Print ISSN: 1681-6900

Online ISSN: 2412-0758

Keywords : Kekre

Image Features Evaluation Using New Algorithm Proposed For Reducing Image Feature Number & Size Stored In Database

Shahlaa T. Abdulwahab

Engineering and Technology Journal, 2011, Volume 29, Issue 6, Pages 1176-1194

This study proposes technique that capable of reducing image features size and
number stored in the database. The proposed technique depends on the image content
of numerical values for the three basic colors (red, green and blue) and then stores it in
the database and to be used for image retrieval. This technique has been developed
based on recent image retrieval procedures that include Color Descriptor Matrix,
YCbCr Color Space and Discrete Cosine Transform. Those procedures have been
applied sequentially on the image and finally Kekre’s Transform has been applied in
the last stage of this technique to evaluate image features and reduce its stored size in
the database. The validity and accuracy of the proposed technique have been
evaluated through experiments by applying Kekre’s Transform on Color Descriptor
Matrix instead of using Kekre’s Transform directly on the image in order to reduce its
feature stored size. Another experiments have been tested and evaluated that include
the application of YCbCr Color Space on the Color Descriptor Matrix and
finally Kekre’s Transform to be executed and explore the image features size and
compare it with the previous stage. The effect of applying the Discrete Cosine
Transform on the YCbCr Color Space and finally the Kekre’s Transform on the image
features size has been studied and compared with the previous step. It is concluded that
the best reduction in image features size stored in the database can be obtained only
when Kekre’s Transform applied in the last step of the proposed technique with
unchanged threshold based image retrieval ratios. Parametric study has been conducted
to investigate the effect of applying the new algorithm on both isolated and mixed
image groups. Good precision ratios of 82% and 65% have been obtained for the
isolated and mixed image groups respectively.