined with HDR images in mind which also works for legacy SDR formats.
- The new measure should support HDR formats as in ISO/WD 22028-5.
- For the HDR case, the measure should correlate with the user experience, quantified as opinion scores on a Just-Noticeable Difference (JND) scale.
- For the legacy SDR formats, the measure should generate results compatible with the ISO/TS 19567-2:2019 measurements.
Outcomes:
- Methodological understanding of how to quantify detail preservation in the context of mobile photography.
- Knowledge about current and emerging image quality assessment standards.
- Knowledge about capture, representation and manipulation of HDR images.
- Algorithmical understanding of and programming experience in extracting features from captured test images to quantify their quality.
References:
[1] Cao F., Guichard F., Hornung Hervé, Measuring texture sharpness of a digital camera, Proc. SPIE. 7250, Digital Photography V (January 18, 2009)
[2] Cao Frédéric, Guichard Frédéric, Hornung Hervé, Dead leaves model for measuring texture quality on a digital camera, Proc. SPIE. 7537, Digital Photography VI (January 17, 2010)
[3] DXOMARK Decodes: Understanding HDR imaging
For whom: MSc students (targeting an MSc. thesis)
Prerequisites: Basics of image and signal processing. Programming experience (C, C#, MATLAB, Python). Knowledge of optics is a plus.
Start date: In May or June.
End date: The work is intended for 6-8 months.
How to apply:
Send your CV and transcript of records to: Ville Pihlajamäki (ville.pihlajamaki@tuni.fi)
For more information about the position, please contact: Atanas Gotchev (atanas.gotchev@tuni.fi)
Project title: Image Quality Evaluation using Large Multi-modality Models (LMMs)
Description:
The quality of images produced by a camera is assessed by two complementary approaches: 1) objective measurements using test charts in controlled environments and 2) evaluation of image quality on captures of natural scenes.
The variation in content in natural scenes often requires a perceptual evaluation approach, where experts observe images under optimal viewing conditions and judge different quality attributes, abstracting from the semantic content. Depending on the methodologies used, these judgments are expressed in the form of Mean Opinion Score (MOS) or Just Objectionable Difference (JOD) scales.
This project aims at exploring machine learning models to predict MOSs and JODs. Such models have demonstrated high promise in cases of pre-defined scene content (e.g. portraits) [1], but have proved limited in the general case of no prior knowledge about the scene content [2], [3]. In recent years, several methods have investigated the vision-language correspondence embedded in rich visual language prior such as CLIP (Radford et al., 2021) to improve generalization ability in image quality assessment (IQA). Q-Align [4] explores the use of large multi-modality models (LMMs) to provide visual rating aligned with human opinions.
Project tasks:
- Study IQA methods exploring LMM.
- Design automatic IQA of some IQ attributes on natural images using LMM.
- Design new IQA method by leveraging the potential of LMM for the problem of generalization to new scenes with few prior knowledge.
Outcomes:
- Methodological understanding of IQA.
- Knowledge about LMM and their use for IQA.
- Methodological and programming experience in designing new IQA methods employing LMMs.
References:
[1] Nicolas Chahine, Sira Ferradans, Javier Vazquez-Corral, Jean Ponce, Generalized Portrait Quality Assessment Signal Processing Letters 2024
[2] Shaolin Su, Qingsen Yan, Yu Zhu, Blindly Assess Image Quality in the Wild Guided by A Self-Adaptive Hyper Network, CVPR 2020
[3] Junjie Ke, Qifei Wang, Yilin Wang, Peyman Milanfar, Feng Yang , MUSIQ: Multi-scale Image Quality Transformer, ICCV 2021
[4] Haoning Wu, Zicheng Zhang, Weixia Zhang , Q-Align: Teaching LMMs for Visual Scoring via Discrete Text-Defined Levels, CVPR 2024
For whom: MSc students (targeting an MSc. thesis)
Prerequisites: Knowledge of image and signal processing. Knowledge of optics is a plus.
Start date: In May or June.
End date: The work is intended for 6-8 months.
How to apply:
Send your CV and transcript of records to: Ville Pihlajamäki (ville.pihlajamaki@tuni.fi)
For more information about the position, please contact: Atanas Gotchev (atanas.gotchev@tuni.fi)
Project title: HDR Noise Measurement
Description:
Noise reduction is one of the most important attributes of image quality for any kind of imaging system. In the context of mobile photography, the degradation of details is often quantified using the notion of “Visual Noise”, as introduced in the IEEE 1858 Standard “Camera Phone Image Quality” (CPIQ) and later expanded in the ISO 15739:2023 standard “Noise Measurement”.
The visual noise measurement requires the determination of a noise curve, i.e., the expected noise variance versus the average signal level, for a range of lightness values. The curve is then used to estimate the variance level at a reference signal level. The resulting variance is then converted in a logarithmic scale to better correlate with human perception.
To account for the effects of the human visual system, prior to the calculation of the noise curve the image under test is pre-filtered with a Contrast Sensitivity Function (CSF). The CSF is determined according to the reference viewing conditions in terms of viewing angle of the observer.
The term HDR generally refers to images represented with more than ten bits per channel per pixel encoded with a non-linear transfer function. Legacy images with fewer bits are usually referred to as Standard Dynamic Range (SDR). When using compatible displays, HDR allows for more vivid and realistic pictures than SDR, with higher contrasts (more than 1000:1) and a wider range of colors.
Several competing formats for storing HDR images are currently being supported by different manufacturers, while a standard ISO/WD 22028-5 “High dynamic range and wide colour gamut encoding for still images (HDR/WCG)” is currently under development.
Project tasks:
- Study HDR image representations and formats.
- Design a new noise reduction measure defined with HDR images in mind that supports the most common HDR formats in compliance with the current draft standard and also supports the legacy SDR formats.
- For the HDR case, the measure should correlate with the user experience, quantified as opinion scores on a Just-Noticeable Difference (JND) scale.
- For the SDR case, the measure should be compatible with the ISO 15739:2023 measurement.
Outcomes:
- Methodological understanding of how to quantify noise appearance in the context of mobile photography.
- Knowledge about current and emerging image quality assessment standards.
- Knowledge about capture, representation and manipulation of HDR images.
- Algorithmical understanding of and programming experience in extracting features from captured test images to quantify noise reduction performance.
References:
[1] Bourbon T., Hillairet C.S., Pochon B., Guichard F., New visual noise measurement on a versatile laboratory setup in HDR conditions for smartphone camera testing, IS&T International Symposium on Electronic Imaging, 2022.
[2] DXOMARK Decodes: Understanding HDR imaging
For whom: MSc students (targeting an MSc. thesis)
Prerequisites: Knowledge of image and signal processing. Knowledge of optics is a plus.
Start date: In May or June.
End date: The work is intended for 6-8 months.
How to apply:
Send your CV and transcript of records to: Ville Pihlajamäki (ville.pihlajamaki@tuni.fi)
For more information about the position, please contact: Atanas Gotchev (atanas.gotchev@tuni.fi)
