Jpeg Analyzer

Ben Applegate CSS497Advisor: Dr. Munehiro Fukuda

Purpose of Research

Create a JPEG analysis program that will run on the UWB Parallel Computing lab's cluster. Potential uses: • Agriculture

Webcam images can be analyzed for fruit that is ready to be harvested. Workers can then save time by only going to areas that have fruit.

• ParkingWebcams can be placed in a parking garage, and their images can be used to identify areas with open spaces.

 • Web Search

Images can be found over the web that are similar to an input image. Images can be analyzed, then stored as keywords to be used in searches.

Agriculture

Sort images according to amount of red, as compared to a baseline photo

What is MASS

  Public static void main( String[ ] args ) {

MASS.init( args );

Places space = new Places( handle, “MySpace”, params, xSize, ySize);

Agents agents = new Agents( handle, “MyAgents”, params, space, population );

space.callAll( MySpace.func1, params );

space.exchangeAll( MySpace.func2, neighbors );

agents.exchangeAll( MyAgents.func3 );

agents.manageAll( );

MASS.finish( );

}

func2( )

func2( )

func2( )

func2( )

func1( ) func1( )func1( )func1( )

func1( ) … …

func3( )

func3( )

JPEG image format

Created to store the maximum amount of visual data in the least amount of space

Three strategies -

• Discrete Cosine Transform• Entropy Coding• Human Vision

Discrete Cosine Transfom

The DCT stores cosine waves reducing them to a set of coefficients

JPEG uses an 8x8 DCT

Stores each visual band as a wave

Image source: Wikipedia

Entropy Coding

DCT information is stored using huffman encoding

Values are stored in a zig zag pattern to minimize space

Higher frequencies tend to go to zero, so the zig zag can terminate quickly

Image source: Wikipedia

Human Vision

The human eye can better detect differences in light and dark than in color.

RGB is converted to YCbCr

Y is the luminance

Cb and Cr are the chrominance values, and can be sub-sampled

RGB

CrCbY

Image source: NASA

Chroma Subsampling

Averages Cb and Cr values to save space

Common sampling ratios are• 2 x 1 : Horizontal• 2 x 2 : Horizontal and vertical

Cheap cameras (i.e. webcams) mostly use 2x2

JPEG MCU

MCU : Minimum Coded Unit

This is the smallest amount of data that can be coded in a given jpeg.

Size depends on subsampling• 2x1 = 16px by 8 px• 2x2 = 16px by 16px

Image source: NASA

Strategy

• Decode MCUs from huffman values• Send individual MCUs to MASS to be decoded in parallel

using the Inverse DCT:           The IDCT (like the DCT) requires 64 calculations per pixel

• Return pixel arrays• Rebuild the pixel arrays into

the image

Image source: Wikipedia

Results

As expected, analysis time decreased logarithmically as processes were increased

Future

• Optimize IDCT• Extend functionality to other chroma subsampling ratios• Improve fault tolerance• Optimize MASS calls by observing network overhead

o Send multiple MCUs to each node• Analyze image data using Munsell color codes

 Web search and apple tree pictures