introduction & motivation - unibas.ch · 2016-02-22 · introduction to signal and image...
TRANSCRIPT
Introduction &Motivation
Introduction toSignal and Image
Processing
Prof. Dr. Philippe Cattin
MIAC, University of Basel
February 23rd, 2016
February 23rd, 2016Introduction to Signal and Image Processing
1 of 17 22.02.2016 09:15
Contents
2
4
5
6
7
8
9
10
11
12
13
14
15
Ph. Cattin: Introduction & Motivation
Contents
Abstract
1 Motivation
Motivation
Application Scenarios
Challenges
Aim of this Lecture
Imaging Examples (1)
Imaging Examples (2)
Definitions/Typology
Typical Application Areas of Computer Vision
Principle of a Computer Vision System
This is, however, not always easy!
This is, however, not always easy! (2)
Computer Vision vs. Computer Graphics
February 23rd, 2016Introduction to Signal and Image Processing
2 of 17 22.02.2016 09:15
February 23rd, 2016Introduction to Signal and Image Processing
(2)
Ph. Cattin: Introduction & Motivation
Abstract
In the Introduction a brief overview of Computer Visionand its possible applications is given. Additionally someimportant definitions and explanations are given.
3 of 17 22.02.2016 09:15
Motivation
February 23rd, 2016Introduction to Signal and Image Processing
(4)Motivation
The human brain is unparallelled in 2D image
analysis and image understanding
Half our brain is devoted to processing and
interpretation of visual data
Limitations of the human visual system
Quantification
Reconstruction of 3D scenes from images
High dimensional image spaces
Practical applications
Quantitative measurements
Automatic analysis of images
4 of 17 22.02.2016 09:15
February 23rd, 2016Introduction to Signal and Image Processing
Motivation
(5)
Ph. Cattin: Introduction & Motivation
Application Scenarios
Inspection
Positioning, registration, metrology
Assembly, navigation, and visual
control
OCR, document processing,
multimedia retrieval
Scene reconstruction, visualisation,
and editing
Image compression
Image enhancement and
restoration
Pattern, object, and event
recognition
Human motion/gesture/face
recognition and interpretation
5 of 17 22.02.2016 09:15
February 23rd, 2016Introduction to Signal and Image Processing
Motivation
(6)
Ph. Cattin: Introduction & Motivation
Challenges
There are still many visual tasks humans can easily do,but that are beyond the reach of computer visionsystems:
Cue Integration
Today: Perfection of cues such as edges, motion,
depth, texture
Future: Integration of multiple cues
Dynamic 3D
Today: Excellent 3D capturing methods available
Future: But the world is 4D (3D+time)
Recognition of object categories
Today: Vision system can recognise individual
objects under varying circumstances
Future: General categorisation problem unsolved
(exceptions: face/OCR)
Shape and scene representations
How to model objects and scenes? For which objects does texture
suffice and where do we need more geometric detail (3D
information)?
Grouping and segmentation
Check-and-Egg Problem: First, segment the different objects,
then recognise the objects. However, how to segment the objects
without exact knowledge about them?
More self-learning and self-diagnosis
Most vision systems don't adapt and don't learn once in
operation. Future systems have to work in less structured
environments. Vision system will have to understand their own
limitations and report them back.
6 of 17 22.02.2016 09:15
February 23rd, 2016Introduction to Signal and Image Processing
7 of 17 22.02.2016 09:15
February 23rd, 2016Introduction to Signal and Image Processing
Motivation
(7)
Ph. Cattin: Introduction & Motivation
Aim of this Lecture
Signal processing background
Present a basic set of methods commonly used in CV
How are they used?
How can I combine them?
How can I determine their performance?
8 of 17 22.02.2016 09:15
February 23rd, 2016Introduction to Signal and Image Processing
Motivation
(8)
Ph. Cattin: Introduction & Motivation
Imaging Examples (1)
Human Visual System
Scene: Real 3D objects
with colour/brightness
Image: Two 2D images
with colour and
brightness
Satellites
Scene: Earth, 2D/3D,...
Image: 2D Images with
different frequency bands,
3D
Camera Images (mimicing
human vision)
Scene: Real 3D objects
with colour/brightness
Image: 2D Image with
brightness, colour/BW
9 of 17 22.02.2016 09:15
February 23rd, 2016Introduction to Signal and Image Processing
Motivation
(9)
Ph. Cattin: Introduction & Motivation
Imaging Examples (2)
Computed Tomography
Scene: Body, 3D,
Absorption of x-ray
Image: 3D matrix with
absorption coefficients
Magnet Resonance Imaging
Scene: Body, 3D, tissue
properties
Image: 3D matrix with
tissue properties
Depth Images
Scene: 3D Objects
Image: 2D Image with
depth information
10 of 17 22.02.2016 09:15
February 23rd, 2016Introduction to Signal and Image Processing
Motivation
(10)
Ph. Cattin: Introduction & Motivation
Definitions/Typology
Computer Vision
Image Processing
(image → image)
Image Analysis (image
→ abstract description)
Image Understanding
(image → object
recognition)
Computer Graphics
Image Synthesis
(abstract
description →
image)
11 of 17 22.02.2016 09:15
February 23rd, 2016Introduction to Signal and Image Processing
Motivation
(11)
Ph. Cattin: Introduction & Motivation
Typical ApplicationAreas of Computer Vision
Image Acquisition
Preprocessing: denoising, remove camera
distortions,...
Storing
Compression
Context based retrieval
Visualisation
Synthetic 3D views
Highlight/segment special regions
Image Analysis
Image Fusion (pixel positions represent the same
spatial region)
Combination of images showing different spectral
bands (Mutual Information)
Combination of images captured with different
imaging modalities (MR/CT)
12 of 17 22.02.2016 09:15
February 23rd, 2016Introduction to Signal and Image Processing
Motivation
(12)
Ph. Cattin: Introduction & Motivation
Principle of a ComputerVision System
Brightness/Colour ←→ Structure
The Aim
We want to extract geometric, topologic, and
semantic object information as well as relations
between objects
Measure structures, identify objects such as
streets, buildings, cars,...
Important Issues
Accuracy, robustness, automation, complexity,
flexibility,...
Means
Generic object properties
Homogeneous properties within the objects
Strong changes at borders between objects
A priori knowledge about the scene
Include experience and expert knowledge:
mathematically difficult
13 of 17 22.02.2016 09:15
February 23rd, 2016Introduction to Signal and Image Processing
Motivation
(13)
Ph. Cattin: Introduction & Motivation
This is, however, notalways easy!
M.C. Escher(1898-1972)
14 of 17 22.02.2016 09:15
February 23rd, 2016Introduction to Signal and Image Processing
Motivation
(14)
Ph. Cattin: Introduction & Motivation
This is, however, notalways easy! (2)
Fig 1.1: M.C. Escher (1898-1972)
15 of 17 22.02.2016 09:15
February 23rd, 2016Introduction to Signal and Image Processing
Motivation
(15)
Ph. Cattin: Introduction & Motivation
Computer Vision vs.Computer Graphics
Digital Image
ComputerVision
→ Building
floor plan
↓
ComputerGraphics
← Abstract
description
of buildings
16 of 17 22.02.2016 09:15
February 23rd, 2016Introduction to Signal and Image Processing
Fig 1.2: Flight through Pompeii (IT)
17 of 17 22.02.2016 09:15