Important: Use custom search function to get better results from our thousands of pages

Use " " for compulsory search eg:"electronics seminar" , use -" " for filter something eg: "electronics seminar" -"/tag/" (used for exclude results from tag pages)

Tags: Distributed, Image, Processing, using, hadoop, MapReduce, framework, mapreduce c, mapreduce combiner, mapreduce cassandra, mapreduce cuda, mapreduce code, mapreduce cloud, mapreduce couchdb, mapreduce criticism, mapreduce counters, mapreduce clustering, mapreduce database, mapreduce diagram, mapreduce drawbacks, mapreduce dean, mapreduce data locality, mapreduce disadvantages, mapreduce debug, mapreduce distributed computing for machine learning, mapreduce design pattern, mapreduce design, mapreduce examples,
Thread Rating:
  • 0 Votes - 0 Average
  • 1
  • 2
  • 3
  • 4
  • 5
Distributed Image Processing using hadoop MapReduce framework
Post: #1

.pdf  Distributed Image Processing using hadoop MapReduce framework.pdf (Size: 412.16 KB / Downloads: 1044)

Distributed Image Processing using Hadoop MapReduce framework

Presented By:
Binoy A Fernandez
Sameer Kumar
Distributed Image Processing

To demonstrate how the hadoop mapreduce framework can be extended to work with image data for distributed image processing.
Distributed Sobel Edge Detection algorithm
Easy to understand and implement, and at the same time, it is computationally expensive.
Has the property that the new value to be calculated for a pixel depends not only on that pixel's original value but also on the values of surrounding pixels.
Distributed Image Processing
Basic Idea
An image can be considered as made up of a two dimensional array of pixels. We can partition the image data into subsets and operate on the data in parallel by distributing subsets to different map tasks.
Convolution at a single pixel
1. Create kernel h indexed from 0 to m-1 horizontally and 0 to n-1 vertically and populate it with kernel coefficients
2. Compute kernel half width, m2 = floor(m/2)
Compute kernel half height, n2 = floor(n/2)
3. sum = 0
for k = -n2 to n2 loop
for j = -m2 to m2 loop
sum = sum + h(j + m2, k + n2) f(x - j, y - k)
end loop end loop
g(x, y) = sum
Distributed Image Processing
Edge Detection Algorithm
Convolution of an image ignoring the borders
1. Create kernel h indexed from 0 to m-1 horizontally and 0 to n-1 vertically and populate it with kernel coefficients
2. Compute kernel half width, m2 = floor(m/2)
Compute kernel half height, n2 = floor(n/2)
3. Create an M x N output image, g
4. for all pixel co-ordinates, x and y, loop
g(x, y) = 0
end loop
5. for y = n2 to N-n2-1 loop
for x = m2 to M-m2-1 loop
Compute g(x, y) using previous algorithm end loop end loop
Responsible for,
1. creating the input splits (getSplits)
2. dividing the splits into records (getRecordReader)
A chunk of the input that is processed by a single map.
An input split is divided into records, and the map processes each record (key-value pair) in turn. A RecordReader is used to generate record key-value pairs.
Base class for all implementations of InputFormat that use files as their data source. Provides,
methods to define which files are included as the input.
an implementation for generating splits for the input files.
lnputFormat<K,V> FilelnputFormat<K,V>
TextlnputFormat ImagelnputFormat
Key: LongWntable Key: Text
Value: Text Value: ImageWritable
Reuse the methods provided for defining which files are to be included as input to a job.
Define a filter to accept only certain image formats.
Define a custom implementation for generating splits for the input files.
Splitting Logic
In putSplit 2
Split Size = 9
[0,13) [0.27) [0,35)
Assumption: Size of an ImageSplit will not be greater than HDFS size.
In putSplit 3
In putSplit 4
Width = 20, Height = 35
Consider image of dimensions 20 x 35 pixels. Based on numSplits argument, the image is divided horizontally
into splits.
Performance: To compute the image splits, we need only the image dimensions. This information can be obtained from the image header thereby avoiding having to read the entire image data into memory.
A record reader implementation for reading sub images based on the computed splits, which is processed independently by the map tasks.
Key: input image path
Value: an ImageWritable
Options :
Read the entire image into memory as a BufferedImage and use the getSubImage method.
Read only the required portion of the image as defined by the split, into memory.
1. Compute the input splits
2. Schedule the map tasks
Submit Job
3- Obtain RecordReader for the split
Child JVM Map / Reduce
TaskTracker Node
Custom hadoop type which implements hadoop's serialization framework - Writable interface.
Data to serialize/de-serialize
Size of the image (in bytes)
Start Y co-ordinate relative to input image
Height of the image (in pixels)
Image path
Image data (byte array)
How to serialize a Buffered I mage using Hadoop's serialization framework?
1. Serialize RGB colour data
Drawback: Color conversion takes place if the
default model does not match the image
2. Serialize raw pixel data as a byte array
Drawback: Requires conditional code based on
image type.
3. Serialize in a preferred image format using ImageIO
1) do not have to deal directly with underlying pixel data representation.
2) intermediate image formats can be specified using custom properties.
3) default PNG format used provides lossless zip compression.
Responsible for,
1. writing output for the job to the file system
2. checking for validity of output specification of job
A RecordWriter is used to write the output key-value pairs to an output file.
Base class for all implementations of OutputFormat that use files as their data sink. Provides,
methods to define the output specifications for the job.
an implementation of checkOutputSpecs.
OutputFormat<K,V> FileQutputFormat<K,V>
/ I \
TextOutputFormat MultipleQutputFormat ImageOutputFormat
ImageOutputFormat is designed similar to the MultipleOutputFormat abstract class, which allows you to
write data to multiple files whose names are derived from
the output keys and values.
ImageOutputFormat provides a RecordWriter implementation for writing the image to the file system.
Key: image path, based on which the output file name is
Value: image and its metadata, written to file system using ImageIO in the preferred output format.
Map and Reduce function
The map function is an implementation of the Sobel edge detection algorithm applied to the input sub image.
The reduce function uses the metadata associated with each image split to combine them into the final processed output
Key: Image path Value: A sub image
Key: Image path
Value: Processed sub image
Map function
Reduce function!
Key: Image path Value: List of processed &ub images
Key: Image path Value: Final image
Distributed Image Processing
Cluster Details
Master node
Intel Pentium 4 CPU @ 2.60 GHz, 1.00 GB RAM
Slave node
Intel Core 2 Duo T9550 @ 2.66 GHz. 1.95 GB RAM
Input Data Set
6 hi-resolution PNG images
Largest: 3000 pixels, Smallest: 1080 pixels (height)
Cluster Settings
Maximum concurrent map tasks per tasktracker = 2
Number of reduce tasks for job = 2
Number of splits per image = 2
DFS replication = 2
Distributed Image Processing Performance
Wall Clock Time
Serial Execution (Master) Begin: 8:38:20, End: 8:43:05 Time: 4 minutes, 45 seconds
Parallel Execution (Cluster) Begin: 7:38:05, End: 7:40:55 Time: 2 minutes, 50 seconds
Distributed Image Processing
Handle global and local boundary conditions
Handle images larger than HDFS block size
Evaluate performance of multi-threaded map runner

