I compared the various image enhancement routines
developed by me to the standard ImageMagick Library functions.
Both the resulting images and the time taken is shown
here.
The method used to find the time taken consisted of -
1. Setting up gettimeofday() system call at appropriate
placesto get the time of each function call. For this I coded each
function I wrote such that it returned
the time it took for processing the image in microseconds. And the times
for the
standard functions were found using
the gettimeofday() system call at the beginning and end of each function
call.
2. All this data was recorded into a data file.
Profiling could not be used for the purpose because
the standard library functions have not been compiled using the `-pg'
option and hence contain no profiling information.
The following pages conatain the results -
1. My functions for image
enhancement in the spatial domain
2. My functions for image
enhancement in the frequency domain
3. ImageMagick standard
functions
The following observations can be made from the results -
1. The standard routines are almost always faster
than mine, except in the median filtering case. This is due fact that all
image
operations can be further optimized
which I have not done due to time constraints.
2. The resulting image however is the same if not
better when obtained using my functions. Also I feel that my routines allow
a greater
flexibility in modifying the various
options/parameters which might effect the outcome.
3. The enhancement routines in the frequency domain
take much much longer than the ones in the spatial domain because of the
obvious
overhead caused due calculations involved
in the various transforms.
Note : Since these values are absolute time values
so they will vary when tested on a different machine (I tested them on
a PIII 600 Mhz)
But the pattern
of results should remain the same.
Improvements Suggested -
1. Apart from the technical optimizations used during
programming various theoretical optimizations can be implemented to speed
up
operations.
2. Reading the source code of the ImageMagick routines
may give better insight as to how these functions can be better implemented
Page last updated on 28 January, 2004. | AT cse.iitd.ac.in | © Parag Chaudhuri , 2009 |