Scribd
Upload
34 views9 pages

Efficient Block Based Motion Estimation: Deepak Turaga Mohamed Alkanhal Tsuhan Chen

This document discusses various motion estimation techniques for video compression. It begins with an overview of block-based motion estimation and describes full-search as the optimal but computationally expensive method. It then explores sub-optimal search strategies like three-step search and four-step search that sacrifice quality for speed. The document also covers hierarchical search methods and techniques that use spatial correlation within frames to further reduce the search space, such as predictive spiral search and adaptive window sizing.

Uploaded by

Ankush Sharma
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
34 views9 pages

Efficient Block Based Motion Estimation: Deepak Turaga Mohamed Alkanhal Tsuhan Chen

This document discusses various motion estimation techniques for video compression. It begins with an overview of block-based motion estimation and describes full-search as the optimal but computationally expensive method. It then explores sub-optimal search strategies like three-step search and four-step search that sacrifice quality for speed. The document also covers hierarchical search methods and techniques that use spatial correlation within frames to further reduce the search space, such as predictive spiral search and adaptive window sizing.

Uploaded by

Ankush Sharma
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 9

Efficient Block Based Motion Estimation

Deepak Turaga Mohamed Alkanhal Tsuhan Chen

Overview of Talk
Motion Estimation Sub-Optimal Search Strategies Hierarchical Search Methods Spatial Correlation Based Searches

2/21/99

EE 796

Motion Estimation
Conditional replenishment to remove temporal redundancy Block Based Motion Estimation
Break Frame into blocks, find best match for block in previous frame. Send only motion vectors and residue

Block Based Motion Estimation Algorithms


Full Search : Optimal, but computationally expensive Sub-Optimal Search Strategies
Sacrifice quality of motion compensation for speed. Three Step Search, Four Step Search, Cross Search, Orthogonal Search, Hierarchical Search etc.

2/21/99

EE 796

Search Strategies - Type I


Pick a large step size around the starting point. At every stage reduce the step size and move in the direction of the best match.
Error Surface

Displacement

2/21/99

EE 796

Search Strategies - Type II


Move in direction of best match.
Error Surface

Reduce step size only on overshoot.

Displacement

2/21/99

EE 796

Search Strategies - Type III

Do a full search over region with best center.

Error Surface

Divide search space into many regions. Pick a center point for each region.

Displacement

2/21/99

EE 796

Best Matches and Motion Vectors

Previous Frame
2/21/99 EE 796

Current Frame

Sub-Optimal Search Strategies


One at a Time Search
Look horizontally first, then look vertically. Very fast, but not very good quality.

2-D Log Search


Use pattern. Reduce step + size at every stage. Move center to best match

Sample Convergence Path


2/21/99 EE 796

Sample Convergence Path

Sub-Optimal Search Strategies


Cross Search
Similar to 2-D Log Search. Use a cross pattern instead.

Orthogonal Search
Alternate between horizontal and vertical directions. Move center to best match.

Sample Convergence Path


2/21/99 EE 796

Sample Convergence Path

Sub-Optimal Search Strategies


Three Step Search(TSS)
Do a coarse to fine search. Move in the direction of best match. Good when best match far from center of search.

Four Step Search(FSS)


Start with a fine step. Move in the direction of best match. Good when best match close to center of search

Sample Convergence Path


2/21/99 EE 796

Sample Convergence Path

Sub-Optimal Search Strategies


Binary Search
Partition the search space into smaller regions and pick one region based on which of the initial blocks selected has the smallest MAD.

2/21/99

EE 796

Hierarchical Search Strategies


Do full Search Here Downsampled Image

Propagate motion vectors down Downsampled Image

Original Image

2/21/99

EE 796

Spatial Correlation Based Searches


Exploit spatial correlation within frames to narrow down the search. Blocks belonging to one object tend to move together. Use information from neighboring blocks to predict location of best match for current block. 2 1 C 3

Use 1,2 and 3 as predictors for block C.


2/21/99 EE 796

MAD Based Spiral Search


Predict motion vector and MAD threshold for current block using predictor blocks. Move center of search to predicted motion. Do a spiral search around new center till MAD smaller than predicted threshold. Have to update to avoid accumulation of error.

Predicted Search Center

2/21/99

EE 796

Adaptive Window Size Search


Use predictors to predict center of search and size of window around search center. Do a search over the intersection of original search window and new search window. Need to do regular update.

2/21/99

EE 796

Majority Voting Schemes


Choose between different algorithms adaptively. Combine their best features. Use spatial correlation information to choose algorithm. e.g. FSS better for small motion while TSS better for large motion. If predictors have large motion vectors pick TSS, else pick FSS.

2/21/99

EE 796

Spatial Correlation Based Searches


Reduce Search Space by moving search center to predicted best match. Do an efficient search over the reduced search space through choice of algorithm. Save in bits for motion vectors (MV for H.263 are differentially coded) Do well in terms of the speed-qualitybitrate tradeoff

2/21/99

EE 796

You might also like