VISION ANALYSIS AND DEEP LEARNING (VADL)
VADL is a Master's level subject designed for students with a background in Signal and Communications Theory. The course equips students with the knowledge to describe, analyze, and manipulate image and video data. It covers a broad range of theoretical concepts in image and video processing, including segmentation, motion analysis, filtering, and object tracking.
DESCRIPTION OF THE COURSE
This course focuses on the part of VADL corresponding to spatial and temporal segmentation techniques (Unit 4). Segmentation is the process of decomposing images and video sequences into meaningful components for further analysis.
Image segmentation is a fundamental task in computer vision with a wide range of applications, including medical image analysis, object recognition, and autonomous driving.
Regarding video segmentation, one of the main aims is to separate moving objects (foreground) from the background on the scene. This allows for a variety of applications, such as object tracking, action recognition, content-based video retrieval, and augmented reality.
The course is divided into two distinct parts.
Part 1: Image segmentation
This first part will address the fundamentals of image segmentation, placing special emphasis on two key methodologies:
- Histogram thresholding-based segmentation: This technique is based on the analysis of the image's histogram, which represents the distribution of pixel intensity levels. By applying appropriate thresholds, regions of the image belonging to different object classes can be separated.
- Gradient analysis-based segmentation: This technique focuses on edge and contour detection in the image. Edge detection operators, such as the Canny detector, allow identifying abrupt transitions between regions of different intensity, thus delimiting the objects present in the image.
Part 2: Moving object segmentation
The second part of the course will delve into the field of moving object segmentation. The main objective will be to isolate moving objects (foreground) from the background of the scene, thus enabling a deeper analysis of video sequences. Here, we will study background subtraction methods in depth. These methods compare each frame of the video with a reference background image, identifying pixels that have changed and therefore belong to moving objects.
FUTURE EXPANSION
While this course emphasizes segmentation, VADL encompasses a broader scope. In the future, the course will be expanded to include the remaining parts of VADL, such as machine learning and deep learning for image analysis.
