Image Representation Using EPANECHNIKOV Density Feature Points Estimator

In image retrieval most of the existing visual content based representation methods are usually application dependent or non robust, making them not suitable for generic applications. These representation methods use visual contents such as colour, texture, shape, size etc. Human image recognition is largely based on shape, thus making it very appealing for image representation algorithms in computer vision. In this paper we propose a generic image representation algorithm using Epanechnikov Density Feature Points Estimator (EDFPE). It is invariant to rotation, scale and translation. The image density feature points within defined rectangular rings around the gravitational centre of the image are obtained in the form of a vector. The EDFPE is applied to the vector representation of the image. The Cosine Angle Distance (CAD) algorithm is used to measure similarity of the images in the database. Quantitative evaluation of the performance of the system and comparison with other algorithms was done.