s4 - 16 ECE 253a Digital Image Processing Pamela Cosman...

Info iconThis preview shows pages 1–3. Sign up to view the full content.

View Full Document Right Arrow Icon
# 16 ECE 253a Digital Image Processing Pamela Cosman 11/11/10 SOLUTIONS FOR HOMEWORK 4 1. Scalar Quantization: Comparing pdfs and distortions of binary quantizers: You are given an input signal X to quantize. It has a pdf given by f X ( x ) = A - x, 0 x 1 A + x, - 1 x 0 0 , otherwise A 2-level quantizer for this input signal has output levels y 1 = - 1 2 and y 2 = 1 2 . (a) To Fnd the constant A: i 1 - 1 f X ( x ) dx = i 0 - 1 ( A + x ) dx + i 1 0 ( A - x ) dx = 1 p Ax + 1 2 x 2 Pv v v v 0 - 1 + p Ax - 1 2 x 2 Pv v v v 1 0 = 2 A - 1 = 1 and A = 1 (b) By symmetry, the distribution of the error should be uniformly distributed from - 1 2 to 1 2 . Mathematically, we start from Fnding the CD± of the error: Pr [ ε α ] = Pr b - 1 < X < 0 , ( - 1 2 - X ) < α B + Pr b 1 > X > 0 , ( 1 2 - X ) < α B = i 0 - 1 2 - α (1 + x ) dx + i 1 1 2 - α (1 - x ) dx = α + 1 2 for - 1 2 < α < 1 2 , and the PD± is: f ε ( α ) = d Pr[ ε < α ] = 1 for - 1 2 < α < 1 2 . (c) The distortion is: D = E ( ε 2 X ) = i - 1 2 - 1 2 x 2 × 1 dx = 1 12 (d) If the quantization levels are - 1 3 and 1 3 , the distortion is: D = E ( ε 2 X ) = i 0 - 1 p x + 1 3 P 2 (1 + x ) dx + i 1 0 p x - 1 3 P 2 (1 - x ) dx = 1 18 which is smaller than the distortion in part (c). This is because the probability density of X is concentrated near 0, and so it is better if we design the absolute value of our quantization level smaller than 1 2 . 1
Background image of page 1

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

View Full DocumentRight Arrow Icon
(e) Since the distortion is defned as the square oF the quantizer error, the sign oF the error wouldn’t matter, and the distortion will be the same. To fnd the new PD± oF the new quantizer error: Pr [ ε α ] = Pr b - 1 < Z < 0 , ( - 1 2 - Z ) < α B + Pr b 1 > Z > 0 , ( 1 2 - Z ) < α B = i 0 - 1 2 - α (1 + z ) dz + i 1 1 2 - α zdz = 3 4 + α - α 2
Background image of page 2
Image of page 3
This is the end of the preview. Sign up to access the rest of the document.

This note was uploaded on 01/15/2012 for the course ECE ece253A taught by Professor Cosman during the Fall '11 term at UCSD.

Page1 / 5

s4 - 16 ECE 253a Digital Image Processing Pamela Cosman...

This preview shows document pages 1 - 3. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online