{[ promptMessage ]}

Bookmark it

{[ promptMessage ]}

lecture02

# lecture02 - ELEC317 Lecture 2 Digital Image Processing...

This preview shows pages 1–7. Sign up to view the full content.

ELEC317 Digital Image Processing Lecture 2 Image Model, sampling and quantization IMAGE SAMPLING & QUANTIZATION --- Continuous image needs be sampled on discrete grid, and --- The value of the sample quantized using finite number of bits Î The result is the representation of an image by a matrix of numbers. Example: An image of size 256x256 is actually a 256x256 matrix. Each picture element of the image is called a PIXEL . Suppose that each pixel is 8 bits, so there are 256 levels. The pixel value can therefore be any integer between 0 and 255. Total amount of data: 256x256 = 65.5 k bytes! Image Scanning A continuous image is 2-D. The sampler can only sample/quantize one pixel at a time Æ Need to scan the image. A common method is to scan row by row. Normally, the pixel locations are on a rectangular grid. However, it is worth noting that some scanning scheme have pixels on other type of grid. This will be

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
discussed later. Image Display Normally, display of sampled image is achieved by D/A conversion. On CRT, after D/A conversion, an array of light spot is displayed on screen. The intensity of light spot is proportional to pixel value. If the spots are too small, image will look discontinuous: To achieve the effect of continuous image, interpolation is needed. Another display method is halftone display . This is used by Newspaper, Printers, and Fax machine. Here, each pixel is marked by a number of much smaller black or white dots. The number of dots is proportional to pixel value. The dots are placed pseudo-randomly:
0,1,2,3 Two Dimensional Sampling Theory 1. Review of 1-D sampling theory Consider a 1-D continuous time signal x(t) with Fourier transform X(f) X(f) = dt e t x f j Π 2 ) ( The signal is called band-limited if X(f)=1 for |f| > B To sample x(t), the following scheme can be used: Where p(t) = −∞ = k kT t ) ( δ A pixel magnified . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . X(f) B -B f p(t) x s (t) x(t)

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Recall from ELEC 211, −∞ = Π = k T kt j e T t p 2 1 ) ( ) ( 1 ) ( 2 t x e T t x k T kt j s −∞ = Π = ) ( 1 ) ( −∞ = = k s T k f X T f X Thus, Xs(f) consists of replicas of X(f). If 1/T >= 2B, i.e. sampling frequency is greater than 2 x Max frequency, then replicas will not overlap. And X(f) [And hence x(t)] can be recovered from Xs(t) by LPF (low pass filter). The minimum sampling rate is called the Nyquist rate. 2. Reconstruction in 1-D case To reconstruct x(t) from xs(t), under the assumption 1/T >= 2B, we see t T x(t) x s (t) X s (f) X(f) k=1 k=-1 k=0 -1/T 1/T -B B
) ( ) ( ) ( f X Tf rect T f X s × = )} ( { ) ( ) ( 1 1 Tf rect F t x T t x s × = df e Tf rect F T T tf j Π = 2 1 2 1 2 1 )} ( { ) ( sin 1 2 2 1 2 1 2 T t c T t j e T T tf j = Π = Π −∞ = = k s kT k kT x t x ) ( ) ( ) ( δ ) ( sin ) ( ) ( ) ( −∞ = = k T t c kT t kT x t x [ ) ( ) ( ) ( T t f T t t f = ] Î ) ( sin ) ( ) ( −∞ = = k T kT t c kT x t x [interpolation formula] 3. Sampling of 2-D signals Similar to 1-D case, a 2-D image is band-limited if ) , ( ) , ( 2 1 ξ F y x f f Rect(Tf) 1 -1/(2T) 1/(2T) T x(kT)

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
and 0 ) , ( 2 1 = ξ F for 0 1 x > or 0 2 y > Sampling is achieved by where ∑∑ −∞ = −∞ = = mn y
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}

### Page1 / 24

lecture02 - ELEC317 Lecture 2 Digital Image Processing...

This preview shows document pages 1 - 7. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online