1	Selected topics in Computer Vision CS3243 Foundations of Artificial Intelligence (Textbook section 24.2, 24.3, 24.5, 24.6) Slides due to Huang Weihua
2	Outline Image formation Low-level Vision Smoothing Edge detection High-level Vision Object recognition Brightness-based approach Feature-based approach Application of Computer Vision Manipulation Navigation
3	Image Formation In computer vision, an image is a two-dimensional grid of pixels.
4	Image Formation Perspective projection: a process of projecting an object in a scene on an image plane.
5	Image Formation Visible light comes in a range of wavelengths: 400 nm (the violet end) to 700 nm (the red end). Discrete representation: Black and white (1 bit) Grayscale: 0-255 brightness (1 byte) RGB combination: each from 0-255 (3 bytes)
6	Image Formation
7	Low-level Vision Smoothing: removing extreme values from the image. Gaussian filter: replacing the original pixel I(x₀, y₀) by summation of I(x, y)G_σ(d) over all pixels, where: , in one dimension. , in two dimensions. And d is the distance between pixel (x, y) and (x₀, y₀) Convolution: h = f * g (Weighted sum)
8	Low-level Vision
9	Low-level Vision Edge detection: finding lines and curves in the image plane that have significant change in brightness. Theorem: (f * g)’ = f * g’ Canny edge detection: combining the Gaussian smoothing process and edge detection process.
10	Low-level Vision
11	Low-level Vision After edge detection, we can segment the edges into visual groups that are single objects or parts of an object. Segmentation is based on similarities of certain visual properties, such as: Brightness Color Texture Gradient
12	High-level Vision Object recognition Applications: Biometric identification, content-based image retrieval, handwriting recognition, etc. Approaches: brightness-based recognition and feature-based recognition.
13	High-level Vision Brightness-based recognition Basic feature: the brightness of pixels. Statistical approach to detect certain objects, such as faces and cars. Disadvantage: great redundancy inherent in the representation.
14	High-level Vision
15	High-level Vision Feature-based recognition Basic feature: regions and edges Classification approach: finding configuration of edges corresponding to views of object. Deformable matching: using simple coordinate transformations. Shape context: arrangement of shapes.
16	High-level Vision
17	Application of Computer Vision Manipulation Direct processing on the objects. Example: manipulating engineering drawings. Navigation Moving without colliding with obstacles. Example: navigation system for an auto-driving vehicle.
18	Application of Computer Vision
19	For more information You can learn more from the following modules: CS3241 Computer Graphics. CS4243 Computer Vision and Pattern Recognition.