An Evaluation of Dominant Color descriptor and Wavelet Transform on YCbCr Color Space for CBIR

Abstract

Content based image retrieval (CBIR) presents an effectual strategy to probe the pictures from the databases. The function extraction and homogeneous attribute measures are the 2 key parameters for retrieval efficiency. A uniform attribute measure performs a paramount function in picture retrieval. “content-based” designates that the quest analyzes the contents of the image alternatively than the metadata equivalent to keyword phrases, tags, or descriptions associated with the picture. In this study, present an evaluation of dominant color descriptor (DCD) and wavelet transform in YCbCr color space for CBIR. In this work, color, feature is extracted using DCD for RGB and HSV color space and auto correlogram (AC). Texture feature is extracted using discrete wavelet transform (DWT) and Gabor Filter. Shape feature is extracted with the use of Kurtosis and Skewness. We take Corel-1000 database-African, Flowers, Food, Elephant and Horse images. Also calculate the Euclidean distance (ED), relative standard deviation (RSD), Correlation, Canberra distance (CD) and Jaccard distance (JD) between giving image and database images. For classification, used support vector machine (SVM) to organize the categories without time consuming. The performance analysis is evaluated on precision and accuracy.