This preview shows page 1. Sign up to view the full content.
Unformatted text preview: n be
described by an affine function of the form d ( x r , y r ) F (ax r by r ) / a / 2 c b. Suppose we divide the right image into square blocks, and assume the 3D points corresponding to each
block can be approximated as a flat surface so that the disparity function for each block can be
modeled using an affine function (as shown in Part a). Propose an algorithm to estimate the affine
parameters. You just need to formulate the problem as an optimization problem and describe in words
how you may solve the optimization problem. Note that you can try this subproblem even if you were
not able to tackle Part a. Solution: d xl x r FB
Z On the other hand, using the relations given, we have
2
Solvign for Z from above yields 1 2 Or 1 1 2
Or
,
2
(b) Here you should solve for a,b,c, for each block by minimizing the matching error defined as ,
, ,
, , , , The optimization problem can be solved, e.g., using a gradient descent method.
5. (20pt) Be brief in your answers to the following questions.
a. (6 pt) Why do different encoder products (hardware or software) following the same video coding
standard have different coding performance? What parts of an encoder does the standard specify? (List
one) What components of an encoder design are not subject to the standard and can differentiate
among different products? (list 2 components at least)
b. (8 pt) What are some of the benefits of scalable coding for video streaming applications? (List one at
least). What are the three types of scalability? Propose one way to realize temporal scalability with
base layer corresponding to a lowframe rate video, and base plus enhancement layer corresponding to
a highframerate video. In such a coder, the base layer can be predicted either from the previous frame
reconstructed from the baselayer, or the previous frame reconstructed using both base and
enhancement layer. What are the pros and cons of each approach?
c. (6 pt) What causes error propagation in the decoded video when a...
View
Full
Document
 Spring '14
 YAOWANG

Click to edit the document details