KDEH Based Image Retrieval Using Invariant Features

Image Retrieval scheme represents images on the basis of histograms derived from photometric color features. The main objective of this research is to apply Kernel Density Estimation function on Histogram (KDEH), to retrieve images from large collection of image database by characterizing the contents. The retrieval process is performed based on the content of images, and more precisely on a low level visual features based method such as color and shape. In this approach, the system extracts the color features in two color spaces namely RGB and HSV. The similarity measure between two color histograms are computed by Histogram Intersection Distance (HID) and their retrieval performances are compared based on precision and recall techniques. Various edge detection methods like gradient-based edge detection and signature-based edge detection for shape feature have been proposed to detect the arbitrary objects in an image. Finally, the KDEH method is compared with a Traditional Histogram (TH) approach by varying the histogram bins with for different testing and training. samples. It has been experimented that extracting the KDEH based color feature histogram provides improved results in terms of retrieval accuracy for the kind of images, which are enriched with color characteristics and