The goal is to provide conceptual tools and algorithms for the elaboration of sophisticated methods for extracting information from images and for the development of techniques for computer vision applications.
Knowledge and understanding: The student must demonstrate knowledge and understanding of the fundamentals of digital image management and the implementation of efficient algorithmic solutions for the classification, identification and semantic segmentation of objects.
Ability to apply knowledge and understanding: The student must demonstrate that he can use his acquired knowledge to solve real problems on real images. Furthermore, he must demonstrate that he possesses adequate skills to formulate and argue original solutions to real case studies.
Independent judgment: The student must be able to independently evaluate the results of the scientific literature and commercial solutions proposed in the field of digital image processing.
Communication skills: The student must be able to argue with scientific rigor and appropriate terminology the approaches and algorithms known in the sector. They must be able to interpret the key aspects of the solutions proposed in the literature and be able to briefly illustrate how they work.
Learning skills: The student must be able to independently update and deepen specific topics and applications in the field of digital image processing in various and diversified study contexts.
Knowledge of Machine and Deep Learning concepts and techniques is required.
Knowledge of the fundamental operations in the field of image processing is required.
A basic knowledge of linear algebra is also required, with specific reference to vectors and matrices and the algebraic operations used for their manipulation. These concepts are considered already possessed by the student and no lesson of the course will be dedicated to these concepts.
Introduction to Computer Vision
-Filters, Edges detection
-Points of Interest and Region Descriptors: Scale-Invariant Feature Transform and Histograms of Gradients
- Deep Learning for Computer Vision: Convolutional Neural Networks and main architectures of CNNs
- Recognition: identification and recognition of objects; Semantic segmentation, Instance segmentation, and panoptic segmentation; Pose estimation
- Neural Style Transfer
- Motion field and Optical flow: Optical flow constraint equation; Lucas-Kanade method
The course provides an introduction to "high level" computer vision activities mainly based on modern deep learning techniques. After reviewing some concepts of low and medium level vision activities, essential components for more advanced level operations, some techniques for the extraction of manual characteristics based on key-points are reviewed. Subsequently, the main Convolutional Network architectures for image classification, object detection and semantic segmentation are introduced.
Advanced applications such as neural style transfer are introduced. Finally, for the evaluation of the motion of objects from static or video images the concept of motion field is introduced and the optical flow is studied.
The teaching is structured in 48 hours of frontal teaching, organized in 2-hour lessons marked by the academic calendar. Frontal teaching consists of theoretical lessons held by the teacher on the topics covered by the course. Theoretical lessons aim to convey to the student the knowledge of algorithms for advanced computer vision tasks as well as the technical and scientific notions useful for understanding and implementing them. During the course, students are offered scientific articles which, by choice, will be discussed in the form of seminars collectively in the classroom, aimed at verifying the understanding of the topics covered. The conduct of seminars on the topics proposed by the teacher allows you to verify the practical application of the topics seen at a theoretical level and the ability of students to propose alternative solutions. Constant attendance of the course is recommended, which is however optional. The exams will be the same for all students, attending or not.
R. Szeleski, Computer Vision: Algorithms and Applications, 2nd ed. (2022)
R.C. Gonzalez, R.E. Woods, Digital Image Processing, 4th Eds, Pearson
The exam consists of two parts: (a) in carrying out a seminar, at the end of the course, on a scientific article whose theme is proposed by the teacher; (b) in an oral test on all the topics addressed during the lessons.
The test is passed if an overall score of 18 is achieved.
The learning material is provided, in English, through the e-learning platform at the following weblink http://e-scienzeetecnologie.uniparthenope.it. To access the published material, the student must be regularly registered at the platform and subscribed at the course.
The teacher is always available at receiving the students during the lessons for clarification or doubts about the topics of the course. Conversely, an appointment between the student and the teacher must be agreed and scheduled by email.