Computa(onal Vision Introduc(on and Overview. Lecture 1: Introduc(on Hamid Dehghani Office: UG38

Transcription

1 Computa(onal Vision Introduc(on and Overview Lecture 1: Introduc(on Hamid Dehghani Office: UG38

2 Schedule 1 Lecture / week 9 am, Fridays@ Nuffield G13 1 Lab / week 11 am UG04, CS Modules webpages hpp://

3 Modules web pages

4 Schedule Lectures (L) and Highly Directed Studies (HDS) Preliminary Date% 17/01/2014' 24/01/2014' 31/01/2014' 07/02/2014' 14/02/2014' 21/02/2014' 28/02/2014' 07/03/2014' 14/03/2014' 21/03/2014' ' Topic% Introduction'to'Computational'Vision'and'Human'Vision'I' Human'Vision'II' Edge'Detection'and'Noise'Filtering' Hough'Transform' ROC'Analysis' Motion' Eigenvalues'and'PCA' Face'Recognition' Guest'Lecture' No'Lecture'

5 Schedule Labs Preliminary All lab work (Lab 1 Lab 4) to be wri7en and submi7ed by the following Friday Date% 17/01/2014' 24/01/2014' 31/01/2014' 07/02/2014' 14/02/2014' 21/02/2014' 28/02/2014' 07/03/2014' 14/03/2014' 15/03/2014' ' MATLAB'Tutorial' Lab'1:'Edge'Detection' Lab'2:'Noise'Filtering' Lab'3:'Hough'Transform' Projects' Projects' Projects' Lab'4:'Face'Recognition' Projects' NO'LABS' Topic%

6 Assessment Exam: 1.5 hours 70 % Date - TBA Assignment 30 % 3 page experimental write- up of a specified projects rela(ng to your labs (Individual Effort) Submission Date: Preliminary: Last day of 12.00

7 Advance No(ce APendance will be monitored Lectures and Labs

8 Module Descrip(on Learning Outcomes: Make informed choices about which sort of algorithms to apply to solve specific problems Use standard vision libraries or sodware to construct working vision systems Apply algorithms to simplified problems by hand Discuss the advantages and drawbacks of different methods, explaining their working

9 What is Computa(onal Vision? First consider Visual Percep(on to know what is where, by looking. vision is the process of discovering from images what is present in the world, and where it is. The acquisi(on of knowledge about objects and events in the environment through informa(on processing of light emiped or reflected from objects

10 What is Computa(onal Vision? To make computers See Automa(c inference of proper(es of the world from images Automa(c inference The world Image Proper(es Inference without (or minimal) human interven(on The real unconstrained 3D physical world Constrained/Engineered environments 2D projec(on of the electromagne(c signal provided by the world Geometric: shape, size, loca(on, distance, Material : color, texture, reflec(vity, transparency Temporal: direc(on of mo(on (in 3D), speed, events Illumina(on: light source specifica(on, light source color Symbolic: objects class, object s ID

11 What is Computa(onal Vision?

12 Is it easy? All people can see equally well Babies can see Really primi(ve animals can see We see effortlessly (at least it feels this way) Vision is immediate Vision appears to be flawless

13 Computa(onal Vision is challenging Vision needs to reverse the imaging process which is a many- to- one mapping ( recover lost informa(on). Vision needs to cope with an inherently imperfect imaging process ( recover lost informa(on) Vision needs to cope with discre(zed images of a prac(cally con(nuous world ( recover lost informa(on). The mere complexity of the task is enormous! Huge por(on of our brain is dedicated to visual percep(on.

14 Approaching the problem computa(onally Constrain/simplify the world Constrain/simplify the task (i.e., the desired output) Devise universal guiding assump(ons or heuris(cs Incorporate explicit knowledge Use experience (learning)

15 Applica(ons Automated naviga(on with obstacle avoidance Object/target detec(on and recogni(on Place/scene recogni(on Manufacturing and assembly Document processing Quality control Biomedical applica(ons Accessibility tools Human computer interfaces

16 Biological Vision Light and image forma(on Re(nal Processing Colour Visual Pathway

17 Electromagne(c Spectrum

18 Visible Light Humans perceive electromagne(c radia(on with wavelengths nm (1 nm = 10-9 m) f = c/λ f = frequency (Hz) λ = wavelength (m) c = speed of light (2.998x10 8 ms - 1 ) E = hf E = Energy (J) h = Plank s constant (6.623x10 34 Js)

19 Light Capturing Devices In the beginning: Forma(on of photopigments (>3BYA) Molecules in which light triggers a physical or chemical change. Captured photons lead to release of energy (of different forms) Released energy is used for Building food (photosynthesis) Behavioral reac(on (nerve reac(on)

20 Light Capturing Devices Photocells Light sensi(ve patch

21 Evolu(on of eyes Single cell 1D capture of light X Mul(ple cell BePer direc(on resolu(on

22 Evolu(on of eyes Mul(ple cell BePer direc(on resolu(on But...where is the image?

23 Evolu(on of eyes A pinhole camera Dilemma:

24 Evolu(on of eyes Solu(on: Use of light refrac(on and hence lenses

25 Refrac(on (Snell s Law) The wavelength changes, but wave crests can't be created or destroyed at the interface, so to make the waves match up, the light has to change direction.

26 Forma(on of lens Evolu(on of eyes

27 The Human Eye

28 Pinhole Camera Model

29 Pinhole Camera: Basic geometry

30 Pinhole Camera: Perspec(ve projec(on

31 Image Forma(on f = the focal length (in meters) 1/f = the power of the lens (dioptres) Human eye has power ~59 dioptres 1/f = 50 dioptres; f = 1/50 = 0.02 m

32 Image Forma(on Most of the refrac(ve power of the human eye comes from the air- cornea boundary(49 of 59 dioptres) As an object moves closer the power of the lens must increase to accommodate So if the object is infinitely far away f = = dioptres But if it is 1m away the lens must change shape to produce a sharp image dioptres f = =

33 Image Forma(on As an object moves in world how does it move across the image plane? If the image plane is curved then as θ gets larger this becomes a worse and worse approxima(on

34 Summary Module Outline Uses of Computa(onal Vision Image forma(on Very early visual processing Filling in and perceptual effects

35 Reading Vicki Bruce, Visual Percep(on, Chapters 1-3 Neil Carlson, Physiology of Behavior, Chapter 3, Vision