H}istogram Partitioning for Feature Vector Dimension Reduction in Bins Approach for {CBIR

Dr. H. B. Kekre; Kavita Sonawane

Feature vector dimensionality is an important issue in any CBIR system. It has great impact on the execution time required by the system to process the given query and generate the retrieval results. We have introduced a novel idea in this paper to extract the feature vectors of the image along with dimensionality reduction. It gives the improved performance as compared to already existing methods. The approach used is called bins approach; designed and implemented using image histogram partitioning. Three image histograms are obtained for each image plane R, G and B separately. Each of them is partitioned using two different techniques namely LP-Linear partitioning and CG –Centre of Gravity partitioning. Performance of these two partitioning techniques is compared by taking 4 different cases into consideration. Four different cases implemented in this paper are based on the variations used in the techniques to extract the image features. Multiple feature vector databases are prepared as pre-processing part of this work. Feature extraction techniques used are based on the original-ORG as well as equalized histogram-EQH and their partitioning based on LP and CG. This partitioning generates 8 bins holding the count of the pixels based on the color contents of the image. Further these 8 bins data is processed by computing the first four statistical moments (Mean, Standard deviation-STD, Skewness-SKEW, and Kurtosis- KURTO) representing the feature vectors of dimension 8. Retrieval results obtained by comparing the query image with database feature vectors by means of three similarity measures Euclidean distance (ED), Absolute distance (AD) and Cosine correlation distance (CD).

Select Volume / Issues:

• Volume 3 Issue 6, Nov. 2012

Year:

2012

Type of Publication:

Article

Keywords:

CBIR; Bins; Linear Partitioning; CG Partitioning; Mean; Standard Deviation; Skewness; Kurtosis

Journal:

IJECCE

Volume:

3

Number:

6

Pages:

1630-1639

Month:

November