Search
Methods of intelligent image and video processing based on visual quality metrics for emerging applications
The goal of this project has been related to the design of methods and algorithms of intelligent image and video processing based on combined visual quality metrics for emerging applications, taking into account peculiarities of HVS. The development of fast no-reference metrics (without reference images) is a crucial element of many practical applications. The first result is the development of the combined HVS-based quality metrics, trained or optimized according to rank-order Spearman (SROCC) or Pearson (PCC) correlation with subjective quality scores, verified for several databases, including multiply distorted images, and tested for different practical situations for emerging applications. Additionally, conclusions concerning applicability of the designed metrics for considered applications have been provided. The next result is the development of prediction methods for multichannel image processing considering degree of inter-channel correlation, especially for filtering and lossy compression, as well as for some other applications, e.g., quality assessment of 3D printing surfaces, stitched images or remote sensing applications.
During mutual visits the extension and sharing of complementary experiences of both teams has taken place, towards development of new no-reference combined metrics which might be applied, e.g., in automatic quality control of products, including 3D printing, video surveillance, ADAS solutions, autonomous robots and UAVs, medical imaging and remote sensing, making it possible to use them finally in some industrial systems, leading to their noticeable improvements.
In particular, during the project the following methods and quality metrics have been developed: combined metrics based on mutual structural similarity for quality assessment of 3D printed surfaces, combined general-purpose full-reference metrics optimized using neural networks and Lasso approach, combined robust full-reference and no-reference metrics for remote sensing applications optimized using various approaches, combined full-reference metrics for multiply distorted images, methods of providing the desired image quality of the BPG compressed images, and intelligent lossy compression method controlled using various image quality metrics.
Details
- Project's acronym:
- NAWA ZUT-KhAI
- Financial Program Name:
- Programy NAWA
- Organization:
- Narodowa Agencja Wymiany Akademickiej
- Agreement:
- PPN/BUA/2019/1/00074/U/00001 z dnia 2020-02-12
- Realisation period:
- 2020-01-01 - 2021-12-31
- Research team leader:
- prof. dr hab. inż. Krzysztof Piotr Okarma
- Team members:
-
- contractor dr inż. Piotr Lech
- contractor dr inż. Jarosław Fastowicz
- contractor dr inż. Wojciech Chlewicki
- Realised in:
- Faculty of Electrical Engineering
- External institutions
participating in project: -
- National Aerospace University, "Kharkiv Aviation Institute" (KhAI), Charków (Ukraine)
- Project's value:
- 23 700.00 PLN
- Request type:
- Different
- Domestic:
- Domestic project
- Verified by:
- No verification
seen 9 times