Instructor: Andrea Loddo, Luca Zedda, Davide Antonio Mura
CFU: 3
Prerequisite:
Image Processing and Analysis
Computer Vision (optional)
Deep Learning (optional)
Objectives:
Nowadays, computer vision-based systems are everywhere and adopted for many purposes, from medicine to security and quality control.
Concerning healthcare systems, they collect and provide most medical data in digital form. The availability of medical data enables many artificial intelligence applications. There is a growing interest in the quantitative analysis of clinical images acquired by positron emission tomography, computed tomography, magnetic resonance imaging, or microscopy techniques.
Computer vision applied to video surveillance deals with the real-time acquisition, processing, and management of video from cameras installed in public and private places for multiple purposes, however, related to the concept of security. It can involve the public sphere, fighting crime and preventing offenses, and the private sphere to ensure personal safety and tranquility. The large category of video surveillance systems includes all the equipment installed in urban contexts (squares, streets, and monuments), in crowded places (stations, subways, airports, stadiums), in banks, in private homes and industrial sites, as well as systems for monitoring traffic (vehicular, air, naval) and those for environmental control (fires, floods, landslides).
Many possible applications of computer vision in agriculture and botany exist, such as detecting pathologies for the preventive identification of diseases on crop plants, optimizing automated irrigation, and recognizing and classifying contamination on seeds, leaves, or plants.
Regardless of the task, however, the increasing amount of available data may lead to a more significant effort to make methods based on computer vision and artificial intelligence useful in real-world applications.
Moreover, this task is even more challenging due to the high variability of data, the availability of various imaging techniques, and the need to consider data from multiple sensors and sources.
After introducing the concept and challenges of image processing, computer vision, and artificial intelligence, the course will address the following critical topics at state of the art.
In particular, it aims to provide an overview of recent advances in the field and analyze how these techniques can be employed in the typical image processing workflow, from image acquisition to classification, including retrieval, detection, and classification on a topic of interest of the student, even with the most advanced.
Topics:
1. Medical image analysis for the following types of diagnosis:
a. Neurodegenerative diseases from CT, MRI, PET (Alzheimer's, Parkinson's, multiple sclerosis);
b. COVID19 from X-ray, CT, or multiple biomarkers;
c. Malaria from peripheral blood samples;
d. Red blood cell diseases or abnormalities from peripheral blood samples;
e. Leukemia or other white blood cell abnormalities from peripheral blood samples;
f. Tumors or other pathologies from histological samples;
g. Macular degeneration, glaucoma, diabetic retinopathy from retinal fundus or optic disc specimens
2. Botanical image analysis:
a. Cataloging plants from leaf images;
b. Cataloging plants from seed images;
c. Detection and classification of plant diseases from leaf images;
d. Detection and classification of plant diseases from fruit images;
e. Fruit quality estimation;
f. Classification of the type of fruit.
3. Analysis of video surveillance images and videos:
a. Counting people in a crowd;
b. Identification of people;
c. Identification of individuals;
d. Re-identification of individuals;
e. Face recognition;
f. Detection of abnormal behavior.
4. Fake news detection:
a. News articles classification as real or fake with both text and images;
b. Deepfake detection.
5. Image retrieval:
a. Biomedical imaging to retrieve similar sections or samples from a biomarker;
b. Biomedical imaging to support the diagnosis.
6. Open issues and research challenges.
Teaching methodology:
The reading course is divided into two parts. In the first part, the instructor will give a seminar, introducing the fundamental topics of the course and the work directions. In the second part, the students will improve their knowledge of the course topics through independent study of scientific papers and periodic meetings with the instructor.
References for each topic (selection):
On medical image analysis topic:
Andrea Loddo, Sara Buttau, Cecilia Di Ruberto, Deep learning-based pipelines for Alzheimer's disease diagnosis: A comparative study and a novel deep-ensemble method, Computers in Biology and Medicine, 2021, 105032
Cecilia Di Ruberto, Andrea Loddo, Lorenzo Putzu, Detection of red and white blood cells from microscopic blood images using a region proposal approach. Comput. Biol. Medicine 116: 103530 (2020)
Andrea Loddo, Cecilia Di Ruberto, Michel Kocher, Recent Advances of Malaria Parasites Detection Systems Based on Mathematical Morphology. Sensors 18(2): 513 (2018)
Cecilia Di Ruberto, Andrea Loddo, Lorenzo Putzu, Histological Image Analysis by Invariant Descriptors. ICIAP (1) 2017: 345-356, 2016
Cecilia Di Ruberto, Andrea Loddo, Lorenzo Putzu, A leucocytes count system from blood smear images - Segmentation and counting of white blood cells based on learning by sampling. Mach. Vis. Appl. 27(8): 1151-1160 (2016)
On botanical image analysis topic:
Andrea Loddo, Mauro Loddo, Cecilia Di Ruberto, A novel deep learning based approach for seed image classification and retrieval. Comput. Electron. Agric. 187: 106269 (2021)
Andrea Loddo, Cecilia Di Ruberto, On the Efficacy of Handcrafted and Deep Features for Seed Image Classification. J. Imaging 7(9): 171 (2021)
On analysis of video surveillance images and videos topic:
Emanuele Ledda, Lorenzo Putzu, Rita Delussu, Andrea Loddo, Giorgio Fumera, How Realistic Should Synthetic Images Be for Training Crowd Counting Models? CAIP (2) 2021: 46-56
Delussu, R., Putzu, L. & Fumera, G. Synthetic Data for Video Surveillance Applications of Computer Vision: A Review. Int J Comput Vis (2024)
On fake news detection:
Luca Zedda, Alessandra Perniciano, Andrea Loddo, Cecilia Di Ruberto, Manuela Sanguinetti, Maurizio Atzori: Snarci at SemEval-2024 Task 4: Themis Model for Binary Classification of Memes. SemEval@NAACL 2024: 853-858
On image retrieval:
Lorenzo Putzu, Andrea Loddo, Cecilia Di Ruberto, Invariant Moments, Textural and Deep Features for Diagnostic MR and CT Image Retrieval. CAIP (1) 2021: 287-297
Assignments:
The exam is composed of two parts: an oral test and a project development test. The first one concerns a topic chosen by the student and agreed with the teacher among those covered in the course. At the end of the first part, the student will present, in a 30 minutes seminar, the contents of the course and the chosen topic, and he/she will answer any questions in a Q&A of about 10 minutes. The student will also have to design and implement a simple prototype of an application task based on computer vision concerning the chosen topic. The project can be carried out individually or by a team of students.
Further information:
"To request additional information or schedule an online/in-person appointment, email: andrea.loddo@unica.it (please include luca.zedda@unica.it and davideantonio.mura@unica.it in CC)"
Informatica