Post: #2
Thanks for uploading. I will learn about it .
Post: #3
Interesting topic. I would be nice to discuss more deeply how the image is split. Because as we all know during convolution edge pixels are not computed, so the split images have to overlap with each other.

P.S. are the source code available online?
Post: #4
Can tell me wat r the hardware and software requirements for the project..?
I have no idea wat a hadoop is?I want to implement this project.Can u Pls guide me.
Post: #5
Do you provide the source code for this process?
I also want to do some application of image processing based on mapreduce, but I'm a beginner of this framework.So it is difficult for me to write the code for input and output of the image based on the mapreduce framework.Could somebody help me? Thank you in advance.
Post: #6
To get more information about the topic " Distributed Image Processing using hadoop MapReduce framework" please refer the link below
Post: #7
Is the source code available?Thank you.
Post: #8
To get more information about the topic " Distributed Image Processing using hadoop MapReduce framework" please refer the link below

Marked Categories : image map reduce, ppt on distributed image processing on hadoop, ppt on distributed image processing in hadoop, code for image processing on hadoop, distributed image processing using hadoop mapreduce framework, mapreduce processing images, image processing mapreduce, mapreduce criticism, image processing using mapreduce, hadoop parallel image processing interface ppt, distributed image proceesing in hadoop, meaning of distributive image processing, mapreduce images, hadoop, distributed image processing, mapreduce image processing, hadoop image processing, image processing using hadoop, map reduce code for image processing, hadoop mapreduce splitting the input,

Quick Reply
Type your reply to this message here.

Image Verification
Image Verification
(case insensitive)
Please enter the text within the image on the left in to the text box below. This process is used to prevent automated posts.

Possibly Related Threads...
Thread: Author Replies: Views: Last Post
  simulation of venturini modulation technique for matrix converter using simulink Guest 0 0 Yesterday 07:31 PM
Last Post: Guest
  matlab source code for cartoon texture image decomposition Guest 0 0 Yesterday 04:54 PM
Last Post: Guest
  object tracking using wireless sensor network in matlab Guest 0 0 Yesterday 01:17 PM
Last Post: Guest
  windmill mini project using opengl with source code Guest 0 0 21-04-2017 08:20 PM
Last Post: Guest
  transportation problem using fuzzy logic pdf Guest 1 197 21-04-2017 04:25 PM
Last Post: jaseela123
  kalman noise image java code Guest 1 0 21-04-2017 02:47 PM
Last Post: jaseela123
  image compression using discrete cosine transform Guest 1 0 21-04-2017 02:07 PM
Last Post: jaseela123
  source code for aquarium in computer graphics using opengl Guest 2 0 21-04-2017 01:57 PM
Last Post: jaseela123
Lightbulb IRS 1C/1D Image Visualization Tool with Filtering Functionality Guest 1 124 21-04-2017 11:49 AM
Last Post: jaseela123
  automatic vehicle security system using alcohol sensor Guest 1 172 21-04-2017 11:35 AM
Last Post: jaseela123
This Page May Contain What is Distributed Image Processing using hadoop MapReduce framework And Latest Information/News About Distributed Image Processing using hadoop MapReduce framework,If Not ...Use Search to get more info about Distributed Image Processing using hadoop MapReduce framework Or Ask Here