ECU 3040 Digital Image Processing

Transcription

1 ECU 3040 Digital Image Processing Dr. Praveen Sankaran Department of ECE NIT Calicut January 8, 2015

2 Ground Rules Grading Policy: Projects 20 Exam 1 15 Exam 2 15 Exam 3 50 Letter Grading:Absolute Textbook: Gonzalez and Woods, Digital Image Processing 3rd Ed., Prentice Hall, 2007.

3 Outcomes 1 Ability to apply the knowledge of imaging systems to implement real world systems. 2 Design image enhancement algorithms and implement systems that utilize your algorithms. 3 Ability to work with and develop open source resources to solve image processing problems. 4 Ability to test and verify (analyze) the soudness of various algorithms.

4 Topics Imaging systems, quantization. Histograms and histogram based modications, spatial ltering, nonlinear spatial image enhancement. Frequency domain, homomorphic ltering, Retinex. Morphological image processing, segmentation. Denosizing, haze, blur removal. HDR imaging, tone mapping. Image quality assesment. Imaging for security. Patterns, classes, decision theory, networks. OpenCV - applications, live projects - end sem.

5 Plagiarism policy Homework assignments and design projects are to be the work of an individual student only. Evidence of foul play, if detected will result in appropriate action against all concerned. Students may discuss among themselves, but the nal work need to be their own.

6 Outline 1 Image and image creation 2

7 Outline 1 Image and image creation 2

8 Image Two-dimensional (2-D) (discrete) representation of a (continuous) physical three-dimensional (3-D) scene.

9 Image Formation By the wavelength (or frequency) of the emitted or reected radiation. Examples: Visible, Infrared, TeraHertz. By the modality with which the image is acquired: Passive: Active visible passive infrared acoustic or ultrasound X-ray TeraHertz.

10 EM Spectrum Figure : The Electromagnetic (EM) Spectrumreproduced from the Lawrence Berkeley Labs website.

11 Visible Spectrum Figure : Narrow visible spectrum Energy E = hν h - Planck's constant, ν - frequncy

12 Visible Spectrum nanometers - wavelength (Some special people able to go from nanometers) terrahertz Maximum sensitivity of eye nanometers - green region.

13 Image Formation - further divides By the image capture device. Examples: CCD or CMOS (visible) uncooled mi- crobolometer (Infrared) ultrasound transducer By the coordinate system of the displayed image: rectangular Cartesian coordinate system for most modalities. ultrasound and radar which are both polar.

14 Outline 1 Image and image creation 2

15 Case of a visible image photo-detector array. continuous amplitude, continuous extent radiance. to continuous amplitude but discrete point dened. Figure : Single imaging sensor Amplitude dependence Amplitude of the signal strength of the radiance eld at that point.

16 Approach to Imaging Spatial sampling: denes the continuous radiance eld only at discrete locations; Brightness quantization: converts the continuous amplitude to a discrete set of values. Achromatic image: shades of gray from black to white f (x, y) = image brightness at spatial location x, y. (real-valued, non-negative, and bounded ).

17 Approach to Imaging Spatial sampling: denes the continuous radiance eld only at discrete locations; Brightness quantization: converts the continuous amplitude to a discrete set of values. Achromatic image: shades of gray from black to white f (x, y) = image brightness at spatial location x, y. (real-valued, non-negative, and bounded ).

18 Approach to Imaging Spatial sampling: denes the continuous radiance eld only at discrete locations; Brightness quantization: converts the continuous amplitude to a discrete set of values. Achromatic image: shades of gray from black to white f (x, y) = image brightness at spatial location x, y. (real-valued, non-negative, and bounded ).

19 Figure : Image brightness function Figure : Rectangular sampling grid

20 Point Spread Function Spatial sampling associates with each pixel [m, n] an average brightness f [m, n] that is determined primarily by the brightness of the points within the pixel. The actual brightness contribution to f [m, n] from points within the pixel and from neighboring points outside the pixel is determined by a point spread or weighting function h. f [m, n] = f (x, y)h (m, n; x, y)dxdy x y

21 Brightness Quantization Analog Digital process. Associate a non-negative integer value, l, with each of the real valued values of f [m, n]. Gray level, l l = 0,1 L 1, where L = 2 b

22 Uniform Brightness Quantization This is the most popular. Decision levels Q l+1 Q l = β α L Values of f [m, n] less than α or greater than β are clipped to 0 or L 1respectively. Total number of bits required to represent the image: b M N.

23 A pixel is a small image area indexed by [m, n]; g [m, n] is the associated pixel value; the possible values of g [m, n] are the gray levels l = 0,1 L 1; a digital image is an M N array of gray levels.

24 Acknowledgement The material taught in this class is heavily inuenced by work of Dr. Zia Rahman. Zia-ur Rahman joined the Electrical and Computer Engineering Department at Old Dominion University, as an Associate Professor in Before that he was a Research Associate Professor with the Department of Applied Science at the College of William & Mary. He received a B.A. in Physics from Ripon College in 1984, and an M.S. and a Ph.D. in Electrical Engineering from the University of Virginia in 1986 and 1989, respectively. His graduate research focused on using neural networks and image processing

25 Acknowledgement techniques for motion detection and target tracking.