recognition on these cropped out images provides us with amuch greater accuracy than with the entire image.Fractional Coefficients are generally obtained as given inFigure 2.
Figure 2. The coloured regions correspond to the fractional coefficientscropped from the original image, seen in black.
There are a number of such transforms that have beenresearched that provide us with these results. Some of themcan be applied to Palm Print Recognition. In our paper, weapply a few of these transforms and check their accuracy forpalm print recognition. The transforms we are using includethe Discrete Cosine Transform, the P.C.A. Eigen VectorTransform, the Haar Transform, the Slant Transform, theHartley Transform, the Kekre Transform and the WalshTransform.III.
Before we get to the actual implementation of thealgorithm, let us see some pre-processing activities. Firstly,the database used consists of 8000 greyscale images of 128x128 resolution which contain the ROI of the palmprintsof the right hand of 400 people. It was obtained from theHong Kong Polytechnic University 2D_3D Database .Here, each subject had ten palm prints taken initially. Afteran average time of one month, the same subject had to comeand provide the palm prints again. Our testing set involvedthe first set of 4000 images from which query images wereextracted and the second involved the next 4000. All theseprocessing mechanisms were carried out in MATLABR2010a. The total size of data structures and variables usedtotalled more than 1.07 GB.One key technique that helped a great deal was theapplication of histogram equalization on the images in orderto make the ridges and lines seem more prominent as seenin Figure 3. These characteristics are highly important asthey form the backbone of most Palm Print Recognitiontechnique parameters. In our findings, we have implicitlyapplied histogram equalization on all images. Without it,accuracy was found to be as low as 74% at average withmost transforms. On the application of histogramequalization, it was found to increase to 94% in certaincases.
Figure 3. Histogram Equalized Image
For our analysis, we carried out a set of operations onthe databank mentioned above. The exact nature of theseoperations has been stated below in the form of analgorithm:
: Obtain the Query Image and perform HistogramEqualization on it.
: Apply the required Transformation on it.Now, this image is to be compared against a training setof 4000 images. These images constitute the images in thedatabase that were taken a month later.
: Obtain the Image Matrix for all images in thetraining set and perform Histogram Equalization on it.
: Apply the required Transform on each Image.
: Calculate the mean square error between eachImage in the Training set and the query image. If partialenergy coefficients are used, calculate the error betweenonly that part of the images which falls inside the fractionalcoefficient. The image with the minimum mean square erroris the closest match.V.
Before providing the results of our study, first let usobtain a brief understanding of the plethora of transformsthat are going to be applied in our study.
Discrete Cosine Transform
A discrete cosine Transform (DCT) is an extension of the fast Fourier Transform that works only in the realdomain. It represents a sequence of finitely arranged datapoints in terms of cosine functions oscillating at differentfrequencies. It is of great use in compression and is oftenused to provide boundary functions for differentialequations and are hence, used greatly in science andengineering. The DCT is found to be symmetric, orthogonaland separable .
The Haar transform is the oldest and possibly thesimplest wavelet basis.  . Like the Fourier Analysisbasis, it consists of square shaped functions whichrepresents functions in the orthonormal function basis. AHaar Wavelet used both high-pass filtering and low-passfiltering and works by incorporating image decompositionon first he image rows and then the image columns. Inessence, the Haar transform is one which when applied to
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 10, October 201185http://sites.google.com/site/ijcsis/ISSN 1947-5500