distributed... image processing gets consistent results fast the csc 410/510 project by lenny...

14
Distributed . . . Image Processing Gets Consistent Results Fast The CSC 410/510 Project by Lenny Scardino Glenn Bond Wayne Arena

Upload: elvin-reed

Post on 31-Dec-2015

215 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Distributed... Image Processing Gets Consistent Results Fast The CSC 410/510 Project by Lenny Scardino Glenn Bond Wayne Arena

Distributed . . .

Image Processing Gets Consistent Results Fast

The CSC 410/510 Project by Lenny Scardino Glenn Bond Wayne Arena

Page 2: Distributed... Image Processing Gets Consistent Results Fast The CSC 410/510 Project by Lenny Scardino Glenn Bond Wayne Arena

Why choose image processing?

Data VisualizationFrom microscopic to cosmic, better images mean

Improved precision/ accuracy of measurement

Reproducible results (consistency)

Greater throughput means more analysis, leadingto greater likelihood that results will be accurate

Page 3: Distributed... Image Processing Gets Consistent Results Fast The CSC 410/510 Project by Lenny Scardino Glenn Bond Wayne Arena

For Example,How many shades of gray can you discern in this image?

Page 4: Distributed... Image Processing Gets Consistent Results Fast The CSC 410/510 Project by Lenny Scardino Glenn Bond Wayne Arena

The human eye. . .

Eyes are good contrast adjusters, but not good at discerning subtle color variations. When colors are close, details are lost.

Eyes can distinguish on the average about 30 shades of grayin a field of pixels. Pixels are discreet and measurable.

Eyes are not good judges of distance, and cannot accurately reproduce measurements. Pixels are evenly spaced at predefined distances.

Page 5: Distributed... Image Processing Gets Consistent Results Fast The CSC 410/510 Project by Lenny Scardino Glenn Bond Wayne Arena

A good image has:

Nice dynamic range

Good Contrast

Low Noise

Few artifacts

Adequate, uniform lighting

Is this a good image?

Page 6: Distributed... Image Processing Gets Consistent Results Fast The CSC 410/510 Project by Lenny Scardino Glenn Bond Wayne Arena

With a little contrast adjustment

details can be discerned more easily.

Page 7: Distributed... Image Processing Gets Consistent Results Fast The CSC 410/510 Project by Lenny Scardino Glenn Bond Wayne Arena

Some assistance with dynamic rangeis helpful for discerning detail. . .

Page 8: Distributed... Image Processing Gets Consistent Results Fast The CSC 410/510 Project by Lenny Scardino Glenn Bond Wayne Arena

Why distributed processing of Images?

Reduces processing time, further boosting throughput.

Allows more ambitious analysis projects.

Perhaps it will open up:

Use of existing, previously unusable data

New classes of problems now solvable

Page 9: Distributed... Image Processing Gets Consistent Results Fast The CSC 410/510 Project by Lenny Scardino Glenn Bond Wayne Arena

Why use the bitmap format?

They’re everywhere and easy to fabricate

Their structure is similar to the RGB format used by SGI

Each pixel is 3 ASCII characters: one for red, one for green, one for blue.

The pixels are one long line in the file, readable by the vi editor and other similar applications. Our C program reads in one long string and begins processing.

Pixels can be converted to integer triplets for easy manipulation, then returned to character type.

Page 10: Distributed... Image Processing Gets Consistent Results Fast The CSC 410/510 Project by Lenny Scardino Glenn Bond Wayne Arena

How our system processes the image

1. A bitmap file is read and loaded into an array

2. The array of pixel values is analyzed on the first pass

3. A new dynamic range is established by substitution

4. The modified array is used to create a new,

hopefully

better image in a new file

Hopefully this occurs in less wall clock time than

a sequential system would take.

Page 11: Distributed... Image Processing Gets Consistent Results Fast The CSC 410/510 Project by Lenny Scardino Glenn Bond Wayne Arena

Process 0 then sends lines of integer values to each node for analysis. As each node sends back information, it is tasked with a new line until all lines are analyzed.

Process0

Page 12: Distributed... Image Processing Gets Consistent Results Fast The CSC 410/510 Project by Lenny Scardino Glenn Bond Wayne Arena

Process 0 builds a dynamic range

array, used to substitute values

in the pixel array. A new bitmap is

constructed, line by line, with greater

contrast and dynamic range.

Page 13: Distributed... Image Processing Gets Consistent Results Fast The CSC 410/510 Project by Lenny Scardino Glenn Bond Wayne Arena

Process 0 accomplishes the second pass like thefirst. This time pixel values are rewritten.As each node sends back conformation, it is tasked with a new line until all lines are rewritten.

Process0

Page 14: Distributed... Image Processing Gets Consistent Results Fast The CSC 410/510 Project by Lenny Scardino Glenn Bond Wayne Arena

The new file is written and theprocess terminates.